Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 11dc66f3

History | View | Annotate | Download (510.2 kB)

1
#
2
#
3

    
4
# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 Google Inc.
5
#
6
# This program is free software; you can redistribute it and/or modify
7
# it under the terms of the GNU General Public License as published by
8
# the Free Software Foundation; either version 2 of the License, or
9
# (at your option) any later version.
10
#
11
# This program is distributed in the hope that it will be useful, but
12
# WITHOUT ANY WARRANTY; without even the implied warranty of
13
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
# General Public License for more details.
15
#
16
# You should have received a copy of the GNU General Public License
17
# along with this program; if not, write to the Free Software
18
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19
# 02110-1301, USA.
20

    
21

    
22
"""Module implementing the master-side code."""
23

    
24
# pylint: disable=W0201,C0302
25

    
26
# W0201 since most LU attributes are defined in CheckPrereq or similar
27
# functions
28

    
29
# C0302: since we have waaaay too many lines in this module
30

    
31
import os
32
import os.path
33
import time
34
import re
35
import platform
36
import logging
37
import copy
38
import OpenSSL
39
import socket
40
import tempfile
41
import shutil
42
import itertools
43
import operator
44

    
45
from ganeti import ssh
46
from ganeti import utils
47
from ganeti import errors
48
from ganeti import hypervisor
49
from ganeti import locking
50
from ganeti import constants
51
from ganeti import objects
52
from ganeti import serializer
53
from ganeti import ssconf
54
from ganeti import uidpool
55
from ganeti import compat
56
from ganeti import masterd
57
from ganeti import netutils
58
from ganeti import query
59
from ganeti import qlang
60
from ganeti import opcodes
61
from ganeti import ht
62
from ganeti import rpc
63

    
64
import ganeti.masterd.instance # pylint: disable=W0611
65

    
66

    
67
#: Size of DRBD meta block device
68
DRBD_META_SIZE = 128
69

    
70
# States of instance
71
INSTANCE_UP = [constants.ADMINST_UP]
72
INSTANCE_DOWN = [constants.ADMINST_DOWN]
73
INSTANCE_OFFLINE = [constants.ADMINST_OFFLINE]
74
INSTANCE_ONLINE = [constants.ADMINST_DOWN, constants.ADMINST_UP]
75
INSTANCE_NOT_RUNNING = [constants.ADMINST_DOWN, constants.ADMINST_OFFLINE]
76

    
77

    
78
class ResultWithJobs:
79
  """Data container for LU results with jobs.
80

81
  Instances of this class returned from L{LogicalUnit.Exec} will be recognized
82
  by L{mcpu.Processor._ProcessResult}. The latter will then submit the jobs
83
  contained in the C{jobs} attribute and include the job IDs in the opcode
84
  result.
85

86
  """
87
  def __init__(self, jobs, **kwargs):
88
    """Initializes this class.
89

90
    Additional return values can be specified as keyword arguments.
91

92
    @type jobs: list of lists of L{opcode.OpCode}
93
    @param jobs: A list of lists of opcode objects
94

95
    """
96
    self.jobs = jobs
97
    self.other = kwargs
98

    
99

    
100
class LogicalUnit(object):
101
  """Logical Unit base class.
102

103
  Subclasses must follow these rules:
104
    - implement ExpandNames
105
    - implement CheckPrereq (except when tasklets are used)
106
    - implement Exec (except when tasklets are used)
107
    - implement BuildHooksEnv
108
    - implement BuildHooksNodes
109
    - redefine HPATH and HTYPE
110
    - optionally redefine their run requirements:
111
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
112

113
  Note that all commands require root permissions.
114

115
  @ivar dry_run_result: the value (if any) that will be returned to the caller
116
      in dry-run mode (signalled by opcode dry_run parameter)
117

118
  """
119
  HPATH = None
120
  HTYPE = None
121
  REQ_BGL = True
122

    
123
  def __init__(self, processor, op, context, rpc_runner):
124
    """Constructor for LogicalUnit.
125

126
    This needs to be overridden in derived classes in order to check op
127
    validity.
128

129
    """
130
    self.proc = processor
131
    self.op = op
132
    self.cfg = context.cfg
133
    self.glm = context.glm
134
    # readability alias
135
    self.owned_locks = context.glm.list_owned
136
    self.context = context
137
    self.rpc = rpc_runner
138
    # Dicts used to declare locking needs to mcpu
139
    self.needed_locks = None
140
    self.share_locks = dict.fromkeys(locking.LEVELS, 0)
141
    self.add_locks = {}
142
    self.remove_locks = {}
143
    # Used to force good behavior when calling helper functions
144
    self.recalculate_locks = {}
145
    # logging
146
    self.Log = processor.Log # pylint: disable=C0103
147
    self.LogWarning = processor.LogWarning # pylint: disable=C0103
148
    self.LogInfo = processor.LogInfo # pylint: disable=C0103
149
    self.LogStep = processor.LogStep # pylint: disable=C0103
150
    # support for dry-run
151
    self.dry_run_result = None
152
    # support for generic debug attribute
153
    if (not hasattr(self.op, "debug_level") or
154
        not isinstance(self.op.debug_level, int)):
155
      self.op.debug_level = 0
156

    
157
    # Tasklets
158
    self.tasklets = None
159

    
160
    # Validate opcode parameters and set defaults
161
    self.op.Validate(True)
162

    
163
    self.CheckArguments()
164

    
165
  def CheckArguments(self):
166
    """Check syntactic validity for the opcode arguments.
167

168
    This method is for doing a simple syntactic check and ensure
169
    validity of opcode parameters, without any cluster-related
170
    checks. While the same can be accomplished in ExpandNames and/or
171
    CheckPrereq, doing these separate is better because:
172

173
      - ExpandNames is left as as purely a lock-related function
174
      - CheckPrereq is run after we have acquired locks (and possible
175
        waited for them)
176

177
    The function is allowed to change the self.op attribute so that
178
    later methods can no longer worry about missing parameters.
179

180
    """
181
    pass
182

    
183
  def ExpandNames(self):
184
    """Expand names for this LU.
185

186
    This method is called before starting to execute the opcode, and it should
187
    update all the parameters of the opcode to their canonical form (e.g. a
188
    short node name must be fully expanded after this method has successfully
189
    completed). This way locking, hooks, logging, etc. can work correctly.
190

191
    LUs which implement this method must also populate the self.needed_locks
192
    member, as a dict with lock levels as keys, and a list of needed lock names
193
    as values. Rules:
194

195
      - use an empty dict if you don't need any lock
196
      - if you don't need any lock at a particular level omit that level
197
      - don't put anything for the BGL level
198
      - if you want all locks at a level use locking.ALL_SET as a value
199

200
    If you need to share locks (rather than acquire them exclusively) at one
201
    level you can modify self.share_locks, setting a true value (usually 1) for
202
    that level. By default locks are not shared.
203

204
    This function can also define a list of tasklets, which then will be
205
    executed in order instead of the usual LU-level CheckPrereq and Exec
206
    functions, if those are not defined by the LU.
207

208
    Examples::
209

210
      # Acquire all nodes and one instance
211
      self.needed_locks = {
212
        locking.LEVEL_NODE: locking.ALL_SET,
213
        locking.LEVEL_INSTANCE: ['instance1.example.com'],
214
      }
215
      # Acquire just two nodes
216
      self.needed_locks = {
217
        locking.LEVEL_NODE: ['node1.example.com', 'node2.example.com'],
218
      }
219
      # Acquire no locks
220
      self.needed_locks = {} # No, you can't leave it to the default value None
221

222
    """
223
    # The implementation of this method is mandatory only if the new LU is
224
    # concurrent, so that old LUs don't need to be changed all at the same
225
    # time.
226
    if self.REQ_BGL:
227
      self.needed_locks = {} # Exclusive LUs don't need locks.
228
    else:
229
      raise NotImplementedError
230

    
231
  def DeclareLocks(self, level):
232
    """Declare LU locking needs for a level
233

234
    While most LUs can just declare their locking needs at ExpandNames time,
235
    sometimes there's the need to calculate some locks after having acquired
236
    the ones before. This function is called just before acquiring locks at a
237
    particular level, but after acquiring the ones at lower levels, and permits
238
    such calculations. It can be used to modify self.needed_locks, and by
239
    default it does nothing.
240

241
    This function is only called if you have something already set in
242
    self.needed_locks for the level.
243

244
    @param level: Locking level which is going to be locked
245
    @type level: member of ganeti.locking.LEVELS
246

247
    """
248

    
249
  def CheckPrereq(self):
250
    """Check prerequisites for this LU.
251

252
    This method should check that the prerequisites for the execution
253
    of this LU are fulfilled. It can do internode communication, but
254
    it should be idempotent - no cluster or system changes are
255
    allowed.
256

257
    The method should raise errors.OpPrereqError in case something is
258
    not fulfilled. Its return value is ignored.
259

260
    This method should also update all the parameters of the opcode to
261
    their canonical form if it hasn't been done by ExpandNames before.
262

263
    """
264
    if self.tasklets is not None:
265
      for (idx, tl) in enumerate(self.tasklets):
266
        logging.debug("Checking prerequisites for tasklet %s/%s",
267
                      idx + 1, len(self.tasklets))
268
        tl.CheckPrereq()
269
    else:
270
      pass
271

    
272
  def Exec(self, feedback_fn):
273
    """Execute the LU.
274

275
    This method should implement the actual work. It should raise
276
    errors.OpExecError for failures that are somewhat dealt with in
277
    code, or expected.
278

279
    """
280
    if self.tasklets is not None:
281
      for (idx, tl) in enumerate(self.tasklets):
282
        logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
283
        tl.Exec(feedback_fn)
284
    else:
285
      raise NotImplementedError
286

    
287
  def BuildHooksEnv(self):
288
    """Build hooks environment for this LU.
289

290
    @rtype: dict
291
    @return: Dictionary containing the environment that will be used for
292
      running the hooks for this LU. The keys of the dict must not be prefixed
293
      with "GANETI_"--that'll be added by the hooks runner. The hooks runner
294
      will extend the environment with additional variables. If no environment
295
      should be defined, an empty dictionary should be returned (not C{None}).
296
    @note: If the C{HPATH} attribute of the LU class is C{None}, this function
297
      will not be called.
298

299
    """
300
    raise NotImplementedError
301

    
302
  def BuildHooksNodes(self):
303
    """Build list of nodes to run LU's hooks.
304

305
    @rtype: tuple; (list, list)
306
    @return: Tuple containing a list of node names on which the hook
307
      should run before the execution and a list of node names on which the
308
      hook should run after the execution. No nodes should be returned as an
309
      empty list (and not None).
310
    @note: If the C{HPATH} attribute of the LU class is C{None}, this function
311
      will not be called.
312

313
    """
314
    raise NotImplementedError
315

    
316
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
317
    """Notify the LU about the results of its hooks.
318

319
    This method is called every time a hooks phase is executed, and notifies
320
    the Logical Unit about the hooks' result. The LU can then use it to alter
321
    its result based on the hooks.  By default the method does nothing and the
322
    previous result is passed back unchanged but any LU can define it if it
323
    wants to use the local cluster hook-scripts somehow.
324

325
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
326
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
327
    @param hook_results: the results of the multi-node hooks rpc call
328
    @param feedback_fn: function used send feedback back to the caller
329
    @param lu_result: the previous Exec result this LU had, or None
330
        in the PRE phase
331
    @return: the new Exec result, based on the previous result
332
        and hook results
333

334
    """
335
    # API must be kept, thus we ignore the unused argument and could
336
    # be a function warnings
337
    # pylint: disable=W0613,R0201
338
    return lu_result
339

    
340
  def _ExpandAndLockInstance(self):
341
    """Helper function to expand and lock an instance.
342

343
    Many LUs that work on an instance take its name in self.op.instance_name
344
    and need to expand it and then declare the expanded name for locking. This
345
    function does it, and then updates self.op.instance_name to the expanded
346
    name. It also initializes needed_locks as a dict, if this hasn't been done
347
    before.
348

349
    """
350
    if self.needed_locks is None:
351
      self.needed_locks = {}
352
    else:
353
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
354
        "_ExpandAndLockInstance called with instance-level locks set"
355
    self.op.instance_name = _ExpandInstanceName(self.cfg,
356
                                                self.op.instance_name)
357
    self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
358

    
359
  def _LockInstancesNodes(self, primary_only=False,
360
                          level=locking.LEVEL_NODE):
361
    """Helper function to declare instances' nodes for locking.
362

363
    This function should be called after locking one or more instances to lock
364
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
365
    with all primary or secondary nodes for instances already locked and
366
    present in self.needed_locks[locking.LEVEL_INSTANCE].
367

368
    It should be called from DeclareLocks, and for safety only works if
369
    self.recalculate_locks[locking.LEVEL_NODE] is set.
370

371
    In the future it may grow parameters to just lock some instance's nodes, or
372
    to just lock primaries or secondary nodes, if needed.
373

374
    If should be called in DeclareLocks in a way similar to::
375

376
      if level == locking.LEVEL_NODE:
377
        self._LockInstancesNodes()
378

379
    @type primary_only: boolean
380
    @param primary_only: only lock primary nodes of locked instances
381
    @param level: Which lock level to use for locking nodes
382

383
    """
384
    assert level in self.recalculate_locks, \
385
      "_LockInstancesNodes helper function called with no nodes to recalculate"
386

    
387
    # TODO: check if we're really been called with the instance locks held
388

    
389
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
390
    # future we might want to have different behaviors depending on the value
391
    # of self.recalculate_locks[locking.LEVEL_NODE]
392
    wanted_nodes = []
393
    locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
394
    for _, instance in self.cfg.GetMultiInstanceInfo(locked_i):
395
      wanted_nodes.append(instance.primary_node)
396
      if not primary_only:
397
        wanted_nodes.extend(instance.secondary_nodes)
398

    
399
    if self.recalculate_locks[level] == constants.LOCKS_REPLACE:
400
      self.needed_locks[level] = wanted_nodes
401
    elif self.recalculate_locks[level] == constants.LOCKS_APPEND:
402
      self.needed_locks[level].extend(wanted_nodes)
403
    else:
404
      raise errors.ProgrammerError("Unknown recalculation mode")
405

    
406
    del self.recalculate_locks[level]
407

    
408

    
409
class NoHooksLU(LogicalUnit): # pylint: disable=W0223
410
  """Simple LU which runs no hooks.
411

412
  This LU is intended as a parent for other LogicalUnits which will
413
  run no hooks, in order to reduce duplicate code.
414

415
  """
416
  HPATH = None
417
  HTYPE = None
418

    
419
  def BuildHooksEnv(self):
420
    """Empty BuildHooksEnv for NoHooksLu.
421

422
    This just raises an error.
423

424
    """
425
    raise AssertionError("BuildHooksEnv called for NoHooksLUs")
426

    
427
  def BuildHooksNodes(self):
428
    """Empty BuildHooksNodes for NoHooksLU.
429

430
    """
431
    raise AssertionError("BuildHooksNodes called for NoHooksLU")
432

    
433

    
434
class Tasklet:
435
  """Tasklet base class.
436

437
  Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
438
  they can mix legacy code with tasklets. Locking needs to be done in the LU,
439
  tasklets know nothing about locks.
440

441
  Subclasses must follow these rules:
442
    - Implement CheckPrereq
443
    - Implement Exec
444

445
  """
446
  def __init__(self, lu):
447
    self.lu = lu
448

    
449
    # Shortcuts
450
    self.cfg = lu.cfg
451
    self.rpc = lu.rpc
452

    
453
  def CheckPrereq(self):
454
    """Check prerequisites for this tasklets.
455

456
    This method should check whether the prerequisites for the execution of
457
    this tasklet are fulfilled. It can do internode communication, but it
458
    should be idempotent - no cluster or system changes are allowed.
459

460
    The method should raise errors.OpPrereqError in case something is not
461
    fulfilled. Its return value is ignored.
462

463
    This method should also update all parameters to their canonical form if it
464
    hasn't been done before.
465

466
    """
467
    pass
468

    
469
  def Exec(self, feedback_fn):
470
    """Execute the tasklet.
471

472
    This method should implement the actual work. It should raise
473
    errors.OpExecError for failures that are somewhat dealt with in code, or
474
    expected.
475

476
    """
477
    raise NotImplementedError
478

    
479

    
480
class _QueryBase:
481
  """Base for query utility classes.
482

483
  """
484
  #: Attribute holding field definitions
485
  FIELDS = None
486

    
487
  def __init__(self, qfilter, fields, use_locking):
488
    """Initializes this class.
489

490
    """
491
    self.use_locking = use_locking
492

    
493
    self.query = query.Query(self.FIELDS, fields, qfilter=qfilter,
494
                             namefield="name")
495
    self.requested_data = self.query.RequestedData()
496
    self.names = self.query.RequestedNames()
497

    
498
    # Sort only if no names were requested
499
    self.sort_by_name = not self.names
500

    
501
    self.do_locking = None
502
    self.wanted = None
503

    
504
  def _GetNames(self, lu, all_names, lock_level):
505
    """Helper function to determine names asked for in the query.
506

507
    """
508
    if self.do_locking:
509
      names = lu.owned_locks(lock_level)
510
    else:
511
      names = all_names
512

    
513
    if self.wanted == locking.ALL_SET:
514
      assert not self.names
515
      # caller didn't specify names, so ordering is not important
516
      return utils.NiceSort(names)
517

    
518
    # caller specified names and we must keep the same order
519
    assert self.names
520
    assert not self.do_locking or lu.glm.is_owned(lock_level)
521

    
522
    missing = set(self.wanted).difference(names)
523
    if missing:
524
      raise errors.OpExecError("Some items were removed before retrieving"
525
                               " their data: %s" % missing)
526

    
527
    # Return expanded names
528
    return self.wanted
529

    
530
  def ExpandNames(self, lu):
531
    """Expand names for this query.
532

533
    See L{LogicalUnit.ExpandNames}.
534

535
    """
536
    raise NotImplementedError()
537

    
538
  def DeclareLocks(self, lu, level):
539
    """Declare locks for this query.
540

541
    See L{LogicalUnit.DeclareLocks}.
542

543
    """
544
    raise NotImplementedError()
545

    
546
  def _GetQueryData(self, lu):
547
    """Collects all data for this query.
548

549
    @return: Query data object
550

551
    """
552
    raise NotImplementedError()
553

    
554
  def NewStyleQuery(self, lu):
555
    """Collect data and execute query.
556

557
    """
558
    return query.GetQueryResponse(self.query, self._GetQueryData(lu),
559
                                  sort_by_name=self.sort_by_name)
560

    
561
  def OldStyleQuery(self, lu):
562
    """Collect data and execute query.
563

564
    """
565
    return self.query.OldStyleQuery(self._GetQueryData(lu),
566
                                    sort_by_name=self.sort_by_name)
567

    
568

    
569
def _ShareAll():
570
  """Returns a dict declaring all lock levels shared.
571

572
  """
573
  return dict.fromkeys(locking.LEVELS, 1)
574

    
575

    
576
def _MakeLegacyNodeInfo(data):
577
  """Formats the data returned by L{rpc.RpcRunner.call_node_info}.
578

579
  Converts the data into a single dictionary. This is fine for most use cases,
580
  but some require information from more than one volume group or hypervisor.
581

582
  """
583
  (bootid, (vg_info, ), (hv_info, )) = data
584

    
585
  return utils.JoinDisjointDicts(utils.JoinDisjointDicts(vg_info, hv_info), {
586
    "bootid": bootid,
587
    })
588

    
589

    
590
def _CheckInstanceNodeGroups(cfg, instance_name, owned_groups):
591
  """Checks if the owned node groups are still correct for an instance.
592

593
  @type cfg: L{config.ConfigWriter}
594
  @param cfg: The cluster configuration
595
  @type instance_name: string
596
  @param instance_name: Instance name
597
  @type owned_groups: set or frozenset
598
  @param owned_groups: List of currently owned node groups
599

600
  """
601
  inst_groups = cfg.GetInstanceNodeGroups(instance_name)
602

    
603
  if not owned_groups.issuperset(inst_groups):
604
    raise errors.OpPrereqError("Instance %s's node groups changed since"
605
                               " locks were acquired, current groups are"
606
                               " are '%s', owning groups '%s'; retry the"
607
                               " operation" %
608
                               (instance_name,
609
                                utils.CommaJoin(inst_groups),
610
                                utils.CommaJoin(owned_groups)),
611
                               errors.ECODE_STATE)
612

    
613
  return inst_groups
614

    
615

    
616
def _CheckNodeGroupInstances(cfg, group_uuid, owned_instances):
617
  """Checks if the instances in a node group are still correct.
618

619
  @type cfg: L{config.ConfigWriter}
620
  @param cfg: The cluster configuration
621
  @type group_uuid: string
622
  @param group_uuid: Node group UUID
623
  @type owned_instances: set or frozenset
624
  @param owned_instances: List of currently owned instances
625

626
  """
627
  wanted_instances = cfg.GetNodeGroupInstances(group_uuid)
628
  if owned_instances != wanted_instances:
629
    raise errors.OpPrereqError("Instances in node group '%s' changed since"
630
                               " locks were acquired, wanted '%s', have '%s';"
631
                               " retry the operation" %
632
                               (group_uuid,
633
                                utils.CommaJoin(wanted_instances),
634
                                utils.CommaJoin(owned_instances)),
635
                               errors.ECODE_STATE)
636

    
637
  return wanted_instances
638

    
639

    
640
def _SupportsOob(cfg, node):
641
  """Tells if node supports OOB.
642

643
  @type cfg: L{config.ConfigWriter}
644
  @param cfg: The cluster configuration
645
  @type node: L{objects.Node}
646
  @param node: The node
647
  @return: The OOB script if supported or an empty string otherwise
648

649
  """
650
  return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
651

    
652

    
653
def _GetWantedNodes(lu, nodes):
654
  """Returns list of checked and expanded node names.
655

656
  @type lu: L{LogicalUnit}
657
  @param lu: the logical unit on whose behalf we execute
658
  @type nodes: list
659
  @param nodes: list of node names or None for all nodes
660
  @rtype: list
661
  @return: the list of nodes, sorted
662
  @raise errors.ProgrammerError: if the nodes parameter is wrong type
663

664
  """
665
  if nodes:
666
    return [_ExpandNodeName(lu.cfg, name) for name in nodes]
667

    
668
  return utils.NiceSort(lu.cfg.GetNodeList())
669

    
670

    
671
def _GetWantedInstances(lu, instances):
672
  """Returns list of checked and expanded instance names.
673

674
  @type lu: L{LogicalUnit}
675
  @param lu: the logical unit on whose behalf we execute
676
  @type instances: list
677
  @param instances: list of instance names or None for all instances
678
  @rtype: list
679
  @return: the list of instances, sorted
680
  @raise errors.OpPrereqError: if the instances parameter is wrong type
681
  @raise errors.OpPrereqError: if any of the passed instances is not found
682

683
  """
684
  if instances:
685
    wanted = [_ExpandInstanceName(lu.cfg, name) for name in instances]
686
  else:
687
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
688
  return wanted
689

    
690

    
691
def _GetUpdatedParams(old_params, update_dict,
692
                      use_default=True, use_none=False):
693
  """Return the new version of a parameter dictionary.
694

695
  @type old_params: dict
696
  @param old_params: old parameters
697
  @type update_dict: dict
698
  @param update_dict: dict containing new parameter values, or
699
      constants.VALUE_DEFAULT to reset the parameter to its default
700
      value
701
  @param use_default: boolean
702
  @type use_default: whether to recognise L{constants.VALUE_DEFAULT}
703
      values as 'to be deleted' values
704
  @param use_none: boolean
705
  @type use_none: whether to recognise C{None} values as 'to be
706
      deleted' values
707
  @rtype: dict
708
  @return: the new parameter dictionary
709

710
  """
711
  params_copy = copy.deepcopy(old_params)
712
  for key, val in update_dict.iteritems():
713
    if ((use_default and val == constants.VALUE_DEFAULT) or
714
        (use_none and val is None)):
715
      try:
716
        del params_copy[key]
717
      except KeyError:
718
        pass
719
    else:
720
      params_copy[key] = val
721
  return params_copy
722

    
723

    
724
def _UpdateAndVerifySubDict(base, updates, type_check):
725
  """Updates and verifies a dict with sub dicts of the same type.
726

727
  @param base: The dict with the old data
728
  @param updates: The dict with the new data
729
  @param type_check: Dict suitable to ForceDictType to verify correct types
730
  @returns: A new dict with updated and verified values
731

732
  """
733
  def fn(old, value):
734
    new = _GetUpdatedParams(old, value)
735
    utils.ForceDictType(new, type_check)
736
    return new
737

    
738
  ret = copy.deepcopy(base)
739
  ret.update(dict((key, fn(base.get(key, {}), value))
740
                  for key, value in updates.items()))
741
  return ret
742

    
743

    
744
def _MergeAndVerifyHvState(op_input, obj_input):
745
  """Combines the hv state from an opcode with the one of the object
746

747
  @param op_input: The input dict from the opcode
748
  @param obj_input: The input dict from the objects
749
  @return: The verified and updated dict
750

751
  """
752
  if op_input:
753
    invalid_hvs = set(op_input) - constants.HYPER_TYPES
754
    if invalid_hvs:
755
      raise errors.OpPrereqError("Invalid hypervisor(s) in hypervisor state:"
756
                                 " %s" % utils.CommaJoin(invalid_hvs),
757
                                 errors.ECODE_INVAL)
758
    if obj_input is None:
759
      obj_input = {}
760
    type_check = constants.HVSTS_PARAMETER_TYPES
761
    return _UpdateAndVerifySubDict(obj_input, op_input, type_check)
762

    
763
  return None
764

    
765

    
766
def _MergeAndVerifyDiskState(op_input, obj_input):
767
  """Combines the disk state from an opcode with the one of the object
768

769
  @param op_input: The input dict from the opcode
770
  @param obj_input: The input dict from the objects
771
  @return: The verified and updated dict
772
  """
773
  if op_input:
774
    invalid_dst = set(op_input) - constants.DS_VALID_TYPES
775
    if invalid_dst:
776
      raise errors.OpPrereqError("Invalid storage type(s) in disk state: %s" %
777
                                 utils.CommaJoin(invalid_dst),
778
                                 errors.ECODE_INVAL)
779
    type_check = constants.DSS_PARAMETER_TYPES
780
    if obj_input is None:
781
      obj_input = {}
782
    return dict((key, _UpdateAndVerifySubDict(obj_input.get(key, {}), value,
783
                                              type_check))
784
                for key, value in op_input.items())
785

    
786
  return None
787

    
788

    
789
def _ReleaseLocks(lu, level, names=None, keep=None):
790
  """Releases locks owned by an LU.
791

792
  @type lu: L{LogicalUnit}
793
  @param level: Lock level
794
  @type names: list or None
795
  @param names: Names of locks to release
796
  @type keep: list or None
797
  @param keep: Names of locks to retain
798

799
  """
800
  assert not (keep is not None and names is not None), \
801
         "Only one of the 'names' and the 'keep' parameters can be given"
802

    
803
  if names is not None:
804
    should_release = names.__contains__
805
  elif keep:
806
    should_release = lambda name: name not in keep
807
  else:
808
    should_release = None
809

    
810
  owned = lu.owned_locks(level)
811
  if not owned:
812
    # Not owning any lock at this level, do nothing
813
    pass
814

    
815
  elif should_release:
816
    retain = []
817
    release = []
818

    
819
    # Determine which locks to release
820
    for name in owned:
821
      if should_release(name):
822
        release.append(name)
823
      else:
824
        retain.append(name)
825

    
826
    assert len(lu.owned_locks(level)) == (len(retain) + len(release))
827

    
828
    # Release just some locks
829
    lu.glm.release(level, names=release)
830

    
831
    assert frozenset(lu.owned_locks(level)) == frozenset(retain)
832
  else:
833
    # Release everything
834
    lu.glm.release(level)
835

    
836
    assert not lu.glm.is_owned(level), "No locks should be owned"
837

    
838

    
839
def _MapInstanceDisksToNodes(instances):
840
  """Creates a map from (node, volume) to instance name.
841

842
  @type instances: list of L{objects.Instance}
843
  @rtype: dict; tuple of (node name, volume name) as key, instance name as value
844

845
  """
846
  return dict(((node, vol), inst.name)
847
              for inst in instances
848
              for (node, vols) in inst.MapLVsByNode().items()
849
              for vol in vols)
850

    
851

    
852
def _RunPostHook(lu, node_name):
853
  """Runs the post-hook for an opcode on a single node.
854

855
  """
856
  hm = lu.proc.BuildHooksManager(lu)
857
  try:
858
    hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
859
  except:
860
    # pylint: disable=W0702
861
    lu.LogWarning("Errors occurred running hooks on %s" % node_name)
862

    
863

    
864
def _CheckOutputFields(static, dynamic, selected):
865
  """Checks whether all selected fields are valid.
866

867
  @type static: L{utils.FieldSet}
868
  @param static: static fields set
869
  @type dynamic: L{utils.FieldSet}
870
  @param dynamic: dynamic fields set
871

872
  """
873
  f = utils.FieldSet()
874
  f.Extend(static)
875
  f.Extend(dynamic)
876

    
877
  delta = f.NonMatching(selected)
878
  if delta:
879
    raise errors.OpPrereqError("Unknown output fields selected: %s"
880
                               % ",".join(delta), errors.ECODE_INVAL)
881

    
882

    
883
def _CheckGlobalHvParams(params):
884
  """Validates that given hypervisor params are not global ones.
885

886
  This will ensure that instances don't get customised versions of
887
  global params.
888

889
  """
890
  used_globals = constants.HVC_GLOBALS.intersection(params)
891
  if used_globals:
892
    msg = ("The following hypervisor parameters are global and cannot"
893
           " be customized at instance level, please modify them at"
894
           " cluster level: %s" % utils.CommaJoin(used_globals))
895
    raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
896

    
897

    
898
def _CheckNodeOnline(lu, node, msg=None):
899
  """Ensure that a given node is online.
900

901
  @param lu: the LU on behalf of which we make the check
902
  @param node: the node to check
903
  @param msg: if passed, should be a message to replace the default one
904
  @raise errors.OpPrereqError: if the node is offline
905

906
  """
907
  if msg is None:
908
    msg = "Can't use offline node"
909
  if lu.cfg.GetNodeInfo(node).offline:
910
    raise errors.OpPrereqError("%s: %s" % (msg, node), errors.ECODE_STATE)
911

    
912

    
913
def _CheckNodeNotDrained(lu, node):
914
  """Ensure that a given node is not drained.
915

916
  @param lu: the LU on behalf of which we make the check
917
  @param node: the node to check
918
  @raise errors.OpPrereqError: if the node is drained
919

920
  """
921
  if lu.cfg.GetNodeInfo(node).drained:
922
    raise errors.OpPrereqError("Can't use drained node %s" % node,
923
                               errors.ECODE_STATE)
924

    
925

    
926
def _CheckNodeVmCapable(lu, node):
927
  """Ensure that a given node is vm capable.
928

929
  @param lu: the LU on behalf of which we make the check
930
  @param node: the node to check
931
  @raise errors.OpPrereqError: if the node is not vm capable
932

933
  """
934
  if not lu.cfg.GetNodeInfo(node).vm_capable:
935
    raise errors.OpPrereqError("Can't use non-vm_capable node %s" % node,
936
                               errors.ECODE_STATE)
937

    
938

    
939
def _CheckNodeHasOS(lu, node, os_name, force_variant):
940
  """Ensure that a node supports a given OS.
941

942
  @param lu: the LU on behalf of which we make the check
943
  @param node: the node to check
944
  @param os_name: the OS to query about
945
  @param force_variant: whether to ignore variant errors
946
  @raise errors.OpPrereqError: if the node is not supporting the OS
947

948
  """
949
  result = lu.rpc.call_os_get(node, os_name)
950
  result.Raise("OS '%s' not in supported OS list for node %s" %
951
               (os_name, node),
952
               prereq=True, ecode=errors.ECODE_INVAL)
953
  if not force_variant:
954
    _CheckOSVariant(result.payload, os_name)
955

    
956

    
957
def _CheckNodeHasSecondaryIP(lu, node, secondary_ip, prereq):
958
  """Ensure that a node has the given secondary ip.
959

960
  @type lu: L{LogicalUnit}
961
  @param lu: the LU on behalf of which we make the check
962
  @type node: string
963
  @param node: the node to check
964
  @type secondary_ip: string
965
  @param secondary_ip: the ip to check
966
  @type prereq: boolean
967
  @param prereq: whether to throw a prerequisite or an execute error
968
  @raise errors.OpPrereqError: if the node doesn't have the ip, and prereq=True
969
  @raise errors.OpExecError: if the node doesn't have the ip, and prereq=False
970

971
  """
972
  result = lu.rpc.call_node_has_ip_address(node, secondary_ip)
973
  result.Raise("Failure checking secondary ip on node %s" % node,
974
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
975
  if not result.payload:
976
    msg = ("Node claims it doesn't have the secondary ip you gave (%s),"
977
           " please fix and re-run this command" % secondary_ip)
978
    if prereq:
979
      raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
980
    else:
981
      raise errors.OpExecError(msg)
982

    
983

    
984
def _GetClusterDomainSecret():
985
  """Reads the cluster domain secret.
986

987
  """
988
  return utils.ReadOneLineFile(constants.CLUSTER_DOMAIN_SECRET_FILE,
989
                               strict=True)
990

    
991

    
992
def _CheckInstanceState(lu, instance, req_states, msg=None):
993
  """Ensure that an instance is in one of the required states.
994

995
  @param lu: the LU on behalf of which we make the check
996
  @param instance: the instance to check
997
  @param msg: if passed, should be a message to replace the default one
998
  @raise errors.OpPrereqError: if the instance is not in the required state
999

1000
  """
1001
  if msg is None:
1002
    msg = "can't use instance from outside %s states" % ", ".join(req_states)
1003
  if instance.admin_state not in req_states:
1004
    raise errors.OpPrereqError("Instance %s is marked to be %s, %s" %
1005
                               (instance, instance.admin_state, msg),
1006
                               errors.ECODE_STATE)
1007

    
1008
  if constants.ADMINST_UP not in req_states:
1009
    pnode = instance.primary_node
1010
    ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
1011
    ins_l.Raise("Can't contact node %s for instance information" % pnode,
1012
                prereq=True, ecode=errors.ECODE_ENVIRON)
1013

    
1014
    if instance.name in ins_l.payload:
1015
      raise errors.OpPrereqError("Instance %s is running, %s" %
1016
                                 (instance.name, msg), errors.ECODE_STATE)
1017

    
1018

    
1019
def _ExpandItemName(fn, name, kind):
1020
  """Expand an item name.
1021

1022
  @param fn: the function to use for expansion
1023
  @param name: requested item name
1024
  @param kind: text description ('Node' or 'Instance')
1025
  @return: the resolved (full) name
1026
  @raise errors.OpPrereqError: if the item is not found
1027

1028
  """
1029
  full_name = fn(name)
1030
  if full_name is None:
1031
    raise errors.OpPrereqError("%s '%s' not known" % (kind, name),
1032
                               errors.ECODE_NOENT)
1033
  return full_name
1034

    
1035

    
1036
def _ExpandNodeName(cfg, name):
1037
  """Wrapper over L{_ExpandItemName} for nodes."""
1038
  return _ExpandItemName(cfg.ExpandNodeName, name, "Node")
1039

    
1040

    
1041
def _ExpandInstanceName(cfg, name):
1042
  """Wrapper over L{_ExpandItemName} for instance."""
1043
  return _ExpandItemName(cfg.ExpandInstanceName, name, "Instance")
1044

    
1045

    
1046
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
1047
                          minmem, maxmem, vcpus, nics, disk_template, disks,
1048
                          bep, hvp, hypervisor_name, tags):
1049
  """Builds instance related env variables for hooks
1050

1051
  This builds the hook environment from individual variables.
1052

1053
  @type name: string
1054
  @param name: the name of the instance
1055
  @type primary_node: string
1056
  @param primary_node: the name of the instance's primary node
1057
  @type secondary_nodes: list
1058
  @param secondary_nodes: list of secondary nodes as strings
1059
  @type os_type: string
1060
  @param os_type: the name of the instance's OS
1061
  @type status: string
1062
  @param status: the desired status of the instance
1063
  @type minmem: string
1064
  @param minmem: the minimum memory size of the instance
1065
  @type maxmem: string
1066
  @param maxmem: the maximum memory size of the instance
1067
  @type vcpus: string
1068
  @param vcpus: the count of VCPUs the instance has
1069
  @type nics: list
1070
  @param nics: list of tuples (ip, mac, mode, link) representing
1071
      the NICs the instance has
1072
  @type disk_template: string
1073
  @param disk_template: the disk template of the instance
1074
  @type disks: list
1075
  @param disks: the list of (size, mode) pairs
1076
  @type bep: dict
1077
  @param bep: the backend parameters for the instance
1078
  @type hvp: dict
1079
  @param hvp: the hypervisor parameters for the instance
1080
  @type hypervisor_name: string
1081
  @param hypervisor_name: the hypervisor for the instance
1082
  @type tags: list
1083
  @param tags: list of instance tags as strings
1084
  @rtype: dict
1085
  @return: the hook environment for this instance
1086

1087
  """
1088
  env = {
1089
    "OP_TARGET": name,
1090
    "INSTANCE_NAME": name,
1091
    "INSTANCE_PRIMARY": primary_node,
1092
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
1093
    "INSTANCE_OS_TYPE": os_type,
1094
    "INSTANCE_STATUS": status,
1095
    "INSTANCE_MINMEM": minmem,
1096
    "INSTANCE_MAXMEM": maxmem,
1097
    # TODO(2.7) remove deprecated "memory" value
1098
    "INSTANCE_MEMORY": maxmem,
1099
    "INSTANCE_VCPUS": vcpus,
1100
    "INSTANCE_DISK_TEMPLATE": disk_template,
1101
    "INSTANCE_HYPERVISOR": hypervisor_name,
1102
  }
1103
  if nics:
1104
    nic_count = len(nics)
1105
    for idx, (ip, mac, mode, link) in enumerate(nics):
1106
      if ip is None:
1107
        ip = ""
1108
      env["INSTANCE_NIC%d_IP" % idx] = ip
1109
      env["INSTANCE_NIC%d_MAC" % idx] = mac
1110
      env["INSTANCE_NIC%d_MODE" % idx] = mode
1111
      env["INSTANCE_NIC%d_LINK" % idx] = link
1112
      if mode == constants.NIC_MODE_BRIDGED:
1113
        env["INSTANCE_NIC%d_BRIDGE" % idx] = link
1114
  else:
1115
    nic_count = 0
1116

    
1117
  env["INSTANCE_NIC_COUNT"] = nic_count
1118

    
1119
  if disks:
1120
    disk_count = len(disks)
1121
    for idx, (size, mode) in enumerate(disks):
1122
      env["INSTANCE_DISK%d_SIZE" % idx] = size
1123
      env["INSTANCE_DISK%d_MODE" % idx] = mode
1124
  else:
1125
    disk_count = 0
1126

    
1127
  env["INSTANCE_DISK_COUNT"] = disk_count
1128

    
1129
  if not tags:
1130
    tags = []
1131

    
1132
  env["INSTANCE_TAGS"] = " ".join(tags)
1133

    
1134
  for source, kind in [(bep, "BE"), (hvp, "HV")]:
1135
    for key, value in source.items():
1136
      env["INSTANCE_%s_%s" % (kind, key)] = value
1137

    
1138
  return env
1139

    
1140

    
1141
def _NICListToTuple(lu, nics):
1142
  """Build a list of nic information tuples.
1143

1144
  This list is suitable to be passed to _BuildInstanceHookEnv or as a return
1145
  value in LUInstanceQueryData.
1146

1147
  @type lu:  L{LogicalUnit}
1148
  @param lu: the logical unit on whose behalf we execute
1149
  @type nics: list of L{objects.NIC}
1150
  @param nics: list of nics to convert to hooks tuples
1151

1152
  """
1153
  hooks_nics = []
1154
  cluster = lu.cfg.GetClusterInfo()
1155
  for nic in nics:
1156
    ip = nic.ip
1157
    mac = nic.mac
1158
    filled_params = cluster.SimpleFillNIC(nic.nicparams)
1159
    mode = filled_params[constants.NIC_MODE]
1160
    link = filled_params[constants.NIC_LINK]
1161
    hooks_nics.append((ip, mac, mode, link))
1162
  return hooks_nics
1163

    
1164

    
1165
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
1166
  """Builds instance related env variables for hooks from an object.
1167

1168
  @type lu: L{LogicalUnit}
1169
  @param lu: the logical unit on whose behalf we execute
1170
  @type instance: L{objects.Instance}
1171
  @param instance: the instance for which we should build the
1172
      environment
1173
  @type override: dict
1174
  @param override: dictionary with key/values that will override
1175
      our values
1176
  @rtype: dict
1177
  @return: the hook environment dictionary
1178

1179
  """
1180
  cluster = lu.cfg.GetClusterInfo()
1181
  bep = cluster.FillBE(instance)
1182
  hvp = cluster.FillHV(instance)
1183
  args = {
1184
    "name": instance.name,
1185
    "primary_node": instance.primary_node,
1186
    "secondary_nodes": instance.secondary_nodes,
1187
    "os_type": instance.os,
1188
    "status": instance.admin_state,
1189
    "maxmem": bep[constants.BE_MAXMEM],
1190
    "minmem": bep[constants.BE_MINMEM],
1191
    "vcpus": bep[constants.BE_VCPUS],
1192
    "nics": _NICListToTuple(lu, instance.nics),
1193
    "disk_template": instance.disk_template,
1194
    "disks": [(disk.size, disk.mode) for disk in instance.disks],
1195
    "bep": bep,
1196
    "hvp": hvp,
1197
    "hypervisor_name": instance.hypervisor,
1198
    "tags": instance.tags,
1199
  }
1200
  if override:
1201
    args.update(override)
1202
  return _BuildInstanceHookEnv(**args) # pylint: disable=W0142
1203

    
1204

    
1205
def _AdjustCandidatePool(lu, exceptions):
1206
  """Adjust the candidate pool after node operations.
1207

1208
  """
1209
  mod_list = lu.cfg.MaintainCandidatePool(exceptions)
1210
  if mod_list:
1211
    lu.LogInfo("Promoted nodes to master candidate role: %s",
1212
               utils.CommaJoin(node.name for node in mod_list))
1213
    for name in mod_list:
1214
      lu.context.ReaddNode(name)
1215
  mc_now, mc_max, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1216
  if mc_now > mc_max:
1217
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
1218
               (mc_now, mc_max))
1219

    
1220

    
1221
def _DecideSelfPromotion(lu, exceptions=None):
1222
  """Decide whether I should promote myself as a master candidate.
1223

1224
  """
1225
  cp_size = lu.cfg.GetClusterInfo().candidate_pool_size
1226
  mc_now, mc_should, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1227
  # the new node will increase mc_max with one, so:
1228
  mc_should = min(mc_should + 1, cp_size)
1229
  return mc_now < mc_should
1230

    
1231

    
1232
def _CheckNicsBridgesExist(lu, target_nics, target_node):
1233
  """Check that the brigdes needed by a list of nics exist.
1234

1235
  """
1236
  cluster = lu.cfg.GetClusterInfo()
1237
  paramslist = [cluster.SimpleFillNIC(nic.nicparams) for nic in target_nics]
1238
  brlist = [params[constants.NIC_LINK] for params in paramslist
1239
            if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
1240
  if brlist:
1241
    result = lu.rpc.call_bridges_exist(target_node, brlist)
1242
    result.Raise("Error checking bridges on destination node '%s'" %
1243
                 target_node, prereq=True, ecode=errors.ECODE_ENVIRON)
1244

    
1245

    
1246
def _CheckInstanceBridgesExist(lu, instance, node=None):
1247
  """Check that the brigdes needed by an instance exist.
1248

1249
  """
1250
  if node is None:
1251
    node = instance.primary_node
1252
  _CheckNicsBridgesExist(lu, instance.nics, node)
1253

    
1254

    
1255
def _CheckOSVariant(os_obj, name):
1256
  """Check whether an OS name conforms to the os variants specification.
1257

1258
  @type os_obj: L{objects.OS}
1259
  @param os_obj: OS object to check
1260
  @type name: string
1261
  @param name: OS name passed by the user, to check for validity
1262

1263
  """
1264
  variant = objects.OS.GetVariant(name)
1265
  if not os_obj.supported_variants:
1266
    if variant:
1267
      raise errors.OpPrereqError("OS '%s' doesn't support variants ('%s'"
1268
                                 " passed)" % (os_obj.name, variant),
1269
                                 errors.ECODE_INVAL)
1270
    return
1271
  if not variant:
1272
    raise errors.OpPrereqError("OS name must include a variant",
1273
                               errors.ECODE_INVAL)
1274

    
1275
  if variant not in os_obj.supported_variants:
1276
    raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)
1277

    
1278

    
1279
def _GetNodeInstancesInner(cfg, fn):
1280
  return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]
1281

    
1282

    
1283
def _GetNodeInstances(cfg, node_name):
1284
  """Returns a list of all primary and secondary instances on a node.
1285

1286
  """
1287

    
1288
  return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)
1289

    
1290

    
1291
def _GetNodePrimaryInstances(cfg, node_name):
1292
  """Returns primary instances on a node.
1293

1294
  """
1295
  return _GetNodeInstancesInner(cfg,
1296
                                lambda inst: node_name == inst.primary_node)
1297

    
1298

    
1299
def _GetNodeSecondaryInstances(cfg, node_name):
1300
  """Returns secondary instances on a node.
1301

1302
  """
1303
  return _GetNodeInstancesInner(cfg,
1304
                                lambda inst: node_name in inst.secondary_nodes)
1305

    
1306

    
1307
def _GetStorageTypeArgs(cfg, storage_type):
1308
  """Returns the arguments for a storage type.
1309

1310
  """
1311
  # Special case for file storage
1312
  if storage_type == constants.ST_FILE:
1313
    # storage.FileStorage wants a list of storage directories
1314
    return [[cfg.GetFileStorageDir(), cfg.GetSharedFileStorageDir()]]
1315

    
1316
  return []
1317

    
1318

    
1319
def _FindFaultyInstanceDisks(cfg, rpc_runner, instance, node_name, prereq):
1320
  faulty = []
1321

    
1322
  for dev in instance.disks:
1323
    cfg.SetDiskID(dev, node_name)
1324

    
1325
  result = rpc_runner.call_blockdev_getmirrorstatus(node_name, instance.disks)
1326
  result.Raise("Failed to get disk status from node %s" % node_name,
1327
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
1328

    
1329
  for idx, bdev_status in enumerate(result.payload):
1330
    if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
1331
      faulty.append(idx)
1332

    
1333
  return faulty
1334

    
1335

    
1336
def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
1337
  """Check the sanity of iallocator and node arguments and use the
1338
  cluster-wide iallocator if appropriate.
1339

1340
  Check that at most one of (iallocator, node) is specified. If none is
1341
  specified, then the LU's opcode's iallocator slot is filled with the
1342
  cluster-wide default iallocator.
1343

1344
  @type iallocator_slot: string
1345
  @param iallocator_slot: the name of the opcode iallocator slot
1346
  @type node_slot: string
1347
  @param node_slot: the name of the opcode target node slot
1348

1349
  """
1350
  node = getattr(lu.op, node_slot, None)
1351
  iallocator = getattr(lu.op, iallocator_slot, None)
1352

    
1353
  if node is not None and iallocator is not None:
1354
    raise errors.OpPrereqError("Do not specify both, iallocator and node",
1355
                               errors.ECODE_INVAL)
1356
  elif node is None and iallocator is None:
1357
    default_iallocator = lu.cfg.GetDefaultIAllocator()
1358
    if default_iallocator:
1359
      setattr(lu.op, iallocator_slot, default_iallocator)
1360
    else:
1361
      raise errors.OpPrereqError("No iallocator or node given and no"
1362
                                 " cluster-wide default iallocator found;"
1363
                                 " please specify either an iallocator or a"
1364
                                 " node, or set a cluster-wide default"
1365
                                 " iallocator")
1366

    
1367

    
1368
def _GetDefaultIAllocator(cfg, iallocator):
1369
  """Decides on which iallocator to use.
1370

1371
  @type cfg: L{config.ConfigWriter}
1372
  @param cfg: Cluster configuration object
1373
  @type iallocator: string or None
1374
  @param iallocator: Iallocator specified in opcode
1375
  @rtype: string
1376
  @return: Iallocator name
1377

1378
  """
1379
  if not iallocator:
1380
    # Use default iallocator
1381
    iallocator = cfg.GetDefaultIAllocator()
1382

    
1383
  if not iallocator:
1384
    raise errors.OpPrereqError("No iallocator was specified, neither in the"
1385
                               " opcode nor as a cluster-wide default",
1386
                               errors.ECODE_INVAL)
1387

    
1388
  return iallocator
1389

    
1390

    
1391
class LUClusterPostInit(LogicalUnit):
1392
  """Logical unit for running hooks after cluster initialization.
1393

1394
  """
1395
  HPATH = "cluster-init"
1396
  HTYPE = constants.HTYPE_CLUSTER
1397

    
1398
  def BuildHooksEnv(self):
1399
    """Build hooks env.
1400

1401
    """
1402
    return {
1403
      "OP_TARGET": self.cfg.GetClusterName(),
1404
      }
1405

    
1406
  def BuildHooksNodes(self):
1407
    """Build hooks nodes.
1408

1409
    """
1410
    return ([], [self.cfg.GetMasterNode()])
1411

    
1412
  def Exec(self, feedback_fn):
1413
    """Nothing to do.
1414

1415
    """
1416
    return True
1417

    
1418

    
1419
class LUClusterDestroy(LogicalUnit):
1420
  """Logical unit for destroying the cluster.
1421

1422
  """
1423
  HPATH = "cluster-destroy"
1424
  HTYPE = constants.HTYPE_CLUSTER
1425

    
1426
  def BuildHooksEnv(self):
1427
    """Build hooks env.
1428

1429
    """
1430
    return {
1431
      "OP_TARGET": self.cfg.GetClusterName(),
1432
      }
1433

    
1434
  def BuildHooksNodes(self):
1435
    """Build hooks nodes.
1436

1437
    """
1438
    return ([], [])
1439

    
1440
  def CheckPrereq(self):
1441
    """Check prerequisites.
1442

1443
    This checks whether the cluster is empty.
1444

1445
    Any errors are signaled by raising errors.OpPrereqError.
1446

1447
    """
1448
    master = self.cfg.GetMasterNode()
1449

    
1450
    nodelist = self.cfg.GetNodeList()
1451
    if len(nodelist) != 1 or nodelist[0] != master:
1452
      raise errors.OpPrereqError("There are still %d node(s) in"
1453
                                 " this cluster." % (len(nodelist) - 1),
1454
                                 errors.ECODE_INVAL)
1455
    instancelist = self.cfg.GetInstanceList()
1456
    if instancelist:
1457
      raise errors.OpPrereqError("There are still %d instance(s) in"
1458
                                 " this cluster." % len(instancelist),
1459
                                 errors.ECODE_INVAL)
1460

    
1461
  def Exec(self, feedback_fn):
1462
    """Destroys the cluster.
1463

1464
    """
1465
    master_params = self.cfg.GetMasterNetworkParameters()
1466

    
1467
    # Run post hooks on master node before it's removed
1468
    _RunPostHook(self, master_params.name)
1469

    
1470
    ems = self.cfg.GetUseExternalMipScript()
1471
    result = self.rpc.call_node_deactivate_master_ip(master_params.name,
1472
                                                     master_params, ems)
1473
    result.Raise("Could not disable the master role")
1474

    
1475
    return master_params.name
1476

    
1477

    
1478
def _VerifyCertificate(filename):
1479
  """Verifies a certificate for L{LUClusterVerifyConfig}.
1480

1481
  @type filename: string
1482
  @param filename: Path to PEM file
1483

1484
  """
1485
  try:
1486
    cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1487
                                           utils.ReadFile(filename))
1488
  except Exception, err: # pylint: disable=W0703
1489
    return (LUClusterVerifyConfig.ETYPE_ERROR,
1490
            "Failed to load X509 certificate %s: %s" % (filename, err))
1491

    
1492
  (errcode, msg) = \
1493
    utils.VerifyX509Certificate(cert, constants.SSL_CERT_EXPIRATION_WARN,
1494
                                constants.SSL_CERT_EXPIRATION_ERROR)
1495

    
1496
  if msg:
1497
    fnamemsg = "While verifying %s: %s" % (filename, msg)
1498
  else:
1499
    fnamemsg = None
1500

    
1501
  if errcode is None:
1502
    return (None, fnamemsg)
1503
  elif errcode == utils.CERT_WARNING:
1504
    return (LUClusterVerifyConfig.ETYPE_WARNING, fnamemsg)
1505
  elif errcode == utils.CERT_ERROR:
1506
    return (LUClusterVerifyConfig.ETYPE_ERROR, fnamemsg)
1507

    
1508
  raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
1509

    
1510

    
1511
def _GetAllHypervisorParameters(cluster, instances):
1512
  """Compute the set of all hypervisor parameters.
1513

1514
  @type cluster: L{objects.Cluster}
1515
  @param cluster: the cluster object
1516
  @param instances: list of L{objects.Instance}
1517
  @param instances: additional instances from which to obtain parameters
1518
  @rtype: list of (origin, hypervisor, parameters)
1519
  @return: a list with all parameters found, indicating the hypervisor they
1520
       apply to, and the origin (can be "cluster", "os X", or "instance Y")
1521

1522
  """
1523
  hvp_data = []
1524

    
1525
  for hv_name in cluster.enabled_hypervisors:
1526
    hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
1527

    
1528
  for os_name, os_hvp in cluster.os_hvp.items():
1529
    for hv_name, hv_params in os_hvp.items():
1530
      if hv_params:
1531
        full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
1532
        hvp_data.append(("os %s" % os_name, hv_name, full_params))
1533

    
1534
  # TODO: collapse identical parameter values in a single one
1535
  for instance in instances:
1536
    if instance.hvparams:
1537
      hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
1538
                       cluster.FillHV(instance)))
1539

    
1540
  return hvp_data
1541

    
1542

    
1543
class _VerifyErrors(object):
1544
  """Mix-in for cluster/group verify LUs.
1545

1546
  It provides _Error and _ErrorIf, and updates the self.bad boolean. (Expects
1547
  self.op and self._feedback_fn to be available.)
1548

1549
  """
1550

    
1551
  ETYPE_FIELD = "code"
1552
  ETYPE_ERROR = "ERROR"
1553
  ETYPE_WARNING = "WARNING"
1554

    
1555
  def _Error(self, ecode, item, msg, *args, **kwargs):
1556
    """Format an error message.
1557

1558
    Based on the opcode's error_codes parameter, either format a
1559
    parseable error code, or a simpler error string.
1560

1561
    This must be called only from Exec and functions called from Exec.
1562

1563
    """
1564
    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
1565
    itype, etxt, _ = ecode
1566
    # first complete the msg
1567
    if args:
1568
      msg = msg % args
1569
    # then format the whole message
1570
    if self.op.error_codes: # This is a mix-in. pylint: disable=E1101
1571
      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
1572
    else:
1573
      if item:
1574
        item = " " + item
1575
      else:
1576
        item = ""
1577
      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
1578
    # and finally report it via the feedback_fn
1579
    self._feedback_fn("  - %s" % msg) # Mix-in. pylint: disable=E1101
1580

    
1581
  def _ErrorIf(self, cond, ecode, *args, **kwargs):
1582
    """Log an error message if the passed condition is True.
1583

1584
    """
1585
    cond = (bool(cond)
1586
            or self.op.debug_simulate_errors) # pylint: disable=E1101
1587

    
1588
    # If the error code is in the list of ignored errors, demote the error to a
1589
    # warning
1590
    (_, etxt, _) = ecode
1591
    if etxt in self.op.ignore_errors:     # pylint: disable=E1101
1592
      kwargs[self.ETYPE_FIELD] = self.ETYPE_WARNING
1593

    
1594
    if cond:
1595
      self._Error(ecode, *args, **kwargs)
1596

    
1597
    # do not mark the operation as failed for WARN cases only
1598
    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
1599
      self.bad = self.bad or cond
1600

    
1601

    
1602
class LUClusterVerify(NoHooksLU):
1603
  """Submits all jobs necessary to verify the cluster.
1604

1605
  """
1606
  REQ_BGL = False
1607

    
1608
  def ExpandNames(self):
1609
    self.needed_locks = {}
1610

    
1611
  def Exec(self, feedback_fn):
1612
    jobs = []
1613

    
1614
    if self.op.group_name:
1615
      groups = [self.op.group_name]
1616
      depends_fn = lambda: None
1617
    else:
1618
      groups = self.cfg.GetNodeGroupList()
1619

    
1620
      # Verify global configuration
1621
      jobs.append([
1622
        opcodes.OpClusterVerifyConfig(ignore_errors=self.op.ignore_errors)
1623
        ])
1624

    
1625
      # Always depend on global verification
1626
      depends_fn = lambda: [(-len(jobs), [])]
1627

    
1628
    jobs.extend([opcodes.OpClusterVerifyGroup(group_name=group,
1629
                                            ignore_errors=self.op.ignore_errors,
1630
                                            depends=depends_fn())]
1631
                for group in groups)
1632

    
1633
    # Fix up all parameters
1634
    for op in itertools.chain(*jobs): # pylint: disable=W0142
1635
      op.debug_simulate_errors = self.op.debug_simulate_errors
1636
      op.verbose = self.op.verbose
1637
      op.error_codes = self.op.error_codes
1638
      try:
1639
        op.skip_checks = self.op.skip_checks
1640
      except AttributeError:
1641
        assert not isinstance(op, opcodes.OpClusterVerifyGroup)
1642

    
1643
    return ResultWithJobs(jobs)
1644

    
1645

    
1646
class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
1647
  """Verifies the cluster config.
1648

1649
  """
1650
  REQ_BGL = True
1651

    
1652
  def _VerifyHVP(self, hvp_data):
1653
    """Verifies locally the syntax of the hypervisor parameters.
1654

1655
    """
1656
    for item, hv_name, hv_params in hvp_data:
1657
      msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
1658
             (item, hv_name))
1659
      try:
1660
        hv_class = hypervisor.GetHypervisor(hv_name)
1661
        utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1662
        hv_class.CheckParameterSyntax(hv_params)
1663
      except errors.GenericError, err:
1664
        self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg % str(err))
1665

    
1666
  def ExpandNames(self):
1667
    # Information can be safely retrieved as the BGL is acquired in exclusive
1668
    # mode
1669
    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER)
1670
    self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
1671
    self.all_node_info = self.cfg.GetAllNodesInfo()
1672
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1673
    self.needed_locks = {}
1674

    
1675
  def Exec(self, feedback_fn):
1676
    """Verify integrity of cluster, performing various test on nodes.
1677

1678
    """
1679
    self.bad = False
1680
    self._feedback_fn = feedback_fn
1681

    
1682
    feedback_fn("* Verifying cluster config")
1683

    
1684
    for msg in self.cfg.VerifyConfig():
1685
      self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg)
1686

    
1687
    feedback_fn("* Verifying cluster certificate files")
1688

    
1689
    for cert_filename in constants.ALL_CERT_FILES:
1690
      (errcode, msg) = _VerifyCertificate(cert_filename)
1691
      self._ErrorIf(errcode, constants.CV_ECLUSTERCERT, None, msg, code=errcode)
1692

    
1693
    feedback_fn("* Verifying hypervisor parameters")
1694

    
1695
    self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
1696
                                                self.all_inst_info.values()))
1697

    
1698
    feedback_fn("* Verifying all nodes belong to an existing group")
1699

    
1700
    # We do this verification here because, should this bogus circumstance
1701
    # occur, it would never be caught by VerifyGroup, which only acts on
1702
    # nodes/instances reachable from existing node groups.
1703

    
1704
    dangling_nodes = set(node.name for node in self.all_node_info.values()
1705
                         if node.group not in self.all_group_info)
1706

    
1707
    dangling_instances = {}
1708
    no_node_instances = []
1709

    
1710
    for inst in self.all_inst_info.values():
1711
      if inst.primary_node in dangling_nodes:
1712
        dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
1713
      elif inst.primary_node not in self.all_node_info:
1714
        no_node_instances.append(inst.name)
1715

    
1716
    pretty_dangling = [
1717
        "%s (%s)" %
1718
        (node.name,
1719
         utils.CommaJoin(dangling_instances.get(node.name,
1720
                                                ["no instances"])))
1721
        for node in dangling_nodes]
1722

    
1723
    self._ErrorIf(bool(dangling_nodes), constants.CV_ECLUSTERDANGLINGNODES,
1724
                  None,
1725
                  "the following nodes (and their instances) belong to a non"
1726
                  " existing group: %s", utils.CommaJoin(pretty_dangling))
1727

    
1728
    self._ErrorIf(bool(no_node_instances), constants.CV_ECLUSTERDANGLINGINST,
1729
                  None,
1730
                  "the following instances have a non-existing primary-node:"
1731
                  " %s", utils.CommaJoin(no_node_instances))
1732

    
1733
    return not self.bad
1734

    
1735

    
1736
class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
1737
  """Verifies the status of a node group.
1738

1739
  """
1740
  HPATH = "cluster-verify"
1741
  HTYPE = constants.HTYPE_CLUSTER
1742
  REQ_BGL = False
1743

    
1744
  _HOOKS_INDENT_RE = re.compile("^", re.M)
1745

    
1746
  class NodeImage(object):
1747
    """A class representing the logical and physical status of a node.
1748

1749
    @type name: string
1750
    @ivar name: the node name to which this object refers
1751
    @ivar volumes: a structure as returned from
1752
        L{ganeti.backend.GetVolumeList} (runtime)
1753
    @ivar instances: a list of running instances (runtime)
1754
    @ivar pinst: list of configured primary instances (config)
1755
    @ivar sinst: list of configured secondary instances (config)
1756
    @ivar sbp: dictionary of {primary-node: list of instances} for all
1757
        instances for which this node is secondary (config)
1758
    @ivar mfree: free memory, as reported by hypervisor (runtime)
1759
    @ivar dfree: free disk, as reported by the node (runtime)
1760
    @ivar offline: the offline status (config)
1761
    @type rpc_fail: boolean
1762
    @ivar rpc_fail: whether the RPC verify call was successfull (overall,
1763
        not whether the individual keys were correct) (runtime)
1764
    @type lvm_fail: boolean
1765
    @ivar lvm_fail: whether the RPC call didn't return valid LVM data
1766
    @type hyp_fail: boolean
1767
    @ivar hyp_fail: whether the RPC call didn't return the instance list
1768
    @type ghost: boolean
1769
    @ivar ghost: whether this is a known node or not (config)
1770
    @type os_fail: boolean
1771
    @ivar os_fail: whether the RPC call didn't return valid OS data
1772
    @type oslist: list
1773
    @ivar oslist: list of OSes as diagnosed by DiagnoseOS
1774
    @type vm_capable: boolean
1775
    @ivar vm_capable: whether the node can host instances
1776

1777
    """
1778
    def __init__(self, offline=False, name=None, vm_capable=True):
1779
      self.name = name
1780
      self.volumes = {}
1781
      self.instances = []
1782
      self.pinst = []
1783
      self.sinst = []
1784
      self.sbp = {}
1785
      self.mfree = 0
1786
      self.dfree = 0
1787
      self.offline = offline
1788
      self.vm_capable = vm_capable
1789
      self.rpc_fail = False
1790
      self.lvm_fail = False
1791
      self.hyp_fail = False
1792
      self.ghost = False
1793
      self.os_fail = False
1794
      self.oslist = {}
1795

    
1796
  def ExpandNames(self):
1797
    # This raises errors.OpPrereqError on its own:
1798
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
1799

    
1800
    # Get instances in node group; this is unsafe and needs verification later
1801
    inst_names = self.cfg.GetNodeGroupInstances(self.group_uuid)
1802

    
1803
    self.needed_locks = {
1804
      locking.LEVEL_INSTANCE: inst_names,
1805
      locking.LEVEL_NODEGROUP: [self.group_uuid],
1806
      locking.LEVEL_NODE: [],
1807
      }
1808

    
1809
    self.share_locks = _ShareAll()
1810

    
1811
  def DeclareLocks(self, level):
1812
    if level == locking.LEVEL_NODE:
1813
      # Get members of node group; this is unsafe and needs verification later
1814
      nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
1815

    
1816
      all_inst_info = self.cfg.GetAllInstancesInfo()
1817

    
1818
      # In Exec(), we warn about mirrored instances that have primary and
1819
      # secondary living in separate node groups. To fully verify that
1820
      # volumes for these instances are healthy, we will need to do an
1821
      # extra call to their secondaries. We ensure here those nodes will
1822
      # be locked.
1823
      for inst in self.owned_locks(locking.LEVEL_INSTANCE):
1824
        # Important: access only the instances whose lock is owned
1825
        if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
1826
          nodes.update(all_inst_info[inst].secondary_nodes)
1827

    
1828
      self.needed_locks[locking.LEVEL_NODE] = nodes
1829

    
1830
  def CheckPrereq(self):
1831
    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
1832
    self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
1833

    
1834
    group_nodes = set(self.group_info.members)
1835
    group_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
1836

    
1837
    unlocked_nodes = \
1838
        group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1839

    
1840
    unlocked_instances = \
1841
        group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
1842

    
1843
    if unlocked_nodes:
1844
      raise errors.OpPrereqError("Missing lock for nodes: %s" %
1845
                                 utils.CommaJoin(unlocked_nodes))
1846

    
1847
    if unlocked_instances:
1848
      raise errors.OpPrereqError("Missing lock for instances: %s" %
1849
                                 utils.CommaJoin(unlocked_instances))
1850

    
1851
    self.all_node_info = self.cfg.GetAllNodesInfo()
1852
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1853

    
1854
    self.my_node_names = utils.NiceSort(group_nodes)
1855
    self.my_inst_names = utils.NiceSort(group_instances)
1856

    
1857
    self.my_node_info = dict((name, self.all_node_info[name])
1858
                             for name in self.my_node_names)
1859

    
1860
    self.my_inst_info = dict((name, self.all_inst_info[name])
1861
                             for name in self.my_inst_names)
1862

    
1863
    # We detect here the nodes that will need the extra RPC calls for verifying
1864
    # split LV volumes; they should be locked.
1865
    extra_lv_nodes = set()
1866

    
1867
    for inst in self.my_inst_info.values():
1868
      if inst.disk_template in constants.DTS_INT_MIRROR:
1869
        group = self.my_node_info[inst.primary_node].group
1870
        for nname in inst.secondary_nodes:
1871
          if self.all_node_info[nname].group != group:
1872
            extra_lv_nodes.add(nname)
1873

    
1874
    unlocked_lv_nodes = \
1875
        extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1876

    
1877
    if unlocked_lv_nodes:
1878
      raise errors.OpPrereqError("these nodes could be locked: %s" %
1879
                                 utils.CommaJoin(unlocked_lv_nodes))
1880
    self.extra_lv_nodes = list(extra_lv_nodes)
1881

    
1882
  def _VerifyNode(self, ninfo, nresult):
1883
    """Perform some basic validation on data returned from a node.
1884

1885
      - check the result data structure is well formed and has all the
1886
        mandatory fields
1887
      - check ganeti version
1888

1889
    @type ninfo: L{objects.Node}
1890
    @param ninfo: the node to check
1891
    @param nresult: the results from the node
1892
    @rtype: boolean
1893
    @return: whether overall this call was successful (and we can expect
1894
         reasonable values in the respose)
1895

1896
    """
1897
    node = ninfo.name
1898
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1899

    
1900
    # main result, nresult should be a non-empty dict
1901
    test = not nresult or not isinstance(nresult, dict)
1902
    _ErrorIf(test, constants.CV_ENODERPC, node,
1903
                  "unable to verify node: no data returned")
1904
    if test:
1905
      return False
1906

    
1907
    # compares ganeti version
1908
    local_version = constants.PROTOCOL_VERSION
1909
    remote_version = nresult.get("version", None)
1910
    test = not (remote_version and
1911
                isinstance(remote_version, (list, tuple)) and
1912
                len(remote_version) == 2)
1913
    _ErrorIf(test, constants.CV_ENODERPC, node,
1914
             "connection to node returned invalid data")
1915
    if test:
1916
      return False
1917

    
1918
    test = local_version != remote_version[0]
1919
    _ErrorIf(test, constants.CV_ENODEVERSION, node,
1920
             "incompatible protocol versions: master %s,"
1921
             " node %s", local_version, remote_version[0])
1922
    if test:
1923
      return False
1924

    
1925
    # node seems compatible, we can actually try to look into its results
1926

    
1927
    # full package version
1928
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1929
                  constants.CV_ENODEVERSION, node,
1930
                  "software version mismatch: master %s, node %s",
1931
                  constants.RELEASE_VERSION, remote_version[1],
1932
                  code=self.ETYPE_WARNING)
1933

    
1934
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1935
    if ninfo.vm_capable and isinstance(hyp_result, dict):
1936
      for hv_name, hv_result in hyp_result.iteritems():
1937
        test = hv_result is not None
1938
        _ErrorIf(test, constants.CV_ENODEHV, node,
1939
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1940

    
1941
    hvp_result = nresult.get(constants.NV_HVPARAMS, None)
1942
    if ninfo.vm_capable and isinstance(hvp_result, list):
1943
      for item, hv_name, hv_result in hvp_result:
1944
        _ErrorIf(True, constants.CV_ENODEHV, node,
1945
                 "hypervisor %s parameter verify failure (source %s): %s",
1946
                 hv_name, item, hv_result)
1947

    
1948
    test = nresult.get(constants.NV_NODESETUP,
1949
                       ["Missing NODESETUP results"])
1950
    _ErrorIf(test, constants.CV_ENODESETUP, node, "node setup error: %s",
1951
             "; ".join(test))
1952

    
1953
    return True
1954

    
1955
  def _VerifyNodeTime(self, ninfo, nresult,
1956
                      nvinfo_starttime, nvinfo_endtime):
1957
    """Check the node time.
1958

1959
    @type ninfo: L{objects.Node}
1960
    @param ninfo: the node to check
1961
    @param nresult: the remote results for the node
1962
    @param nvinfo_starttime: the start time of the RPC call
1963
    @param nvinfo_endtime: the end time of the RPC call
1964

1965
    """
1966
    node = ninfo.name
1967
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1968

    
1969
    ntime = nresult.get(constants.NV_TIME, None)
1970
    try:
1971
      ntime_merged = utils.MergeTime(ntime)
1972
    except (ValueError, TypeError):
1973
      _ErrorIf(True, constants.CV_ENODETIME, node, "Node returned invalid time")
1974
      return
1975

    
1976
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1977
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1978
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1979
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1980
    else:
1981
      ntime_diff = None
1982

    
1983
    _ErrorIf(ntime_diff is not None, constants.CV_ENODETIME, node,
1984
             "Node time diverges by at least %s from master node time",
1985
             ntime_diff)
1986

    
1987
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1988
    """Check the node LVM results.
1989

1990
    @type ninfo: L{objects.Node}
1991
    @param ninfo: the node to check
1992
    @param nresult: the remote results for the node
1993
    @param vg_name: the configured VG name
1994

1995
    """
1996
    if vg_name is None:
1997
      return
1998

    
1999
    node = ninfo.name
2000
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2001

    
2002
    # checks vg existence and size > 20G
2003
    vglist = nresult.get(constants.NV_VGLIST, None)
2004
    test = not vglist
2005
    _ErrorIf(test, constants.CV_ENODELVM, node, "unable to check volume groups")
2006
    if not test:
2007
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
2008
                                            constants.MIN_VG_SIZE)
2009
      _ErrorIf(vgstatus, constants.CV_ENODELVM, node, vgstatus)
2010

    
2011
    # check pv names
2012
    pvlist = nresult.get(constants.NV_PVLIST, None)
2013
    test = pvlist is None
2014
    _ErrorIf(test, constants.CV_ENODELVM, node, "Can't get PV list from node")
2015
    if not test:
2016
      # check that ':' is not present in PV names, since it's a
2017
      # special character for lvcreate (denotes the range of PEs to
2018
      # use on the PV)
2019
      for _, pvname, owner_vg in pvlist:
2020
        test = ":" in pvname
2021
        _ErrorIf(test, constants.CV_ENODELVM, node,
2022
                 "Invalid character ':' in PV '%s' of VG '%s'",
2023
                 pvname, owner_vg)
2024

    
2025
  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
2026
    """Check the node bridges.
2027

2028
    @type ninfo: L{objects.Node}
2029
    @param ninfo: the node to check
2030
    @param nresult: the remote results for the node
2031
    @param bridges: the expected list of bridges
2032

2033
    """
2034
    if not bridges:
2035
      return
2036

    
2037
    node = ninfo.name
2038
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2039

    
2040
    missing = nresult.get(constants.NV_BRIDGES, None)
2041
    test = not isinstance(missing, list)
2042
    _ErrorIf(test, constants.CV_ENODENET, node,
2043
             "did not return valid bridge information")
2044
    if not test:
2045
      _ErrorIf(bool(missing), constants.CV_ENODENET, node,
2046
               "missing bridges: %s" % utils.CommaJoin(sorted(missing)))
2047

    
2048
  def _VerifyNodeUserScripts(self, ninfo, nresult):
2049
    """Check the results of user scripts presence and executability on the node
2050

2051
    @type ninfo: L{objects.Node}
2052
    @param ninfo: the node to check
2053
    @param nresult: the remote results for the node
2054

2055
    """
2056
    node = ninfo.name
2057

    
2058
    test = not constants.NV_USERSCRIPTS in nresult
2059
    self._ErrorIf(test, constants.CV_ENODEUSERSCRIPTS, node,
2060
                  "did not return user scripts information")
2061

    
2062
    broken_scripts = nresult.get(constants.NV_USERSCRIPTS, None)
2063
    if not test:
2064
      self._ErrorIf(broken_scripts, constants.CV_ENODEUSERSCRIPTS, node,
2065
                    "user scripts not present or not executable: %s" %
2066
                    utils.CommaJoin(sorted(broken_scripts)))
2067

    
2068
  def _VerifyNodeNetwork(self, ninfo, nresult):
2069
    """Check the node network connectivity results.
2070

2071
    @type ninfo: L{objects.Node}
2072
    @param ninfo: the node to check
2073
    @param nresult: the remote results for the node
2074

2075
    """
2076
    node = ninfo.name
2077
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2078

    
2079
    test = constants.NV_NODELIST not in nresult
2080
    _ErrorIf(test, constants.CV_ENODESSH, node,
2081
             "node hasn't returned node ssh connectivity data")
2082
    if not test:
2083
      if nresult[constants.NV_NODELIST]:
2084
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
2085
          _ErrorIf(True, constants.CV_ENODESSH, node,
2086
                   "ssh communication with node '%s': %s", a_node, a_msg)
2087

    
2088
    test = constants.NV_NODENETTEST not in nresult
2089
    _ErrorIf(test, constants.CV_ENODENET, node,
2090
             "node hasn't returned node tcp connectivity data")
2091
    if not test:
2092
      if nresult[constants.NV_NODENETTEST]:
2093
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
2094
        for anode in nlist:
2095
          _ErrorIf(True, constants.CV_ENODENET, node,
2096
                   "tcp communication with node '%s': %s",
2097
                   anode, nresult[constants.NV_NODENETTEST][anode])
2098

    
2099
    test = constants.NV_MASTERIP not in nresult
2100
    _ErrorIf(test, constants.CV_ENODENET, node,
2101
             "node hasn't returned node master IP reachability data")
2102
    if not test:
2103
      if not nresult[constants.NV_MASTERIP]:
2104
        if node == self.master_node:
2105
          msg = "the master node cannot reach the master IP (not configured?)"
2106
        else:
2107
          msg = "cannot reach the master IP"
2108
        _ErrorIf(True, constants.CV_ENODENET, node, msg)
2109

    
2110
  def _VerifyInstance(self, instance, instanceconfig, node_image,
2111
                      diskstatus):
2112
    """Verify an instance.
2113

2114
    This function checks to see if the required block devices are
2115
    available on the instance's node.
2116

2117
    """
2118
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2119
    node_current = instanceconfig.primary_node
2120

    
2121
    node_vol_should = {}
2122
    instanceconfig.MapLVsByNode(node_vol_should)
2123

    
2124
    for node in node_vol_should:
2125
      n_img = node_image[node]
2126
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2127
        # ignore missing volumes on offline or broken nodes
2128
        continue
2129
      for volume in node_vol_should[node]:
2130
        test = volume not in n_img.volumes
2131
        _ErrorIf(test, constants.CV_EINSTANCEMISSINGDISK, instance,
2132
                 "volume %s missing on node %s", volume, node)
2133

    
2134
    if instanceconfig.admin_state == constants.ADMINST_UP:
2135
      pri_img = node_image[node_current]
2136
      test = instance not in pri_img.instances and not pri_img.offline
2137
      _ErrorIf(test, constants.CV_EINSTANCEDOWN, instance,
2138
               "instance not running on its primary node %s",
2139
               node_current)
2140

    
2141
    diskdata = [(nname, success, status, idx)
2142
                for (nname, disks) in diskstatus.items()
2143
                for idx, (success, status) in enumerate(disks)]
2144

    
2145
    for nname, success, bdev_status, idx in diskdata:
2146
      # the 'ghost node' construction in Exec() ensures that we have a
2147
      # node here
2148
      snode = node_image[nname]
2149
      bad_snode = snode.ghost or snode.offline
2150
      _ErrorIf(instanceconfig.admin_state == constants.ADMINST_UP and
2151
               not success and not bad_snode,
2152
               constants.CV_EINSTANCEFAULTYDISK, instance,
2153
               "couldn't retrieve status for disk/%s on %s: %s",
2154
               idx, nname, bdev_status)
2155
      _ErrorIf((instanceconfig.admin_state == constants.ADMINST_UP and
2156
                success and bdev_status.ldisk_status == constants.LDS_FAULTY),
2157
               constants.CV_EINSTANCEFAULTYDISK, instance,
2158
               "disk/%s on %s is faulty", idx, nname)
2159

    
2160
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
2161
    """Verify if there are any unknown volumes in the cluster.
2162

2163
    The .os, .swap and backup volumes are ignored. All other volumes are
2164
    reported as unknown.
2165

2166
    @type reserved: L{ganeti.utils.FieldSet}
2167
    @param reserved: a FieldSet of reserved volume names
2168

2169
    """
2170
    for node, n_img in node_image.items():
2171
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2172
        # skip non-healthy nodes
2173
        continue
2174
      for volume in n_img.volumes:
2175
        test = ((node not in node_vol_should or
2176
                volume not in node_vol_should[node]) and
2177
                not reserved.Matches(volume))
2178
        self._ErrorIf(test, constants.CV_ENODEORPHANLV, node,
2179
                      "volume %s is unknown", volume)
2180

    
2181
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
2182
    """Verify N+1 Memory Resilience.
2183

2184
    Check that if one single node dies we can still start all the
2185
    instances it was primary for.
2186

2187
    """
2188
    cluster_info = self.cfg.GetClusterInfo()
2189
    for node, n_img in node_image.items():
2190
      # This code checks that every node which is now listed as
2191
      # secondary has enough memory to host all instances it is
2192
      # supposed to should a single other node in the cluster fail.
2193
      # FIXME: not ready for failover to an arbitrary node
2194
      # FIXME: does not support file-backed instances
2195
      # WARNING: we currently take into account down instances as well
2196
      # as up ones, considering that even if they're down someone
2197
      # might want to start them even in the event of a node failure.
2198
      if n_img.offline:
2199
        # we're skipping offline nodes from the N+1 warning, since
2200
        # most likely we don't have good memory infromation from them;
2201
        # we already list instances living on such nodes, and that's
2202
        # enough warning
2203
        continue
2204
      #TODO(dynmem): use MINMEM for checking
2205
      #TODO(dynmem): also consider ballooning out other instances
2206
      for prinode, instances in n_img.sbp.items():
2207
        needed_mem = 0
2208
        for instance in instances:
2209
          bep = cluster_info.FillBE(instance_cfg[instance])
2210
          if bep[constants.BE_AUTO_BALANCE]:
2211
            needed_mem += bep[constants.BE_MAXMEM]
2212
        test = n_img.mfree < needed_mem
2213
        self._ErrorIf(test, constants.CV_ENODEN1, node,
2214
                      "not enough memory to accomodate instance failovers"
2215
                      " should node %s fail (%dMiB needed, %dMiB available)",
2216
                      prinode, needed_mem, n_img.mfree)
2217

    
2218
  @classmethod
2219
  def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
2220
                   (files_all, files_opt, files_mc, files_vm)):
2221
    """Verifies file checksums collected from all nodes.
2222

2223
    @param errorif: Callback for reporting errors
2224
    @param nodeinfo: List of L{objects.Node} objects
2225
    @param master_node: Name of master node
2226
    @param all_nvinfo: RPC results
2227

2228
    """
2229
    # Define functions determining which nodes to consider for a file
2230
    files2nodefn = [
2231
      (files_all, None),
2232
      (files_mc, lambda node: (node.master_candidate or
2233
                               node.name == master_node)),
2234
      (files_vm, lambda node: node.vm_capable),
2235
      ]
2236

    
2237
    # Build mapping from filename to list of nodes which should have the file
2238
    nodefiles = {}
2239
    for (files, fn) in files2nodefn:
2240
      if fn is None:
2241
        filenodes = nodeinfo
2242
      else:
2243
        filenodes = filter(fn, nodeinfo)
2244
      nodefiles.update((filename,
2245
                        frozenset(map(operator.attrgetter("name"), filenodes)))
2246
                       for filename in files)
2247

    
2248
    assert set(nodefiles) == (files_all | files_mc | files_vm)
2249

    
2250
    fileinfo = dict((filename, {}) for filename in nodefiles)
2251
    ignore_nodes = set()
2252

    
2253
    for node in nodeinfo:
2254
      if node.offline:
2255
        ignore_nodes.add(node.name)
2256
        continue
2257

    
2258
      nresult = all_nvinfo[node.name]
2259

    
2260
      if nresult.fail_msg or not nresult.payload:
2261
        node_files = None
2262
      else:
2263
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
2264

    
2265
      test = not (node_files and isinstance(node_files, dict))
2266
      errorif(test, constants.CV_ENODEFILECHECK, node.name,
2267
              "Node did not return file checksum data")
2268
      if test:
2269
        ignore_nodes.add(node.name)
2270
        continue
2271

    
2272
      # Build per-checksum mapping from filename to nodes having it
2273
      for (filename, checksum) in node_files.items():
2274
        assert filename in nodefiles
2275
        fileinfo[filename].setdefault(checksum, set()).add(node.name)
2276

    
2277
    for (filename, checksums) in fileinfo.items():
2278
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2279

    
2280
      # Nodes having the file
2281
      with_file = frozenset(node_name
2282
                            for nodes in fileinfo[filename].values()
2283
                            for node_name in nodes) - ignore_nodes
2284

    
2285
      expected_nodes = nodefiles[filename] - ignore_nodes
2286

    
2287
      # Nodes missing file
2288
      missing_file = expected_nodes - with_file
2289

    
2290
      if filename in files_opt:
2291
        # All or no nodes
2292
        errorif(missing_file and missing_file != expected_nodes,
2293
                constants.CV_ECLUSTERFILECHECK, None,
2294
                "File %s is optional, but it must exist on all or no"
2295
                " nodes (not found on %s)",
2296
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2297
      else:
2298
        errorif(missing_file, constants.CV_ECLUSTERFILECHECK, None,
2299
                "File %s is missing from node(s) %s", filename,
2300
                utils.CommaJoin(utils.NiceSort(missing_file)))
2301

    
2302
        # Warn if a node has a file it shouldn't
2303
        unexpected = with_file - expected_nodes
2304
        errorif(unexpected,
2305
                constants.CV_ECLUSTERFILECHECK, None,
2306
                "File %s should not exist on node(s) %s",
2307
                filename, utils.CommaJoin(utils.NiceSort(unexpected)))
2308

    
2309
      # See if there are multiple versions of the file
2310
      test = len(checksums) > 1
2311
      if test:
2312
        variants = ["variant %s on %s" %
2313
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2314
                    for (idx, (checksum, nodes)) in
2315
                      enumerate(sorted(checksums.items()))]
2316
      else:
2317
        variants = []
2318

    
2319
      errorif(test, constants.CV_ECLUSTERFILECHECK, None,
2320
              "File %s found with %s different checksums (%s)",
2321
              filename, len(checksums), "; ".join(variants))
2322

    
2323
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2324
                      drbd_map):
2325
    """Verifies and the node DRBD status.
2326

2327
    @type ninfo: L{objects.Node}
2328
    @param ninfo: the node to check
2329
    @param nresult: the remote results for the node
2330
    @param instanceinfo: the dict of instances
2331
    @param drbd_helper: the configured DRBD usermode helper
2332
    @param drbd_map: the DRBD map as returned by
2333
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2334

2335
    """
2336
    node = ninfo.name
2337
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2338

    
2339
    if drbd_helper:
2340
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2341
      test = (helper_result == None)
2342
      _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2343
               "no drbd usermode helper returned")
2344
      if helper_result:
2345
        status, payload = helper_result
2346
        test = not status
2347
        _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2348
                 "drbd usermode helper check unsuccessful: %s", payload)
2349
        test = status and (payload != drbd_helper)
2350
        _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
2351
                 "wrong drbd usermode helper: %s", payload)
2352

    
2353
    # compute the DRBD minors
2354
    node_drbd = {}
2355
    for minor, instance in drbd_map[node].items():
2356
      test = instance not in instanceinfo
2357
      _ErrorIf(test, constants.CV_ECLUSTERCFG, None,
2358
               "ghost instance '%s' in temporary DRBD map", instance)
2359
        # ghost instance should not be running, but otherwise we
2360
        # don't give double warnings (both ghost instance and
2361
        # unallocated minor in use)
2362
      if test:
2363
        node_drbd[minor] = (instance, False)
2364
      else:
2365
        instance = instanceinfo[instance]
2366
        node_drbd[minor] = (instance.name,
2367
                            instance.admin_state == constants.ADMINST_UP)
2368

    
2369
    # and now check them
2370
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2371
    test = not isinstance(used_minors, (tuple, list))
2372
    _ErrorIf(test, constants.CV_ENODEDRBD, node,
2373
             "cannot parse drbd status file: %s", str(used_minors))
2374
    if test:
2375
      # we cannot check drbd status
2376
      return
2377

    
2378
    for minor, (iname, must_exist) in node_drbd.items():
2379
      test = minor not in used_minors and must_exist
2380
      _ErrorIf(test, constants.CV_ENODEDRBD, node,
2381
               "drbd minor %d of instance %s is not active", minor, iname)
2382
    for minor in used_minors:
2383
      test = minor not in node_drbd
2384
      _ErrorIf(test, constants.CV_ENODEDRBD, node,
2385
               "unallocated drbd minor %d is in use", minor)
2386

    
2387
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2388
    """Builds the node OS structures.
2389

2390
    @type ninfo: L{objects.Node}
2391
    @param ninfo: the node to check
2392
    @param nresult: the remote results for the node
2393
    @param nimg: the node image object
2394

2395
    """
2396
    node = ninfo.name
2397
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2398

    
2399
    remote_os = nresult.get(constants.NV_OSLIST, None)
2400
    test = (not isinstance(remote_os, list) or
2401
            not compat.all(isinstance(v, list) and len(v) == 7
2402
                           for v in remote_os))
2403

    
2404
    _ErrorIf(test, constants.CV_ENODEOS, node,
2405
             "node hasn't returned valid OS data")
2406

    
2407
    nimg.os_fail = test
2408

    
2409
    if test:
2410
      return
2411

    
2412
    os_dict = {}
2413

    
2414
    for (name, os_path, status, diagnose,
2415
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2416

    
2417
      if name not in os_dict:
2418
        os_dict[name] = []
2419

    
2420
      # parameters is a list of lists instead of list of tuples due to
2421
      # JSON lacking a real tuple type, fix it:
2422
      parameters = [tuple(v) for v in parameters]
2423
      os_dict[name].append((os_path, status, diagnose,
2424
                            set(variants), set(parameters), set(api_ver)))
2425

    
2426
    nimg.oslist = os_dict
2427

    
2428
  def _VerifyNodeOS(self, ninfo, nimg, base):
2429
    """Verifies the node OS list.
2430

2431
    @type ninfo: L{objects.Node}
2432
    @param ninfo: the node to check
2433
    @param nimg: the node image object
2434
    @param base: the 'template' node we match against (e.g. from the master)
2435

2436
    """
2437
    node = ninfo.name
2438
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2439

    
2440
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2441

    
2442
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2443
    for os_name, os_data in nimg.oslist.items():
2444
      assert os_data, "Empty OS status for OS %s?!" % os_name
2445
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2446
      _ErrorIf(not f_status, constants.CV_ENODEOS, node,
2447
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2448
      _ErrorIf(len(os_data) > 1, constants.CV_ENODEOS, node,
2449
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2450
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2451
      # comparisons with the 'base' image
2452
      test = os_name not in base.oslist
2453
      _ErrorIf(test, constants.CV_ENODEOS, node,
2454
               "Extra OS %s not present on reference node (%s)",
2455
               os_name, base.name)
2456
      if test:
2457
        continue
2458
      assert base.oslist[os_name], "Base node has empty OS status?"
2459
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2460
      if not b_status:
2461
        # base OS is invalid, skipping
2462
        continue
2463
      for kind, a, b in [("API version", f_api, b_api),
2464
                         ("variants list", f_var, b_var),
2465
                         ("parameters", beautify_params(f_param),
2466
                          beautify_params(b_param))]:
2467
        _ErrorIf(a != b, constants.CV_ENODEOS, node,
2468
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2469
                 kind, os_name, base.name,
2470
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2471

    
2472
    # check any missing OSes
2473
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2474
    _ErrorIf(missing, constants.CV_ENODEOS, node,
2475
             "OSes present on reference node %s but missing on this node: %s",
2476
             base.name, utils.CommaJoin(missing))
2477

    
2478
  def _VerifyOob(self, ninfo, nresult):
2479
    """Verifies out of band functionality of a node.
2480

2481
    @type ninfo: L{objects.Node}
2482
    @param ninfo: the node to check
2483
    @param nresult: the remote results for the node
2484

2485
    """
2486
    node = ninfo.name
2487
    # We just have to verify the paths on master and/or master candidates
2488
    # as the oob helper is invoked on the master
2489
    if ((ninfo.master_candidate or ninfo.master_capable) and
2490
        constants.NV_OOB_PATHS in nresult):
2491
      for path_result in nresult[constants.NV_OOB_PATHS]:
2492
        self._ErrorIf(path_result, constants.CV_ENODEOOBPATH, node, path_result)
2493

    
2494
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2495
    """Verifies and updates the node volume data.
2496

2497
    This function will update a L{NodeImage}'s internal structures
2498
    with data from the remote call.
2499

2500
    @type ninfo: L{objects.Node}
2501
    @param ninfo: the node to check
2502
    @param nresult: the remote results for the node
2503
    @param nimg: the node image object
2504
    @param vg_name: the configured VG name
2505

2506
    """
2507
    node = ninfo.name
2508
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2509

    
2510
    nimg.lvm_fail = True
2511
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2512
    if vg_name is None:
2513
      pass
2514
    elif isinstance(lvdata, basestring):
2515
      _ErrorIf(True, constants.CV_ENODELVM, node, "LVM problem on node: %s",
2516
               utils.SafeEncode(lvdata))
2517
    elif not isinstance(lvdata, dict):
2518
      _ErrorIf(True, constants.CV_ENODELVM, node,
2519
               "rpc call to node failed (lvlist)")
2520
    else:
2521
      nimg.volumes = lvdata
2522
      nimg.lvm_fail = False
2523

    
2524
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2525
    """Verifies and updates the node instance list.
2526

2527
    If the listing was successful, then updates this node's instance
2528
    list. Otherwise, it marks the RPC call as failed for the instance
2529
    list key.
2530

2531
    @type ninfo: L{objects.Node}
2532
    @param ninfo: the node to check
2533
    @param nresult: the remote results for the node
2534
    @param nimg: the node image object
2535

2536
    """
2537
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2538
    test = not isinstance(idata, list)
2539
    self._ErrorIf(test, constants.CV_ENODEHV, ninfo.name,
2540
                  "rpc call to node failed (instancelist): %s",
2541
                  utils.SafeEncode(str(idata)))
2542
    if test:
2543
      nimg.hyp_fail = True
2544
    else:
2545
      nimg.instances = idata
2546

    
2547
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2548
    """Verifies and computes a node information map
2549

2550
    @type ninfo: L{objects.Node}
2551
    @param ninfo: the node to check
2552
    @param nresult: the remote results for the node
2553
    @param nimg: the node image object
2554
    @param vg_name: the configured VG name
2555

2556
    """
2557
    node = ninfo.name
2558
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2559

    
2560
    # try to read free memory (from the hypervisor)
2561
    hv_info = nresult.get(constants.NV_HVINFO, None)
2562
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2563
    _ErrorIf(test, constants.CV_ENODEHV, node,
2564
             "rpc call to node failed (hvinfo)")
2565
    if not test:
2566
      try:
2567
        nimg.mfree = int(hv_info["memory_free"])
2568
      except (ValueError, TypeError):
2569
        _ErrorIf(True, constants.CV_ENODERPC, node,
2570
                 "node returned invalid nodeinfo, check hypervisor")
2571

    
2572
    # FIXME: devise a free space model for file based instances as well
2573
    if vg_name is not None:
2574
      test = (constants.NV_VGLIST not in nresult or
2575
              vg_name not in nresult[constants.NV_VGLIST])
2576
      _ErrorIf(test, constants.CV_ENODELVM, node,
2577
               "node didn't return data for the volume group '%s'"
2578
               " - it is either missing or broken", vg_name)
2579
      if not test:
2580
        try:
2581
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2582
        except (ValueError, TypeError):
2583
          _ErrorIf(True, constants.CV_ENODERPC, node,
2584
                   "node returned invalid LVM info, check LVM status")
2585

    
2586
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2587
    """Gets per-disk status information for all instances.
2588

2589
    @type nodelist: list of strings
2590
    @param nodelist: Node names
2591
    @type node_image: dict of (name, L{objects.Node})
2592
    @param node_image: Node objects
2593
    @type instanceinfo: dict of (name, L{objects.Instance})
2594
    @param instanceinfo: Instance objects
2595
    @rtype: {instance: {node: [(succes, payload)]}}
2596
    @return: a dictionary of per-instance dictionaries with nodes as
2597
        keys and disk information as values; the disk information is a
2598
        list of tuples (success, payload)
2599

2600
    """
2601
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2602

    
2603
    node_disks = {}
2604
    node_disks_devonly = {}
2605
    diskless_instances = set()
2606
    diskless = constants.DT_DISKLESS
2607

    
2608
    for nname in nodelist:
2609
      node_instances = list(itertools.chain(node_image[nname].pinst,
2610
                                            node_image[nname].sinst))
2611
      diskless_instances.update(inst for inst in node_instances
2612
                                if instanceinfo[inst].disk_template == diskless)
2613
      disks = [(inst, disk)
2614
               for inst in node_instances
2615
               for disk in instanceinfo[inst].disks]
2616

    
2617
      if not disks:
2618
        # No need to collect data
2619
        continue
2620

    
2621
      node_disks[nname] = disks
2622

    
2623
      # Creating copies as SetDiskID below will modify the objects and that can
2624
      # lead to incorrect data returned from nodes
2625
      devonly = [dev.Copy() for (_, dev) in disks]
2626

    
2627
      for dev in devonly:
2628
        self.cfg.SetDiskID(dev, nname)
2629

    
2630
      node_disks_devonly[nname] = devonly
2631

    
2632
    assert len(node_disks) == len(node_disks_devonly)
2633

    
2634
    # Collect data from all nodes with disks
2635
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2636
                                                          node_disks_devonly)
2637

    
2638
    assert len(result) == len(node_disks)
2639

    
2640
    instdisk = {}
2641

    
2642
    for (nname, nres) in result.items():
2643
      disks = node_disks[nname]
2644

    
2645
      if nres.offline:
2646
        # No data from this node
2647
        data = len(disks) * [(False, "node offline")]
2648
      else:
2649
        msg = nres.fail_msg
2650
        _ErrorIf(msg, constants.CV_ENODERPC, nname,
2651
                 "while getting disk information: %s", msg)
2652
        if msg:
2653
          # No data from this node
2654
          data = len(disks) * [(False, msg)]
2655
        else:
2656
          data = []
2657
          for idx, i in enumerate(nres.payload):
2658
            if isinstance(i, (tuple, list)) and len(i) == 2:
2659
              data.append(i)
2660
            else:
2661
              logging.warning("Invalid result from node %s, entry %d: %s",
2662
                              nname, idx, i)
2663
              data.append((False, "Invalid result from the remote node"))
2664

    
2665
      for ((inst, _), status) in zip(disks, data):
2666
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2667

    
2668
    # Add empty entries for diskless instances.
2669
    for inst in diskless_instances:
2670
      assert inst not in instdisk
2671
      instdisk[inst] = {}
2672

    
2673
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2674
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2675
                      compat.all(isinstance(s, (tuple, list)) and
2676
                                 len(s) == 2 for s in statuses)
2677
                      for inst, nnames in instdisk.items()
2678
                      for nname, statuses in nnames.items())
2679
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2680

    
2681
    return instdisk
2682

    
2683
  @staticmethod
2684
  def _SshNodeSelector(group_uuid, all_nodes):
2685
    """Create endless iterators for all potential SSH check hosts.
2686

2687
    """
2688
    nodes = [node for node in all_nodes
2689
             if (node.group != group_uuid and
2690
                 not node.offline)]
2691
    keyfunc = operator.attrgetter("group")
2692

    
2693
    return map(itertools.cycle,
2694
               [sorted(map(operator.attrgetter("name"), names))
2695
                for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
2696
                                                  keyfunc)])
2697

    
2698
  @classmethod
2699
  def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
2700
    """Choose which nodes should talk to which other nodes.
2701

2702
    We will make nodes contact all nodes in their group, and one node from
2703
    every other group.
2704

2705
    @warning: This algorithm has a known issue if one node group is much
2706
      smaller than others (e.g. just one node). In such a case all other
2707
      nodes will talk to the single node.
2708

2709
    """
2710
    online_nodes = sorted(node.name for node in group_nodes if not node.offline)
2711
    sel = cls._SshNodeSelector(group_uuid, all_nodes)
2712

    
2713
    return (online_nodes,
2714
            dict((name, sorted([i.next() for i in sel]))
2715
                 for name in online_nodes))
2716

    
2717
  def BuildHooksEnv(self):
2718
    """Build hooks env.
2719

2720
    Cluster-Verify hooks just ran in the post phase and their failure makes
2721
    the output be logged in the verify output and the verification to fail.
2722

2723
    """
2724
    env = {
2725
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2726
      }
2727

    
2728
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2729
               for node in self.my_node_info.values())
2730

    
2731
    return env
2732

    
2733
  def BuildHooksNodes(self):
2734
    """Build hooks nodes.
2735

2736
    """
2737
    return ([], self.my_node_names)
2738

    
2739
  def Exec(self, feedback_fn):
2740
    """Verify integrity of the node group, performing various test on nodes.
2741

2742
    """
2743
    # This method has too many local variables. pylint: disable=R0914
2744
    feedback_fn("* Verifying group '%s'" % self.group_info.name)
2745

    
2746
    if not self.my_node_names:
2747
      # empty node group
2748
      feedback_fn("* Empty node group, skipping verification")
2749
      return True
2750

    
2751
    self.bad = False
2752
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2753
    verbose = self.op.verbose
2754
    self._feedback_fn = feedback_fn
2755

    
2756
    vg_name = self.cfg.GetVGName()
2757
    drbd_helper = self.cfg.GetDRBDHelper()
2758
    cluster = self.cfg.GetClusterInfo()
2759
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2760
    hypervisors = cluster.enabled_hypervisors
2761
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2762

    
2763
    i_non_redundant = [] # Non redundant instances
2764
    i_non_a_balanced = [] # Non auto-balanced instances
2765
    i_offline = 0 # Count of offline instances
2766
    n_offline = 0 # Count of offline nodes
2767
    n_drained = 0 # Count of nodes being drained
2768
    node_vol_should = {}
2769

    
2770
    # FIXME: verify OS list
2771

    
2772
    # File verification
2773
    filemap = _ComputeAncillaryFiles(cluster, False)
2774

    
2775
    # do local checksums
2776
    master_node = self.master_node = self.cfg.GetMasterNode()
2777
    master_ip = self.cfg.GetMasterIP()
2778

    
2779
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2780

    
2781
    user_scripts = []
2782
    if self.cfg.GetUseExternalMipScript():
2783
      user_scripts.append(constants.EXTERNAL_MASTER_SETUP_SCRIPT)
2784

    
2785
    node_verify_param = {
2786
      constants.NV_FILELIST:
2787
        utils.UniqueSequence(filename
2788
                             for files in filemap
2789
                             for filename in files),
2790
      constants.NV_NODELIST:
2791
        self._SelectSshCheckNodes(node_data_list, self.group_uuid,
2792
                                  self.all_node_info.values()),
2793
      constants.NV_HYPERVISOR: hypervisors,
2794
      constants.NV_HVPARAMS:
2795
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2796
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2797
                                 for node in node_data_list
2798
                                 if not node.offline],
2799
      constants.NV_INSTANCELIST: hypervisors,
2800
      constants.NV_VERSION: None,
2801
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2802
      constants.NV_NODESETUP: None,
2803
      constants.NV_TIME: None,
2804
      constants.NV_MASTERIP: (master_node, master_ip),
2805
      constants.NV_OSLIST: None,
2806
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2807
      constants.NV_USERSCRIPTS: user_scripts,
2808
      }
2809

    
2810
    if vg_name is not None:
2811
      node_verify_param[constants.NV_VGLIST] = None
2812
      node_verify_param[constants.NV_LVLIST] = vg_name
2813
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2814
      node_verify_param[constants.NV_DRBDLIST] = None
2815

    
2816
    if drbd_helper:
2817
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2818

    
2819
    # bridge checks
2820
    # FIXME: this needs to be changed per node-group, not cluster-wide
2821
    bridges = set()
2822
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2823
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2824
      bridges.add(default_nicpp[constants.NIC_LINK])
2825
    for instance in self.my_inst_info.values():
2826
      for nic in instance.nics:
2827
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2828
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2829
          bridges.add(full_nic[constants.NIC_LINK])
2830

    
2831
    if bridges:
2832
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2833

    
2834
    # Build our expected cluster state
2835
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2836
                                                 name=node.name,
2837
                                                 vm_capable=node.vm_capable))
2838
                      for node in node_data_list)
2839

    
2840
    # Gather OOB paths
2841
    oob_paths = []
2842
    for node in self.all_node_info.values():
2843
      path = _SupportsOob(self.cfg, node)
2844
      if path and path not in oob_paths:
2845
        oob_paths.append(path)
2846

    
2847
    if oob_paths:
2848
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2849

    
2850
    for instance in self.my_inst_names:
2851
      inst_config = self.my_inst_info[instance]
2852

    
2853
      for nname in inst_config.all_nodes:
2854
        if nname not in node_image:
2855
          gnode = self.NodeImage(name=nname)
2856
          gnode.ghost = (nname not in self.all_node_info)
2857
          node_image[nname] = gnode
2858

    
2859
      inst_config.MapLVsByNode(node_vol_should)
2860

    
2861
      pnode = inst_config.primary_node
2862
      node_image[pnode].pinst.append(instance)
2863

    
2864
      for snode in inst_config.secondary_nodes:
2865
        nimg = node_image[snode]
2866
        nimg.sinst.append(instance)
2867
        if pnode not in nimg.sbp:
2868
          nimg.sbp[pnode] = []
2869
        nimg.sbp[pnode].append(instance)
2870

    
2871
    # At this point, we have the in-memory data structures complete,
2872
    # except for the runtime information, which we'll gather next
2873

    
2874
    # Due to the way our RPC system works, exact response times cannot be
2875
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2876
    # time before and after executing the request, we can at least have a time
2877
    # window.
2878
    nvinfo_starttime = time.time()
2879
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2880
                                           node_verify_param,
2881
                                           self.cfg.GetClusterName())
2882
    nvinfo_endtime = time.time()
2883

    
2884
    if self.extra_lv_nodes and vg_name is not None:
2885
      extra_lv_nvinfo = \
2886
          self.rpc.call_node_verify(self.extra_lv_nodes,
2887
                                    {constants.NV_LVLIST: vg_name},
2888
                                    self.cfg.GetClusterName())
2889
    else:
2890
      extra_lv_nvinfo = {}
2891

    
2892
    all_drbd_map = self.cfg.ComputeDRBDMap()
2893

    
2894
    feedback_fn("* Gathering disk information (%s nodes)" %
2895
                len(self.my_node_names))
2896
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2897
                                     self.my_inst_info)
2898

    
2899
    feedback_fn("* Verifying configuration file consistency")
2900

    
2901
    # If not all nodes are being checked, we need to make sure the master node
2902
    # and a non-checked vm_capable node are in the list.
2903
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2904
    if absent_nodes:
2905
      vf_nvinfo = all_nvinfo.copy()
2906
      vf_node_info = list(self.my_node_info.values())
2907
      additional_nodes = []
2908
      if master_node not in self.my_node_info:
2909
        additional_nodes.append(master_node)
2910
        vf_node_info.append(self.all_node_info[master_node])
2911
      # Add the first vm_capable node we find which is not included
2912
      for node in absent_nodes:
2913
        nodeinfo = self.all_node_info[node]
2914
        if nodeinfo.vm_capable and not nodeinfo.offline:
2915
          additional_nodes.append(node)
2916
          vf_node_info.append(self.all_node_info[node])
2917
          break
2918
      key = constants.NV_FILELIST
2919
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2920
                                                 {key: node_verify_param[key]},
2921
                                                 self.cfg.GetClusterName()))
2922
    else:
2923
      vf_nvinfo = all_nvinfo
2924
      vf_node_info = self.my_node_info.values()
2925

    
2926
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2927

    
2928
    feedback_fn("* Verifying node status")
2929

    
2930
    refos_img = None
2931

    
2932
    for node_i in node_data_list:
2933
      node = node_i.name
2934
      nimg = node_image[node]
2935

    
2936
      if node_i.offline:
2937
        if verbose:
2938
          feedback_fn("* Skipping offline node %s" % (node,))
2939
        n_offline += 1
2940
        continue
2941

    
2942
      if node == master_node:
2943
        ntype = "master"
2944
      elif node_i.master_candidate:
2945
        ntype = "master candidate"
2946
      elif node_i.drained:
2947
        ntype = "drained"
2948
        n_drained += 1
2949
      else:
2950
        ntype = "regular"
2951
      if verbose:
2952
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2953

    
2954
      msg = all_nvinfo[node].fail_msg
2955
      _ErrorIf(msg, constants.CV_ENODERPC, node, "while contacting node: %s",
2956
               msg)
2957
      if msg:
2958
        nimg.rpc_fail = True
2959
        continue
2960

    
2961
      nresult = all_nvinfo[node].payload
2962

    
2963
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2964
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2965
      self._VerifyNodeNetwork(node_i, nresult)
2966
      self._VerifyNodeUserScripts(node_i, nresult)
2967
      self._VerifyOob(node_i, nresult)
2968

    
2969
      if nimg.vm_capable:
2970
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2971
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2972
                             all_drbd_map)
2973

    
2974
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2975
        self._UpdateNodeInstances(node_i, nresult, nimg)
2976
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2977
        self._UpdateNodeOS(node_i, nresult, nimg)
2978

    
2979
        if not nimg.os_fail:
2980
          if refos_img is None:
2981
            refos_img = nimg
2982
          self._VerifyNodeOS(node_i, nimg, refos_img)
2983
        self._VerifyNodeBridges(node_i, nresult, bridges)
2984

    
2985
        # Check whether all running instancies are primary for the node. (This
2986
        # can no longer be done from _VerifyInstance below, since some of the
2987
        # wrong instances could be from other node groups.)
2988
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2989

    
2990
        for inst in non_primary_inst:
2991
          # FIXME: investigate best way to handle offline insts
2992
          if inst.admin_state == constants.ADMINST_OFFLINE:
2993
            if verbose:
2994
              feedback_fn("* Skipping offline instance %s" % inst.name)
2995
            i_offline += 1
2996
            continue
2997
          test = inst in self.all_inst_info
2998
          _ErrorIf(test, constants.CV_EINSTANCEWRONGNODE, inst,
2999
                   "instance should not run on node %s", node_i.name)
3000
          _ErrorIf(not test, constants.CV_ENODEORPHANINSTANCE, node_i.name,
3001
                   "node is running unknown instance %s", inst)
3002

    
3003
    for node, result in extra_lv_nvinfo.items():
3004
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
3005
                              node_image[node], vg_name)
3006

    
3007
    feedback_fn("* Verifying instance status")
3008
    for instance in self.my_inst_names:
3009
      if verbose:
3010
        feedback_fn("* Verifying instance %s" % instance)
3011
      inst_config = self.my_inst_info[instance]
3012
      self._VerifyInstance(instance, inst_config, node_image,
3013
                           instdisk[instance])
3014
      inst_nodes_offline = []
3015

    
3016
      pnode = inst_config.primary_node
3017
      pnode_img = node_image[pnode]
3018
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
3019
               constants.CV_ENODERPC, pnode, "instance %s, connection to"
3020
               " primary node failed", instance)
3021

    
3022
      _ErrorIf(inst_config.admin_state == constants.ADMINST_UP and
3023
               pnode_img.offline,
3024
               constants.CV_EINSTANCEBADNODE, instance,
3025
               "instance is marked as running and lives on offline node %s",
3026
               inst_config.primary_node)
3027

    
3028
      # If the instance is non-redundant we cannot survive losing its primary
3029
      # node, so we are not N+1 compliant. On the other hand we have no disk
3030
      # templates with more than one secondary so that situation is not well
3031
      # supported either.
3032
      # FIXME: does not support file-backed instances
3033
      if not inst_config.secondary_nodes:
3034
        i_non_redundant.append(instance)
3035

    
3036
      _ErrorIf(len(inst_config.secondary_nodes) > 1,
3037
               constants.CV_EINSTANCELAYOUT,
3038
               instance, "instance has multiple secondary nodes: %s",
3039
               utils.CommaJoin(inst_config.secondary_nodes),
3040
               code=self.ETYPE_WARNING)
3041

    
3042
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
3043
        pnode = inst_config.primary_node
3044
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
3045
        instance_groups = {}
3046

    
3047
        for node in instance_nodes:
3048
          instance_groups.setdefault(self.all_node_info[node].group,
3049
                                     []).append(node)
3050

    
3051
        pretty_list = [
3052
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
3053
          # Sort so that we always list the primary node first.
3054
          for group, nodes in sorted(instance_groups.items(),
3055
                                     key=lambda (_, nodes): pnode in nodes,
3056
                                     reverse=True)]
3057

    
3058
        self._ErrorIf(len(instance_groups) > 1,
3059
                      constants.CV_EINSTANCESPLITGROUPS,
3060
                      instance, "instance has primary and secondary nodes in"
3061
                      " different groups: %s", utils.CommaJoin(pretty_list),
3062
                      code=self.ETYPE_WARNING)
3063

    
3064
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
3065
        i_non_a_balanced.append(instance)
3066

    
3067
      for snode in inst_config.secondary_nodes:
3068
        s_img = node_image[snode]
3069
        _ErrorIf(s_img.rpc_fail and not s_img.offline, constants.CV_ENODERPC,
3070
                 snode, "instance %s, connection to secondary node failed",
3071
                 instance)
3072

    
3073
        if s_img.offline:
3074
          inst_nodes_offline.append(snode)
3075

    
3076
      # warn that the instance lives on offline nodes
3077
      _ErrorIf(inst_nodes_offline, constants.CV_EINSTANCEBADNODE, instance,
3078
               "instance has offline secondary node(s) %s",
3079
               utils.CommaJoin(inst_nodes_offline))
3080
      # ... or ghost/non-vm_capable nodes
3081
      for node in inst_config.all_nodes:
3082
        _ErrorIf(node_image[node].ghost, constants.CV_EINSTANCEBADNODE,
3083
                 instance, "instance lives on ghost node %s", node)
3084
        _ErrorIf(not node_image[node].vm_capable, constants.CV_EINSTANCEBADNODE,
3085
                 instance, "instance lives on non-vm_capable node %s", node)
3086

    
3087
    feedback_fn("* Verifying orphan volumes")
3088
    reserved = utils.FieldSet(*cluster.reserved_lvs)
3089

    
3090
    # We will get spurious "unknown volume" warnings if any node of this group
3091
    # is secondary for an instance whose primary is in another group. To avoid
3092
    # them, we find these instances and add their volumes to node_vol_should.
3093
    for inst in self.all_inst_info.values():
3094
      for secondary in inst.secondary_nodes:
3095
        if (secondary in self.my_node_info
3096
            and inst.name not in self.my_inst_info):
3097
          inst.MapLVsByNode(node_vol_should)
3098
          break
3099

    
3100
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
3101

    
3102
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
3103
      feedback_fn("* Verifying N+1 Memory redundancy")
3104
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
3105

    
3106
    feedback_fn("* Other Notes")
3107
    if i_non_redundant:
3108
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
3109
                  % len(i_non_redundant))
3110

    
3111
    if i_non_a_balanced:
3112
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
3113
                  % len(i_non_a_balanced))
3114

    
3115
    if i_offline:
3116
      feedback_fn("  - NOTICE: %d offline instance(s) found." % i_offline)
3117

    
3118
    if n_offline:
3119
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
3120

    
3121
    if n_drained:
3122
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
3123

    
3124
    return not self.bad
3125

    
3126
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
3127
    """Analyze the post-hooks' result
3128

3129
    This method analyses the hook result, handles it, and sends some
3130
    nicely-formatted feedback back to the user.
3131

3132
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
3133
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
3134
    @param hooks_results: the results of the multi-node hooks rpc call
3135
    @param feedback_fn: function used send feedback back to the caller
3136
    @param lu_result: previous Exec result
3137
    @return: the new Exec result, based on the previous result
3138
        and hook results
3139

3140
    """
3141
    # We only really run POST phase hooks, only for non-empty groups,
3142
    # and are only interested in their results
3143
    if not self.my_node_names:
3144
      # empty node group
3145
      pass
3146
    elif phase == constants.HOOKS_PHASE_POST:
3147
      # Used to change hooks' output to proper indentation
3148
      feedback_fn("* Hooks Results")
3149
      assert hooks_results, "invalid result from hooks"
3150

    
3151
      for node_name in hooks_results:
3152
        res = hooks_results[node_name]
3153
        msg = res.fail_msg
3154
        test = msg and not res.offline
3155
        self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
3156
                      "Communication failure in hooks execution: %s", msg)
3157
        if res.offline or msg:
3158
          # No need to investigate payload if node is offline or gave
3159
          # an error.
3160
          continue
3161
        for script, hkr, output in res.payload:
3162
          test = hkr == constants.HKR_FAIL
3163
          self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
3164
                        "Script %s failed, output:", script)
3165
          if test:
3166
            output = self._HOOKS_INDENT_RE.sub("      ", output)
3167
            feedback_fn("%s" % output)
3168
            lu_result = False
3169

    
3170
    return lu_result
3171

    
3172

    
3173
class LUClusterVerifyDisks(NoHooksLU):
3174
  """Verifies the cluster disks status.
3175

3176
  """
3177
  REQ_BGL = False
3178

    
3179
  def ExpandNames(self):
3180
    self.share_locks = _ShareAll()
3181
    self.needed_locks = {
3182
      locking.LEVEL_NODEGROUP: locking.ALL_SET,
3183
      }
3184

    
3185
  def Exec(self, feedback_fn):
3186
    group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3187

    
3188
    # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3189
    return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3190
                           for group in group_names])
3191

    
3192

    
3193
class LUGroupVerifyDisks(NoHooksLU):
3194
  """Verifies the status of all disks in a node group.
3195

3196
  """
3197
  REQ_BGL = False
3198

    
3199
  def ExpandNames(self):
3200
    # Raises errors.OpPrereqError on its own if group can't be found
3201
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3202

    
3203
    self.share_locks = _ShareAll()
3204
    self.needed_locks = {
3205
      locking.LEVEL_INSTANCE: [],
3206
      locking.LEVEL_NODEGROUP: [],
3207
      locking.LEVEL_NODE: [],
3208
      }
3209

    
3210
  def DeclareLocks(self, level):
3211
    if level == locking.LEVEL_INSTANCE:
3212
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
3213

    
3214
      # Lock instances optimistically, needs verification once node and group
3215
      # locks have been acquired
3216
      self.needed_locks[locking.LEVEL_INSTANCE] = \
3217
        self.cfg.GetNodeGroupInstances(self.group_uuid)
3218

    
3219
    elif level == locking.LEVEL_NODEGROUP:
3220
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3221

    
3222
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
3223
        set([self.group_uuid] +
3224
            # Lock all groups used by instances optimistically; this requires
3225
            # going via the node before it's locked, requiring verification
3226
            # later on
3227
            [group_uuid
3228
             for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3229
             for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3230

    
3231
    elif level == locking.LEVEL_NODE:
3232
      # This will only lock the nodes in the group to be verified which contain
3233
      # actual instances
3234
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3235
      self._LockInstancesNodes()
3236

    
3237
      # Lock all nodes in group to be verified
3238
      assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3239
      member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3240
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3241

    
3242
  def CheckPrereq(self):
3243
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3244
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3245
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3246

    
3247
    assert self.group_uuid in owned_groups
3248

    
3249
    # Check if locked instances are still correct
3250
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3251

    
3252
    # Get instance information
3253
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3254

    
3255
    # Check if node groups for locked instances are still correct
3256
    for (instance_name, inst) in self.instances.items():
3257
      assert owned_nodes.issuperset(inst.all_nodes), \
3258
        "Instance %s's nodes changed while we kept the lock" % instance_name
3259

    
3260
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
3261
                                             owned_groups)
3262

    
3263
      assert self.group_uuid in inst_groups, \
3264
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
3265

    
3266
  def Exec(self, feedback_fn):
3267
    """Verify integrity of cluster disks.
3268

3269
    @rtype: tuple of three items
3270
    @return: a tuple of (dict of node-to-node_error, list of instances
3271
        which need activate-disks, dict of instance: (node, volume) for
3272
        missing volumes
3273

3274
    """
3275
    res_nodes = {}
3276
    res_instances = set()
3277
    res_missing = {}
3278

    
3279
    nv_dict = _MapInstanceDisksToNodes([inst
3280
            for inst in self.instances.values()
3281
            if inst.admin_state == constants.ADMINST_UP])
3282

    
3283
    if nv_dict:
3284
      nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3285
                             set(self.cfg.GetVmCapableNodeList()))
3286

    
3287
      node_lvs = self.rpc.call_lv_list(nodes, [])
3288

    
3289
      for (node, node_res) in node_lvs.items():
3290
        if node_res.offline:
3291
          continue
3292

    
3293
        msg = node_res.fail_msg
3294
        if msg:
3295
          logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3296
          res_nodes[node] = msg
3297
          continue
3298

    
3299
        for lv_name, (_, _, lv_online) in node_res.payload.items():
3300
          inst = nv_dict.pop((node, lv_name), None)
3301
          if not (lv_online or inst is None):
3302
            res_instances.add(inst)
3303

    
3304
      # any leftover items in nv_dict are missing LVs, let's arrange the data
3305
      # better
3306
      for key, inst in nv_dict.iteritems():
3307
        res_missing.setdefault(inst, []).append(list(key))
3308

    
3309
    return (res_nodes, list(res_instances), res_missing)
3310

    
3311

    
3312
class LUClusterRepairDiskSizes(NoHooksLU):
3313
  """Verifies the cluster disks sizes.
3314

3315
  """
3316
  REQ_BGL = False
3317

    
3318
  def ExpandNames(self):
3319
    if self.op.instances:
3320
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
3321
      self.needed_locks = {
3322
        locking.LEVEL_NODE_RES: [],
3323
        locking.LEVEL_INSTANCE: self.wanted_names,
3324
        }
3325
      self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
3326
    else:
3327
      self.wanted_names = None
3328
      self.needed_locks = {
3329
        locking.LEVEL_NODE_RES: locking.ALL_SET,
3330
        locking.LEVEL_INSTANCE: locking.ALL_SET,
3331
        }
3332
    self.share_locks = {
3333
      locking.LEVEL_NODE_RES: 1,
3334
      locking.LEVEL_INSTANCE: 0,
3335
      }
3336

    
3337
  def DeclareLocks(self, level):
3338
    if level == locking.LEVEL_NODE_RES and self.wanted_names is not None:
3339
      self._LockInstancesNodes(primary_only=True, level=level)
3340

    
3341
  def CheckPrereq(self):
3342
    """Check prerequisites.
3343

3344
    This only checks the optional instance list against the existing names.
3345

3346
    """
3347
    if self.wanted_names is None:
3348
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3349

    
3350
    self.wanted_instances = \
3351
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3352

    
3353
  def _EnsureChildSizes(self, disk):
3354
    """Ensure children of the disk have the needed disk size.
3355

3356
    This is valid mainly for DRBD8 and fixes an issue where the
3357
    children have smaller disk size.
3358

3359
    @param disk: an L{ganeti.objects.Disk} object
3360

3361
    """
3362
    if disk.dev_type == constants.LD_DRBD8:
3363
      assert disk.children, "Empty children for DRBD8?"
3364
      fchild = disk.children[0]
3365
      mismatch = fchild.size < disk.size
3366
      if mismatch:
3367
        self.LogInfo("Child disk has size %d, parent %d, fixing",
3368
                     fchild.size, disk.size)
3369
        fchild.size = disk.size
3370

    
3371
      # and we recurse on this child only, not on the metadev
3372
      return self._EnsureChildSizes(fchild) or mismatch
3373
    else:
3374
      return False
3375

    
3376
  def Exec(self, feedback_fn):
3377
    """Verify the size of cluster disks.
3378

3379
    """
3380
    # TODO: check child disks too
3381
    # TODO: check differences in size between primary/secondary nodes
3382
    per_node_disks = {}
3383
    for instance in self.wanted_instances:
3384
      pnode = instance.primary_node
3385
      if pnode not in per_node_disks:
3386
        per_node_disks[pnode] = []
3387
      for idx, disk in enumerate(instance.disks):
3388
        per_node_disks[pnode].append((instance, idx, disk))
3389

    
3390
    assert not (frozenset(per_node_disks.keys()) -
3391
                self.owned_locks(locking.LEVEL_NODE_RES)), \
3392
      "Not owning correct locks"
3393
    assert not self.owned_locks(locking.LEVEL_NODE)
3394

    
3395
    changed = []
3396
    for node, dskl in per_node_disks.items():
3397
      newl = [v[2].Copy() for v in dskl]
3398
      for dsk in newl:
3399
        self.cfg.SetDiskID(dsk, node)
3400
      result = self.rpc.call_blockdev_getsize(node, newl)
3401
      if result.fail_msg:
3402
        self.LogWarning("Failure in blockdev_getsize call to node"
3403
                        " %s, ignoring", node)
3404
        continue
3405
      if len(result.payload) != len(dskl):
3406
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
3407
                        " result.payload=%s", node, len(dskl), result.payload)
3408
        self.LogWarning("Invalid result from node %s, ignoring node results",
3409
                        node)
3410
        continue
3411
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
3412
        if size is None:
3413
          self.LogWarning("Disk %d of instance %s did not return size"
3414
                          " information, ignoring", idx, instance.name)
3415
          continue
3416
        if not isinstance(size, (int, long)):
3417
          self.LogWarning("Disk %d of instance %s did not return valid"
3418
                          " size information, ignoring", idx, instance.name)
3419
          continue
3420
        size = size >> 20
3421
        if size != disk.size:
3422
          self.LogInfo("Disk %d of instance %s has mismatched size,"
3423
                       " correcting: recorded %d, actual %d", idx,
3424
                       instance.name, disk.size, size)
3425
          disk.size = size
3426
          self.cfg.Update(instance, feedback_fn)
3427
          changed.append((instance.name, idx, size))
3428
        if self._EnsureChildSizes(disk):
3429
          self.cfg.Update(instance, feedback_fn)
3430
          changed.append((instance.name, idx, disk.size))
3431
    return changed
3432

    
3433

    
3434
class LUClusterRename(LogicalUnit):
3435
  """Rename the cluster.
3436

3437
  """
3438
  HPATH = "cluster-rename"
3439
  HTYPE = constants.HTYPE_CLUSTER
3440

    
3441
  def BuildHooksEnv(self):
3442
    """Build hooks env.
3443

3444
    """
3445
    return {
3446
      "OP_TARGET": self.cfg.GetClusterName(),
3447
      "NEW_NAME": self.op.name,
3448
      }
3449

    
3450
  def BuildHooksNodes(self):
3451
    """Build hooks nodes.
3452

3453
    """
3454
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3455

    
3456
  def CheckPrereq(self):
3457
    """Verify that the passed name is a valid one.
3458

3459
    """
3460
    hostname = netutils.GetHostname(name=self.op.name,
3461
                                    family=self.cfg.GetPrimaryIPFamily())
3462

    
3463
    new_name = hostname.name
3464
    self.ip = new_ip = hostname.ip
3465
    old_name = self.cfg.GetClusterName()
3466
    old_ip = self.cfg.GetMasterIP()
3467
    if new_name == old_name and new_ip == old_ip:
3468
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3469
                                 " cluster has changed",
3470
                                 errors.ECODE_INVAL)
3471
    if new_ip != old_ip:
3472
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3473
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3474
                                   " reachable on the network" %
3475
                                   new_ip, errors.ECODE_NOTUNIQUE)
3476

    
3477
    self.op.name = new_name
3478

    
3479
  def Exec(self, feedback_fn):
3480
    """Rename the cluster.
3481

3482
    """
3483
    clustername = self.op.name
3484
    new_ip = self.ip
3485

    
3486
    # shutdown the master IP
3487
    master_params = self.cfg.GetMasterNetworkParameters()
3488
    ems = self.cfg.GetUseExternalMipScript()
3489
    result = self.rpc.call_node_deactivate_master_ip(master_params.name,
3490
                                                     master_params, ems)
3491
    result.Raise("Could not disable the master role")
3492

    
3493
    try:
3494
      cluster = self.cfg.GetClusterInfo()
3495
      cluster.cluster_name = clustername
3496
      cluster.master_ip = new_ip
3497
      self.cfg.Update(cluster, feedback_fn)
3498

    
3499
      # update the known hosts file
3500
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3501
      node_list = self.cfg.GetOnlineNodeList()
3502
      try:
3503
        node_list.remove(master_params.name)
3504
      except ValueError:
3505
        pass
3506
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3507
    finally:
3508
      master_params.ip = new_ip
3509
      result = self.rpc.call_node_activate_master_ip(master_params.name,
3510
                                                     master_params, ems)
3511
      msg = result.fail_msg
3512
      if msg:
3513
        self.LogWarning("Could not re-enable the master role on"
3514
                        " the master, please restart manually: %s", msg)
3515

    
3516
    return clustername
3517

    
3518

    
3519
def _ValidateNetmask(cfg, netmask):
3520
  """Checks if a netmask is valid.
3521

3522
  @type cfg: L{config.ConfigWriter}
3523
  @param cfg: The cluster configuration
3524
  @type netmask: int
3525
  @param netmask: the netmask to be verified
3526
  @raise errors.OpPrereqError: if the validation fails
3527

3528
  """
3529
  ip_family = cfg.GetPrimaryIPFamily()
3530
  try:
3531
    ipcls = netutils.IPAddress.GetClassFromIpFamily(ip_family)
3532
  except errors.ProgrammerError:
3533
    raise errors.OpPrereqError("Invalid primary ip family: %s." %
3534
                               ip_family)
3535
  if not ipcls.ValidateNetmask(netmask):
3536
    raise errors.OpPrereqError("CIDR netmask (%s) not valid" %
3537
                                (netmask))
3538

    
3539

    
3540
class LUClusterSetParams(LogicalUnit):
3541
  """Change the parameters of the cluster.
3542

3543
  """
3544
  HPATH = "cluster-modify"
3545
  HTYPE = constants.HTYPE_CLUSTER
3546
  REQ_BGL = False
3547

    
3548
  def CheckArguments(self):
3549
    """Check parameters
3550

3551
    """
3552
    if self.op.uid_pool:
3553
      uidpool.CheckUidPool(self.op.uid_pool)
3554

    
3555
    if self.op.add_uids:
3556
      uidpool.CheckUidPool(self.op.add_uids)
3557

    
3558
    if self.op.remove_uids:
3559
      uidpool.CheckUidPool(self.op.remove_uids)
3560

    
3561
    if self.op.master_netmask is not None:
3562
      _ValidateNetmask(self.cfg, self.op.master_netmask)
3563

    
3564
    if self.op.diskparams:
3565
      for dt_params in self.op.diskparams.values():
3566
        utils.ForceDictType(dt_params, constants.DISK_DT_TYPES)
3567

    
3568
  def ExpandNames(self):
3569
    # FIXME: in the future maybe other cluster params won't require checking on
3570
    # all nodes to be modified.
3571
    self.needed_locks = {
3572
      locking.LEVEL_NODE: locking.ALL_SET,
3573
    }
3574
    self.share_locks[locking.LEVEL_NODE] = 1
3575

    
3576
  def BuildHooksEnv(self):
3577
    """Build hooks env.
3578

3579
    """
3580
    return {
3581
      "OP_TARGET": self.cfg.GetClusterName(),
3582
      "NEW_VG_NAME": self.op.vg_name,
3583
      }
3584

    
3585
  def BuildHooksNodes(self):
3586
    """Build hooks nodes.
3587

3588
    """
3589
    mn = self.cfg.GetMasterNode()
3590
    return ([mn], [mn])
3591

    
3592
  def CheckPrereq(self):
3593
    """Check prerequisites.
3594

3595
    This checks whether the given params don't conflict and
3596
    if the given volume group is valid.
3597

3598
    """
3599
    if self.op.vg_name is not None and not self.op.vg_name:
3600
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3601
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3602
                                   " instances exist", errors.ECODE_INVAL)
3603

    
3604
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3605
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3606
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3607
                                   " drbd-based instances exist",
3608
                                   errors.ECODE_INVAL)
3609

    
3610
    node_list = self.owned_locks(locking.LEVEL_NODE)
3611

    
3612
    # if vg_name not None, checks given volume group on all nodes
3613
    if self.op.vg_name:
3614
      vglist = self.rpc.call_vg_list(node_list)
3615
      for node in node_list:
3616
        msg = vglist[node].fail_msg
3617
        if msg:
3618
          # ignoring down node
3619
          self.LogWarning("Error while gathering data on node %s"
3620
                          " (ignoring node): %s", node, msg)
3621
          continue
3622
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3623
                                              self.op.vg_name,
3624
                                              constants.MIN_VG_SIZE)
3625
        if vgstatus:
3626
          raise errors.OpPrereqError("Error on node '%s': %s" %
3627
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3628

    
3629
    if self.op.drbd_helper:
3630
      # checks given drbd helper on all nodes
3631
      helpers = self.rpc.call_drbd_helper(node_list)
3632
      for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3633
        if ninfo.offline:
3634
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3635
          continue
3636
        msg = helpers[node].fail_msg
3637
        if msg:
3638
          raise errors.OpPrereqError("Error checking drbd helper on node"
3639
                                     " '%s': %s" % (node, msg),
3640
                                     errors.ECODE_ENVIRON)
3641
        node_helper = helpers[node].payload
3642
        if node_helper != self.op.drbd_helper:
3643
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3644
                                     (node, node_helper), errors.ECODE_ENVIRON)
3645

    
3646
    self.cluster = cluster = self.cfg.GetClusterInfo()
3647
    # validate params changes
3648
    if self.op.beparams:
3649
      objects.UpgradeBeParams(self.op.beparams)
3650
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3651
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3652

    
3653
    if self.op.ndparams:
3654
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3655
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3656

    
3657
      # TODO: we need a more general way to handle resetting
3658
      # cluster-level parameters to default values
3659
      if self.new_ndparams["oob_program"] == "":
3660
        self.new_ndparams["oob_program"] = \
3661
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3662

    
3663
    if self.op.hv_state:
3664
      new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
3665
                                            self.cluster.hv_state_static)
3666
      self.new_hv_state = dict((hv, cluster.SimpleFillHvState(values))
3667
                               for hv, values in new_hv_state.items())
3668

    
3669
    if self.op.disk_state:
3670
      new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state,
3671
                                                self.cluster.disk_state_static)
3672
      self.new_disk_state = \
3673
        dict((storage, dict((name, cluster.SimpleFillDiskState(values))
3674
                            for name, values in svalues.items()))
3675
             for storage, svalues in new_disk_state.items())
3676

    
3677
    if self.op.nicparams:
3678
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3679
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3680
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3681
      nic_errors = []
3682

    
3683
      # check all instances for consistency
3684
      for instance in self.cfg.GetAllInstancesInfo().values():
3685
        for nic_idx, nic in enumerate(instance.nics):
3686
          params_copy = copy.deepcopy(nic.nicparams)
3687
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3688

    
3689
          # check parameter syntax
3690
          try:
3691
            objects.NIC.CheckParameterSyntax(params_filled)
3692
          except errors.ConfigurationError, err:
3693
            nic_errors.append("Instance %s, nic/%d: %s" %
3694
                              (instance.name, nic_idx, err))
3695

    
3696
          # if we're moving instances to routed, check that they have an ip
3697
          target_mode = params_filled[constants.NIC_MODE]
3698
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3699
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3700
                              " address" % (instance.name, nic_idx))
3701
      if nic_errors:
3702
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3703
                                   "\n".join(nic_errors))
3704

    
3705
    # hypervisor list/parameters
3706
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3707
    if self.op.hvparams:
3708
      for hv_name, hv_dict in self.op.hvparams.items():
3709
        if hv_name not in self.new_hvparams:
3710
          self.new_hvparams[hv_name] = hv_dict
3711
        else:
3712
          self.new_hvparams[hv_name].update(hv_dict)
3713

    
3714
    # disk template parameters
3715
    self.new_diskparams = objects.FillDict(cluster.diskparams, {})
3716
    if self.op.diskparams:
3717
      for dt_name, dt_params in self.op.diskparams.items():
3718
        if dt_name not in self.op.diskparams:
3719
          self.new_diskparams[dt_name] = dt_params
3720
        else:
3721
          self.new_diskparams[dt_name].update(dt_params)
3722

    
3723
    # os hypervisor parameters
3724
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3725
    if self.op.os_hvp:
3726
      for os_name, hvs in self.op.os_hvp.items():
3727
        if os_name not in self.new_os_hvp:
3728
          self.new_os_hvp[os_name] = hvs
3729
        else:
3730
          for hv_name, hv_dict in hvs.items():
3731
            if hv_name not in self.new_os_hvp[os_name]:
3732
              self.new_os_hvp[os_name][hv_name] = hv_dict
3733
            else:
3734
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3735

    
3736
    # os parameters
3737
    self.new_osp = objects.FillDict(cluster.osparams, {})
3738
    if self.op.osparams:
3739
      for os_name, osp in self.op.osparams.items():
3740
        if os_name not in self.new_osp:
3741
          self.new_osp[os_name] = {}
3742

    
3743
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3744
                                                  use_none=True)
3745

    
3746
        if not self.new_osp[os_name]:
3747
          # we removed all parameters
3748
          del self.new_osp[os_name]
3749
        else:
3750
          # check the parameter validity (remote check)
3751
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3752
                         os_name, self.new_osp[os_name])
3753

    
3754
    # changes to the hypervisor list
3755
    if self.op.enabled_hypervisors is not None:
3756
      self.hv_list = self.op.enabled_hypervisors
3757
      for hv in self.hv_list:
3758
        # if the hypervisor doesn't already exist in the cluster
3759
        # hvparams, we initialize it to empty, and then (in both
3760
        # cases) we make sure to fill the defaults, as we might not
3761
        # have a complete defaults list if the hypervisor wasn't
3762
        # enabled before
3763
        if hv not in new_hvp:
3764
          new_hvp[hv] = {}
3765
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3766
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3767
    else:
3768
      self.hv_list = cluster.enabled_hypervisors
3769

    
3770
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3771
      # either the enabled list has changed, or the parameters have, validate
3772
      for hv_name, hv_params in self.new_hvparams.items():
3773
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3774
            (self.op.enabled_hypervisors and
3775
             hv_name in self.op.enabled_hypervisors)):
3776
          # either this is a new hypervisor, or its parameters have changed
3777
          hv_class = hypervisor.GetHypervisor(hv_name)
3778
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3779
          hv_class.CheckParameterSyntax(hv_params)
3780
          _CheckHVParams(self, node_list, hv_name, hv_params)
3781

    
3782
    if self.op.os_hvp:
3783
      # no need to check any newly-enabled hypervisors, since the
3784
      # defaults have already been checked in the above code-block
3785
      for os_name, os_hvp in self.new_os_hvp.items():
3786
        for hv_name, hv_params in os_hvp.items():
3787
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3788
          # we need to fill in the new os_hvp on top of the actual hv_p
3789
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3790
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3791
          hv_class = hypervisor.GetHypervisor(hv_name)
3792
          hv_class.CheckParameterSyntax(new_osp)
3793
          _CheckHVParams(self, node_list, hv_name, new_osp)
3794

    
3795
    if self.op.default_iallocator:
3796
      alloc_script = utils.FindFile(self.op.default_iallocator,
3797
                                    constants.IALLOCATOR_SEARCH_PATH,
3798
                                    os.path.isfile)
3799
      if alloc_script is None:
3800
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3801
                                   " specified" % self.op.default_iallocator,
3802
                                   errors.ECODE_INVAL)
3803

    
3804
  def Exec(self, feedback_fn):
3805
    """Change the parameters of the cluster.
3806

3807
    """
3808
    if self.op.vg_name is not None:
3809
      new_volume = self.op.vg_name
3810
      if not new_volume:
3811
        new_volume = None
3812
      if new_volume != self.cfg.GetVGName():
3813
        self.cfg.SetVGName(new_volume)
3814
      else:
3815
        feedback_fn("Cluster LVM configuration already in desired"
3816
                    " state, not changing")
3817
    if self.op.drbd_helper is not None:
3818
      new_helper = self.op.drbd_helper
3819
      if not new_helper:
3820
        new_helper = None
3821
      if new_helper != self.cfg.GetDRBDHelper():
3822
        self.cfg.SetDRBDHelper(new_helper)
3823
      else:
3824
        feedback_fn("Cluster DRBD helper already in desired state,"
3825
                    " not changing")
3826
    if self.op.hvparams:
3827
      self.cluster.hvparams = self.new_hvparams
3828
    if self.op.os_hvp:
3829
      self.cluster.os_hvp = self.new_os_hvp
3830
    if self.op.enabled_hypervisors is not None:
3831
      self.cluster.hvparams = self.new_hvparams
3832
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3833
    if self.op.beparams:
3834
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3835
    if self.op.nicparams:
3836
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3837
    if self.op.osparams:
3838
      self.cluster.osparams = self.new_osp
3839
    if self.op.ndparams:
3840
      self.cluster.ndparams = self.new_ndparams
3841
    if self.op.diskparams:
3842
      self.cluster.diskparams = self.new_diskparams
3843
    if self.op.hv_state:
3844
      self.cluster.hv_state_static = self.new_hv_state
3845
    if self.op.disk_state:
3846
      self.cluster.disk_state_static = self.new_disk_state
3847

    
3848
    if self.op.candidate_pool_size is not None:
3849
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3850
      # we need to update the pool size here, otherwise the save will fail
3851
      _AdjustCandidatePool(self, [])
3852

    
3853
    if self.op.maintain_node_health is not None:
3854
      if self.op.maintain_node_health and not constants.ENABLE_CONFD:
3855
        feedback_fn("Note: CONFD was disabled at build time, node health"
3856
                    " maintenance is not useful (still enabling it)")
3857
      self.cluster.maintain_node_health = self.op.maintain_node_health
3858

    
3859
    if self.op.prealloc_wipe_disks is not None:
3860
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3861

    
3862
    if self.op.add_uids is not None:
3863
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3864

    
3865
    if self.op.remove_uids is not None:
3866
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3867

    
3868
    if self.op.uid_pool is not None:
3869
      self.cluster.uid_pool = self.op.uid_pool
3870

    
3871
    if self.op.default_iallocator is not None:
3872
      self.cluster.default_iallocator = self.op.default_iallocator
3873

    
3874
    if self.op.reserved_lvs is not None:
3875
      self.cluster.reserved_lvs = self.op.reserved_lvs
3876

    
3877
    if self.op.use_external_mip_script is not None:
3878
      self.cluster.use_external_mip_script = self.op.use_external_mip_script
3879

    
3880
    def helper_os(aname, mods, desc):
3881
      desc += " OS list"
3882
      lst = getattr(self.cluster, aname)
3883
      for key, val in mods:
3884
        if key == constants.DDM_ADD:
3885
          if val in lst:
3886
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3887
          else:
3888
            lst.append(val)
3889
        elif key == constants.DDM_REMOVE:
3890
          if val in lst:
3891
            lst.remove(val)
3892
          else:
3893
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3894
        else:
3895
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3896

    
3897
    if self.op.hidden_os:
3898
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3899

    
3900
    if self.op.blacklisted_os:
3901
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3902

    
3903
    if self.op.master_netdev:
3904
      master_params = self.cfg.GetMasterNetworkParameters()
3905
      ems = self.cfg.GetUseExternalMipScript()
3906
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3907
                  self.cluster.master_netdev)
3908
      result = self.rpc.call_node_deactivate_master_ip(master_params.name,
3909
                                                       master_params, ems)
3910
      result.Raise("Could not disable the master ip")
3911
      feedback_fn("Changing master_netdev from %s to %s" %
3912
                  (master_params.netdev, self.op.master_netdev))
3913
      self.cluster.master_netdev = self.op.master_netdev
3914

    
3915
    if self.op.master_netmask:
3916
      master_params = self.cfg.GetMasterNetworkParameters()
3917
      feedback_fn("Changing master IP netmask to %s" % self.op.master_netmask)
3918
      result = self.rpc.call_node_change_master_netmask(master_params.name,
3919
                                                        master_params.netmask,
3920
                                                        self.op.master_netmask,
3921
                                                        master_params.ip,
3922
                                                        master_params.netdev)
3923
      if result.fail_msg:
3924
        msg = "Could not change the master IP netmask: %s" % result.fail_msg
3925
        feedback_fn(msg)
3926

    
3927
      self.cluster.master_netmask = self.op.master_netmask
3928

    
3929
    self.cfg.Update(self.cluster, feedback_fn)
3930

    
3931
    if self.op.master_netdev:
3932
      master_params = self.cfg.GetMasterNetworkParameters()
3933
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3934
                  self.op.master_netdev)
3935
      ems = self.cfg.GetUseExternalMipScript()
3936
      result = self.rpc.call_node_activate_master_ip(master_params.name,
3937
                                                     master_params, ems)
3938
      if result.fail_msg:
3939
        self.LogWarning("Could not re-enable the master ip on"
3940
                        " the master, please restart manually: %s",
3941
                        result.fail_msg)
3942

    
3943

    
3944
def _UploadHelper(lu, nodes, fname):
3945
  """Helper for uploading a file and showing warnings.
3946

3947
  """
3948
  if os.path.exists(fname):
3949
    result = lu.rpc.call_upload_file(nodes, fname)
3950
    for to_node, to_result in result.items():
3951
      msg = to_result.fail_msg
3952
      if msg:
3953
        msg = ("Copy of file %s to node %s failed: %s" %
3954
               (fname, to_node, msg))
3955
        lu.proc.LogWarning(msg)
3956

    
3957

    
3958
def _ComputeAncillaryFiles(cluster, redist):
3959
  """Compute files external to Ganeti which need to be consistent.
3960

3961
  @type redist: boolean
3962
  @param redist: Whether to include files which need to be redistributed
3963

3964
  """
3965
  # Compute files for all nodes
3966
  files_all = set([
3967
    constants.SSH_KNOWN_HOSTS_FILE,
3968
    constants.CONFD_HMAC_KEY,
3969
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3970
    constants.SPICE_CERT_FILE,
3971
    constants.SPICE_CACERT_FILE,
3972
    constants.RAPI_USERS_FILE,
3973
    ])
3974

    
3975
  if not redist:
3976
    files_all.update(constants.ALL_CERT_FILES)
3977
    files_all.update(ssconf.SimpleStore().GetFileList())
3978
  else:
3979
    # we need to ship at least the RAPI certificate
3980
    files_all.add(constants.RAPI_CERT_FILE)
3981

    
3982
  if cluster.modify_etc_hosts:
3983
    files_all.add(constants.ETC_HOSTS)
3984

    
3985
  # Files which are optional, these must:
3986
  # - be present in one other category as well
3987
  # - either exist or not exist on all nodes of that category (mc, vm all)
3988
  files_opt = set([
3989
    constants.RAPI_USERS_FILE,
3990
    ])
3991

    
3992
  # Files which should only be on master candidates
3993
  files_mc = set()
3994

    
3995
  if not redist:
3996
    files_mc.add(constants.CLUSTER_CONF_FILE)
3997

    
3998
    # FIXME: this should also be replicated but Ganeti doesn't support files_mc
3999
    # replication
4000
    files_mc.add(constants.DEFAULT_MASTER_SETUP_SCRIPT)
4001

    
4002
  # Files which should only be on VM-capable nodes
4003
  files_vm = set(filename
4004
    for hv_name in cluster.enabled_hypervisors
4005
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
4006

    
4007
  files_opt |= set(filename
4008
    for hv_name in cluster.enabled_hypervisors
4009
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
4010

    
4011
  # Filenames in each category must be unique
4012
  all_files_set = files_all | files_mc | files_vm
4013
  assert (len(all_files_set) ==
4014
          sum(map(len, [files_all, files_mc, files_vm]))), \
4015
         "Found file listed in more than one file list"
4016

    
4017
  # Optional files must be present in one other category
4018
  assert all_files_set.issuperset(files_opt), \
4019
         "Optional file not in a different required list"
4020

    
4021
  return (files_all, files_opt, files_mc, files_vm)
4022

    
4023

    
4024
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
4025
  """Distribute additional files which are part of the cluster configuration.
4026

4027
  ConfigWriter takes care of distributing the config and ssconf files, but
4028
  there are more files which should be distributed to all nodes. This function
4029
  makes sure those are copied.
4030

4031
  @param lu: calling logical unit
4032
  @param additional_nodes: list of nodes not in the config to distribute to
4033
  @type additional_vm: boolean
4034
  @param additional_vm: whether the additional nodes are vm-capable or not
4035

4036
  """
4037
  # Gather target nodes
4038
  cluster = lu.cfg.GetClusterInfo()
4039
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
4040

    
4041
  online_nodes = lu.cfg.GetOnlineNodeList()
4042
  vm_nodes = lu.cfg.GetVmCapableNodeList()
4043

    
4044
  if additional_nodes is not None:
4045
    online_nodes.extend(additional_nodes)
4046
    if additional_vm:
4047
      vm_nodes.extend(additional_nodes)
4048

    
4049
  # Never distribute to master node
4050
  for nodelist in [online_nodes, vm_nodes]:
4051
    if master_info.name in nodelist:
4052
      nodelist.remove(master_info.name)
4053

    
4054
  # Gather file lists
4055
  (files_all, _, files_mc, files_vm) = \
4056
    _ComputeAncillaryFiles(cluster, True)
4057

    
4058
  # Never re-distribute configuration file from here
4059
  assert not (constants.CLUSTER_CONF_FILE in files_all or
4060
              constants.CLUSTER_CONF_FILE in files_vm)
4061
  assert not files_mc, "Master candidates not handled in this function"
4062

    
4063
  filemap = [
4064
    (online_nodes, files_all),
4065
    (vm_nodes, files_vm),
4066
    ]
4067

    
4068
  # Upload the files
4069
  for (node_list, files) in filemap:
4070
    for fname in files:
4071
      _UploadHelper(lu, node_list, fname)
4072

    
4073

    
4074
class LUClusterRedistConf(NoHooksLU):
4075
  """Force the redistribution of cluster configuration.
4076

4077
  This is a very simple LU.
4078

4079
  """
4080
  REQ_BGL = False
4081

    
4082
  def ExpandNames(self):
4083
    self.needed_locks = {
4084
      locking.LEVEL_NODE: locking.ALL_SET,
4085
    }
4086
    self.share_locks[locking.LEVEL_NODE] = 1
4087

    
4088
  def Exec(self, feedback_fn):
4089
    """Redistribute the configuration.
4090

4091
    """
4092
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
4093
    _RedistributeAncillaryFiles(self)
4094

    
4095

    
4096
class LUClusterActivateMasterIp(NoHooksLU):
4097
  """Activate the master IP on the master node.
4098

4099
  """
4100
  def Exec(self, feedback_fn):
4101
    """Activate the master IP.
4102

4103
    """
4104
    master_params = self.cfg.GetMasterNetworkParameters()
4105
    ems = self.cfg.GetUseExternalMipScript()
4106
    result = self.rpc.call_node_activate_master_ip(master_params.name,
4107
                                                   master_params, ems)
4108
    result.Raise("Could not activate the master IP")
4109

    
4110

    
4111
class LUClusterDeactivateMasterIp(NoHooksLU):
4112
  """Deactivate the master IP on the master node.
4113

4114
  """
4115
  def Exec(self, feedback_fn):
4116
    """Deactivate the master IP.
4117

4118
    """
4119
    master_params = self.cfg.GetMasterNetworkParameters()
4120
    ems = self.cfg.GetUseExternalMipScript()
4121
    result = self.rpc.call_node_deactivate_master_ip(master_params.name,
4122
                                                     master_params, ems)
4123
    result.Raise("Could not deactivate the master IP")
4124

    
4125

    
4126
def _WaitForSync(lu, instance, disks=None, oneshot=False):
4127
  """Sleep and poll for an instance's disk to sync.
4128

4129
  """
4130
  if not instance.disks or disks is not None and not disks:
4131
    return True
4132

    
4133
  disks = _ExpandCheckDisks(instance, disks)
4134

    
4135
  if not oneshot:
4136
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
4137

    
4138
  node = instance.primary_node
4139

    
4140
  for dev in disks:
4141
    lu.cfg.SetDiskID(dev, node)
4142

    
4143
  # TODO: Convert to utils.Retry
4144

    
4145
  retries = 0
4146
  degr_retries = 10 # in seconds, as we sleep 1 second each time
4147
  while True:
4148
    max_time = 0
4149
    done = True
4150
    cumul_degraded = False
4151
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
4152
    msg = rstats.fail_msg
4153
    if msg:
4154
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
4155
      retries += 1
4156
      if retries >= 10:
4157
        raise errors.RemoteError("Can't contact node %s for mirror data,"
4158
                                 " aborting." % node)
4159
      time.sleep(6)
4160
      continue
4161
    rstats = rstats.payload
4162
    retries = 0
4163
    for i, mstat in enumerate(rstats):
4164
      if mstat is None:
4165
        lu.LogWarning("Can't compute data for node %s/%s",
4166
                           node, disks[i].iv_name)
4167
        continue
4168

    
4169
      cumul_degraded = (cumul_degraded or
4170
                        (mstat.is_degraded and mstat.sync_percent is None))
4171
      if mstat.sync_percent is not None:
4172
        done = False
4173
        if mstat.estimated_time is not None:
4174
          rem_time = ("%s remaining (estimated)" %
4175
                      utils.FormatSeconds(mstat.estimated_time))
4176
          max_time = mstat.estimated_time
4177
        else:
4178
          rem_time = "no time estimate"
4179
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
4180
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
4181

    
4182
    # if we're done but degraded, let's do a few small retries, to
4183
    # make sure we see a stable and not transient situation; therefore
4184
    # we force restart of the loop
4185
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
4186
      logging.info("Degraded disks found, %d retries left", degr_retries)
4187
      degr_retries -= 1
4188
      time.sleep(1)
4189
      continue
4190

    
4191
    if done or oneshot:
4192
      break
4193

    
4194
    time.sleep(min(60, max_time))
4195

    
4196
  if done:
4197
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
4198
  return not cumul_degraded
4199

    
4200

    
4201
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
4202
  """Check that mirrors are not degraded.
4203

4204
  The ldisk parameter, if True, will change the test from the
4205
  is_degraded attribute (which represents overall non-ok status for
4206
  the device(s)) to the ldisk (representing the local storage status).
4207

4208
  """
4209
  lu.cfg.SetDiskID(dev, node)
4210

    
4211
  result = True
4212

    
4213
  if on_primary or dev.AssembleOnSecondary():
4214
    rstats = lu.rpc.call_blockdev_find(node, dev)
4215
    msg = rstats.fail_msg
4216
    if msg:
4217
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
4218
      result = False
4219
    elif not rstats.payload:
4220
      lu.LogWarning("Can't find disk on node %s", node)
4221
      result = False
4222
    else:
4223
      if ldisk:
4224
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
4225
      else:
4226
        result = result and not rstats.payload.is_degraded
4227

    
4228
  if dev.children:
4229
    for child in dev.children:
4230
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
4231

    
4232
  return result
4233

    
4234

    
4235
class LUOobCommand(NoHooksLU):
4236
  """Logical unit for OOB handling.
4237

4238
  """
4239
  REG_BGL = False
4240
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
4241

    
4242
  def ExpandNames(self):
4243
    """Gather locks we need.
4244

4245
    """
4246
    if self.op.node_names:
4247
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
4248
      lock_names = self.op.node_names
4249
    else:
4250
      lock_names = locking.ALL_SET
4251

    
4252
    self.needed_locks = {
4253
      locking.LEVEL_NODE: lock_names,
4254
      }
4255

    
4256
  def CheckPrereq(self):
4257
    """Check prerequisites.
4258

4259
    This checks:
4260
     - the node exists in the configuration
4261
     - OOB is supported
4262

4263
    Any errors are signaled by raising errors.OpPrereqError.
4264

4265
    """
4266
    self.nodes = []
4267
    self.master_node = self.cfg.GetMasterNode()
4268

    
4269
    assert self.op.power_delay >= 0.0
4270

    
4271
    if self.op.node_names:
4272
      if (self.op.command in self._SKIP_MASTER and
4273
          self.master_node in self.op.node_names):
4274
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
4275
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
4276

    
4277
        if master_oob_handler:
4278
          additional_text = ("run '%s %s %s' if you want to operate on the"
4279
                             " master regardless") % (master_oob_handler,
4280
                                                      self.op.command,
4281
                                                      self.master_node)
4282
        else:
4283
          additional_text = "it does not support out-of-band operations"
4284

    
4285
        raise errors.OpPrereqError(("Operating on the master node %s is not"
4286
                                    " allowed for %s; %s") %
4287
                                   (self.master_node, self.op.command,
4288
                                    additional_text), errors.ECODE_INVAL)
4289
    else:
4290
      self.op.node_names = self.cfg.GetNodeList()
4291
      if self.op.command in self._SKIP_MASTER:
4292
        self.op.node_names.remove(self.master_node)
4293

    
4294
    if self.op.command in self._SKIP_MASTER:
4295
      assert self.master_node not in self.op.node_names
4296

    
4297
    for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
4298
      if node is None:
4299
        raise errors.OpPrereqError("Node %s not found" % node_name,
4300
                                   errors.ECODE_NOENT)
4301
      else:
4302
        self.nodes.append(node)
4303

    
4304
      if (not self.op.ignore_status and
4305
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
4306
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
4307
                                    " not marked offline") % node_name,
4308
                                   errors.ECODE_STATE)
4309

    
4310
  def Exec(self, feedback_fn):
4311
    """Execute OOB and return result if we expect any.
4312

4313
    """
4314
    master_node = self.master_node
4315
    ret = []
4316

    
4317
    for idx, node in enumerate(utils.NiceSort(self.nodes,
4318
                                              key=lambda node: node.name)):
4319
      node_entry = [(constants.RS_NORMAL, node.name)]
4320
      ret.append(node_entry)
4321

    
4322
      oob_program = _SupportsOob(self.cfg, node)
4323

    
4324
      if not oob_program:
4325
        node_entry.append((constants.RS_UNAVAIL, None))
4326
        continue
4327

    
4328
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
4329
                   self.op.command, oob_program, node.name)
4330
      result = self.rpc.call_run_oob(master_node, oob_program,
4331
                                     self.op.command, node.name,
4332
                                     self.op.timeout)
4333

    
4334
      if result.fail_msg:
4335
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4336
                        node.name, result.fail_msg)
4337
        node_entry.append((constants.RS_NODATA, None))
4338
      else:
4339
        try:
4340
          self._CheckPayload(result)
4341
        except errors.OpExecError, err:
4342
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
4343
                          node.name, err)
4344
          node_entry.append((constants.RS_NODATA, None))
4345
        else:
4346
          if self.op.command == constants.OOB_HEALTH:
4347
            # For health we should log important events
4348
            for item, status in result.payload:
4349
              if status in [constants.OOB_STATUS_WARNING,
4350
                            constants.OOB_STATUS_CRITICAL]:
4351
                self.LogWarning("Item '%s' on node '%s' has status '%s'",
4352
                                item, node.name, status)
4353

    
4354
          if self.op.command == constants.OOB_POWER_ON:
4355
            node.powered = True
4356
          elif self.op.command == constants.OOB_POWER_OFF:
4357
            node.powered = False
4358
          elif self.op.command == constants.OOB_POWER_STATUS:
4359
            powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
4360
            if powered != node.powered:
4361
              logging.warning(("Recorded power state (%s) of node '%s' does not"
4362
                               " match actual power state (%s)"), node.powered,
4363
                              node.name, powered)
4364

    
4365
          # For configuration changing commands we should update the node
4366
          if self.op.command in (constants.OOB_POWER_ON,
4367
                                 constants.OOB_POWER_OFF):
4368
            self.cfg.Update(node, feedback_fn)
4369

    
4370
          node_entry.append((constants.RS_NORMAL, result.payload))
4371

    
4372
          if (self.op.command == constants.OOB_POWER_ON and
4373
              idx < len(self.nodes) - 1):
4374
            time.sleep(self.op.power_delay)
4375

    
4376
    return ret
4377

    
4378
  def _CheckPayload(self, result):
4379
    """Checks if the payload is valid.
4380

4381
    @param result: RPC result
4382
    @raises errors.OpExecError: If payload is not valid
4383

4384
    """
4385
    errs = []
4386
    if self.op.command == constants.OOB_HEALTH:
4387
      if not isinstance(result.payload, list):
4388
        errs.append("command 'health' is expected to return a list but got %s" %
4389
                    type(result.payload))
4390
      else:
4391
        for item, status in result.payload:
4392
          if status not in constants.OOB_STATUSES:
4393
            errs.append("health item '%s' has invalid status '%s'" %
4394
                        (item, status))
4395

    
4396
    if self.op.command == constants.OOB_POWER_STATUS:
4397
      if not isinstance(result.payload, dict):
4398
        errs.append("power-status is expected to return a dict but got %s" %
4399
                    type(result.payload))
4400

    
4401
    if self.op.command in [
4402
        constants.OOB_POWER_ON,
4403
        constants.OOB_POWER_OFF,
4404
        constants.OOB_POWER_CYCLE,
4405
        ]:
4406
      if result.payload is not None:
4407
        errs.append("%s is expected to not return payload but got '%s'" %
4408
                    (self.op.command, result.payload))
4409

    
4410
    if errs:
4411
      raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
4412
                               utils.CommaJoin(errs))
4413

    
4414

    
4415
class _OsQuery(_QueryBase):
4416
  FIELDS = query.OS_FIELDS
4417

    
4418
  def ExpandNames(self, lu):
4419
    # Lock all nodes in shared mode
4420
    # Temporary removal of locks, should be reverted later
4421
    # TODO: reintroduce locks when they are lighter-weight
4422
    lu.needed_locks = {}
4423
    #self.share_locks[locking.LEVEL_NODE] = 1
4424
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4425

    
4426
    # The following variables interact with _QueryBase._GetNames
4427
    if self.names:
4428
      self.wanted = self.names
4429
    else:
4430
      self.wanted = locking.ALL_SET
4431

    
4432
    self.do_locking = self.use_locking
4433

    
4434
  def DeclareLocks(self, lu, level):
4435
    pass
4436

    
4437
  @staticmethod
4438
  def _DiagnoseByOS(rlist):
4439
    """Remaps a per-node return list into an a per-os per-node dictionary
4440

4441
    @param rlist: a map with node names as keys and OS objects as values
4442

4443
    @rtype: dict
4444
    @return: a dictionary with osnames as keys and as value another
4445
        map, with nodes as keys and tuples of (path, status, diagnose,
4446
        variants, parameters, api_versions) as values, eg::
4447

4448
          {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
4449
                                     (/srv/..., False, "invalid api")],
4450
                           "node2": [(/srv/..., True, "", [], [])]}
4451
          }
4452

4453
    """
4454
    all_os = {}
4455
    # we build here the list of nodes that didn't fail the RPC (at RPC
4456
    # level), so that nodes with a non-responding node daemon don't
4457
    # make all OSes invalid
4458
    good_nodes = [node_name for node_name in rlist
4459
                  if not rlist[node_name].fail_msg]
4460
    for node_name, nr in rlist.items():
4461
      if nr.fail_msg or not nr.payload:
4462
        continue
4463
      for (name, path, status, diagnose, variants,
4464
           params, api_versions) in nr.payload:
4465
        if name not in all_os:
4466
          # build a list of nodes for this os containing empty lists
4467
          # for each node in node_list
4468
          all_os[name] = {}
4469
          for nname in good_nodes:
4470
            all_os[name][nname] = []
4471
        # convert params from [name, help] to (name, help)
4472
        params = [tuple(v) for v in params]
4473
        all_os[name][node_name].append((path, status, diagnose,
4474
                                        variants, params, api_versions))
4475
    return all_os
4476

    
4477
  def _GetQueryData(self, lu):
4478
    """Computes the list of nodes and their attributes.
4479

4480
    """
4481
    # Locking is not used
4482
    assert not (compat.any(lu.glm.is_owned(level)
4483
                           for level in locking.LEVELS
4484
                           if level != locking.LEVEL_CLUSTER) or
4485
                self.do_locking or self.use_locking)
4486

    
4487
    valid_nodes = [node.name
4488
                   for node in lu.cfg.GetAllNodesInfo().values()
4489
                   if not node.offline and node.vm_capable]
4490
    pol = self._DiagnoseByOS(lu.rpc.call_os_diagnose(valid_nodes))
4491
    cluster = lu.cfg.GetClusterInfo()
4492

    
4493
    data = {}
4494

    
4495
    for (os_name, os_data) in pol.items():
4496
      info = query.OsInfo(name=os_name, valid=True, node_status=os_data,
4497
                          hidden=(os_name in cluster.hidden_os),
4498
                          blacklisted=(os_name in cluster.blacklisted_os))
4499

    
4500
      variants = set()
4501
      parameters = set()
4502
      api_versions = set()
4503

    
4504
      for idx, osl in enumerate(os_data.values()):
4505
        info.valid = bool(info.valid and osl and osl[0][1])
4506
        if not info.valid:
4507
          break
4508

    
4509
        (node_variants, node_params, node_api) = osl[0][3:6]
4510
        if idx == 0:
4511
          # First entry
4512
          variants.update(node_variants)
4513
          parameters.update(node_params)
4514
          api_versions.update(node_api)
4515
        else:
4516
          # Filter out inconsistent values
4517
          variants.intersection_update(node_variants)
4518
          parameters.intersection_update(node_params)
4519
          api_versions.intersection_update(node_api)
4520

    
4521
      info.variants = list(variants)
4522
      info.parameters = list(parameters)
4523
      info.api_versions = list(api_versions)
4524

    
4525
      data[os_name] = info
4526

    
4527
    # Prepare data in requested order
4528
    return [data[name] for name in self._GetNames(lu, pol.keys(), None)
4529
            if name in data]
4530

    
4531

    
4532
class LUOsDiagnose(NoHooksLU):
4533
  """Logical unit for OS diagnose/query.
4534

4535
  """
4536
  REQ_BGL = False
4537

    
4538
  @staticmethod
4539
  def _BuildFilter(fields, names):
4540
    """Builds a filter for querying OSes.
4541

4542
    """
4543
    name_filter = qlang.MakeSimpleFilter("name", names)
4544

    
4545
    # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the
4546
    # respective field is not requested
4547
    status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]
4548
                     for fname in ["hidden", "blacklisted"]
4549
                     if fname not in fields]
4550
    if "valid" not in fields:
4551
      status_filter.append([qlang.OP_TRUE, "valid"])
4552

    
4553
    if status_filter:
4554
      status_filter.insert(0, qlang.OP_AND)
4555
    else:
4556
      status_filter = None
4557

    
4558
    if name_filter and status_filter:
4559
      return [qlang.OP_AND, name_filter, status_filter]
4560
    elif name_filter:
4561
      return name_filter
4562
    else:
4563
      return status_filter
4564

    
4565
  def CheckArguments(self):
4566
    self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),
4567
                       self.op.output_fields, False)
4568

    
4569
  def ExpandNames(self):
4570
    self.oq.ExpandNames(self)
4571

    
4572
  def Exec(self, feedback_fn):
4573
    return self.oq.OldStyleQuery(self)
4574

    
4575

    
4576
class LUNodeRemove(LogicalUnit):
4577
  """Logical unit for removing a node.
4578

4579
  """
4580
  HPATH = "node-remove"
4581
  HTYPE = constants.HTYPE_NODE
4582

    
4583
  def BuildHooksEnv(self):
4584
    """Build hooks env.
4585

4586
    This doesn't run on the target node in the pre phase as a failed
4587
    node would then be impossible to remove.
4588

4589
    """
4590
    return {
4591
      "OP_TARGET": self.op.node_name,
4592
      "NODE_NAME": self.op.node_name,
4593
      }
4594

    
4595
  def BuildHooksNodes(self):
4596
    """Build hooks nodes.
4597

4598
    """
4599
    all_nodes = self.cfg.GetNodeList()
4600
    try:
4601
      all_nodes.remove(self.op.node_name)
4602
    except ValueError:
4603
      logging.warning("Node '%s', which is about to be removed, was not found"
4604
                      " in the list of all nodes", self.op.node_name)
4605
    return (all_nodes, all_nodes)
4606

    
4607
  def CheckPrereq(self):
4608
    """Check prerequisites.
4609

4610
    This checks:
4611
     - the node exists in the configuration
4612
     - it does not have primary or secondary instances
4613
     - it's not the master
4614

4615
    Any errors are signaled by raising errors.OpPrereqError.
4616

4617
    """
4618
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4619
    node = self.cfg.GetNodeInfo(self.op.node_name)
4620
    assert node is not None
4621

    
4622
    masternode = self.cfg.GetMasterNode()
4623
    if node.name == masternode:
4624
      raise errors.OpPrereqError("Node is the master node, failover to another"
4625
                                 " node is required", errors.ECODE_INVAL)
4626

    
4627
    for instance_name, instance in self.cfg.GetAllInstancesInfo().items():
4628
      if node.name in instance.all_nodes:
4629
        raise errors.OpPrereqError("Instance %s is still running on the node,"
4630
                                   " please remove first" % instance_name,
4631
                                   errors.ECODE_INVAL)
4632
    self.op.node_name = node.name
4633
    self.node = node
4634

    
4635
  def Exec(self, feedback_fn):
4636
    """Removes the node from the cluster.
4637

4638
    """
4639
    node = self.node
4640
    logging.info("Stopping the node daemon and removing configs from node %s",
4641
                 node.name)
4642

    
4643
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
4644

    
4645
    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
4646
      "Not owning BGL"
4647

    
4648
    # Promote nodes to master candidate as needed
4649
    _AdjustCandidatePool(self, exceptions=[node.name])
4650
    self.context.RemoveNode(node.name)
4651

    
4652
    # Run post hooks on the node before it's removed
4653
    _RunPostHook(self, node.name)
4654

    
4655
    result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
4656
    msg = result.fail_msg
4657
    if msg:
4658
      self.LogWarning("Errors encountered on the remote node while leaving"
4659
                      " the cluster: %s", msg)
4660

    
4661
    # Remove node from our /etc/hosts
4662
    if self.cfg.GetClusterInfo().modify_etc_hosts:
4663
      master_node = self.cfg.GetMasterNode()
4664
      result = self.rpc.call_etc_hosts_modify(master_node,
4665
                                              constants.ETC_HOSTS_REMOVE,
4666
                                              node.name, None)
4667
      result.Raise("Can't update hosts file with new host data")
4668
      _RedistributeAncillaryFiles(self)
4669

    
4670

    
4671
class _NodeQuery(_QueryBase):
4672
  FIELDS = query.NODE_FIELDS
4673

    
4674
  def ExpandNames(self, lu):
4675
    lu.needed_locks = {}
4676
    lu.share_locks = _ShareAll()
4677

    
4678
    if self.names:
4679
      self.wanted = _GetWantedNodes(lu, self.names)
4680
    else:
4681
      self.wanted = locking.ALL_SET
4682

    
4683
    self.do_locking = (self.use_locking and
4684
                       query.NQ_LIVE in self.requested_data)
4685

    
4686
    if self.do_locking:
4687
      # If any non-static field is requested we need to lock the nodes
4688
      lu.needed_locks[locking.LEVEL_NODE] = self.wanted
4689

    
4690
  def DeclareLocks(self, lu, level):
4691
    pass
4692

    
4693
  def _GetQueryData(self, lu):
4694
    """Computes the list of nodes and their attributes.
4695

4696
    """
4697
    all_info = lu.cfg.GetAllNodesInfo()
4698

    
4699
    nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
4700

    
4701
    # Gather data as requested
4702
    if query.NQ_LIVE in self.requested_data:
4703
      # filter out non-vm_capable nodes
4704
      toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
4705

    
4706
      node_data = lu.rpc.call_node_info(toquery_nodes, [lu.cfg.GetVGName()],
4707
                                        [lu.cfg.GetHypervisorType()])
4708
      live_data = dict((name, _MakeLegacyNodeInfo(nresult.payload))
4709
                       for (name, nresult) in node_data.items()
4710
                       if not nresult.fail_msg and nresult.payload)
4711
    else:
4712
      live_data = None
4713

    
4714
    if query.NQ_INST in self.requested_data:
4715
      node_to_primary = dict([(name, set()) for name in nodenames])
4716
      node_to_secondary = dict([(name, set()) for name in nodenames])
4717

    
4718
      inst_data = lu.cfg.GetAllInstancesInfo()
4719

    
4720
      for inst in inst_data.values():
4721
        if inst.primary_node in node_to_primary:
4722
          node_to_primary[inst.primary_node].add(inst.name)
4723
        for secnode in inst.secondary_nodes:
4724
          if secnode in node_to_secondary:
4725
            node_to_secondary[secnode].add(inst.name)
4726
    else:
4727
      node_to_primary = None
4728
      node_to_secondary = None
4729

    
4730
    if query.NQ_OOB in self.requested_data:
4731
      oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
4732
                         for name, node in all_info.iteritems())
4733
    else:
4734
      oob_support = None
4735

    
4736
    if query.NQ_GROUP in self.requested_data:
4737
      groups = lu.cfg.GetAllNodeGroupsInfo()
4738
    else:
4739
      groups = {}
4740

    
4741
    return query.NodeQueryData([all_info[name] for name in nodenames],
4742
                               live_data, lu.cfg.GetMasterNode(),
4743
                               node_to_primary, node_to_secondary, groups,
4744
                               oob_support, lu.cfg.GetClusterInfo())
4745

    
4746

    
4747
class LUNodeQuery(NoHooksLU):
4748
  """Logical unit for querying nodes.
4749

4750
  """
4751
  # pylint: disable=W0142
4752
  REQ_BGL = False
4753

    
4754
  def CheckArguments(self):
4755
    self.nq = _NodeQuery(qlang.MakeSimpleFilter("name", self.op.names),
4756
                         self.op.output_fields, self.op.use_locking)
4757

    
4758
  def ExpandNames(self):
4759
    self.nq.ExpandNames(self)
4760

    
4761
  def DeclareLocks(self, level):
4762
    self.nq.DeclareLocks(self, level)
4763

    
4764
  def Exec(self, feedback_fn):
4765
    return self.nq.OldStyleQuery(self)
4766

    
4767

    
4768
class LUNodeQueryvols(NoHooksLU):
4769
  """Logical unit for getting volumes on node(s).
4770

4771
  """
4772
  REQ_BGL = False
4773
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
4774
  _FIELDS_STATIC = utils.FieldSet("node")
4775

    
4776
  def CheckArguments(self):
4777
    _CheckOutputFields(static=self._FIELDS_STATIC,
4778
                       dynamic=self._FIELDS_DYNAMIC,
4779
                       selected=self.op.output_fields)
4780

    
4781
  def ExpandNames(self):
4782
    self.share_locks = _ShareAll()
4783
    self.needed_locks = {}
4784

    
4785
    if not self.op.nodes:
4786
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4787
    else:
4788
      self.needed_locks[locking.LEVEL_NODE] = \
4789
        _GetWantedNodes(self, self.op.nodes)
4790

    
4791
  def Exec(self, feedback_fn):
4792
    """Computes the list of nodes and their attributes.
4793

4794
    """
4795
    nodenames = self.owned_locks(locking.LEVEL_NODE)
4796
    volumes = self.rpc.call_node_volumes(nodenames)
4797

    
4798
    ilist = self.cfg.GetAllInstancesInfo()
4799
    vol2inst = _MapInstanceDisksToNodes(ilist.values())
4800

    
4801
    output = []
4802
    for node in nodenames:
4803
      nresult = volumes[node]
4804
      if nresult.offline:
4805
        continue
4806
      msg = nresult.fail_msg
4807
      if msg:
4808
        self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
4809
        continue
4810

    
4811
      node_vols = sorted(nresult.payload,
4812
                         key=operator.itemgetter("dev"))
4813

    
4814
      for vol in node_vols:
4815
        node_output = []
4816
        for field in self.op.output_fields:
4817
          if field == "node":
4818
            val = node
4819
          elif field == "phys":
4820
            val = vol["dev"]
4821
          elif field == "vg":
4822
            val = vol["vg"]
4823
          elif field == "name":
4824
            val = vol["name"]
4825
          elif field == "size":
4826
            val = int(float(vol["size"]))
4827
          elif field == "instance":
4828
            val = vol2inst.get((node, vol["vg"] + "/" + vol["name"]), "-")
4829
          else:
4830
            raise errors.ParameterError(field)
4831
          node_output.append(str(val))
4832

    
4833
        output.append(node_output)
4834

    
4835
    return output
4836

    
4837

    
4838
class LUNodeQueryStorage(NoHooksLU):
4839
  """Logical unit for getting information on storage units on node(s).
4840

4841
  """
4842
  _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
4843
  REQ_BGL = False
4844

    
4845
  def CheckArguments(self):
4846
    _CheckOutputFields(static=self._FIELDS_STATIC,
4847
                       dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
4848
                       selected=self.op.output_fields)
4849

    
4850
  def ExpandNames(self):
4851
    self.share_locks = _ShareAll()
4852
    self.needed_locks = {}
4853

    
4854
    if self.op.nodes:
4855
      self.needed_locks[locking.LEVEL_NODE] = \
4856
        _GetWantedNodes(self, self.op.nodes)
4857
    else:
4858
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4859

    
4860
  def Exec(self, feedback_fn):
4861
    """Computes the list of nodes and their attributes.
4862

4863
    """
4864
    self.nodes = self.owned_locks(locking.LEVEL_NODE)
4865

    
4866
    # Always get name to sort by
4867
    if constants.SF_NAME in self.op.output_fields:
4868
      fields = self.op.output_fields[:]
4869
    else:
4870
      fields = [constants.SF_NAME] + self.op.output_fields
4871

    
4872
    # Never ask for node or type as it's only known to the LU
4873
    for extra in [constants.SF_NODE, constants.SF_TYPE]:
4874
      while extra in fields:
4875
        fields.remove(extra)
4876

    
4877
    field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
4878
    name_idx = field_idx[constants.SF_NAME]
4879

    
4880
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4881
    data = self.rpc.call_storage_list(self.nodes,
4882
                                      self.op.storage_type, st_args,
4883
                                      self.op.name, fields)
4884

    
4885
    result = []
4886

    
4887
    for node in utils.NiceSort(self.nodes):
4888
      nresult = data[node]
4889
      if nresult.offline:
4890
        continue
4891

    
4892
      msg = nresult.fail_msg
4893
      if msg:
4894
        self.LogWarning("Can't get storage data from node %s: %s", node, msg)
4895
        continue
4896

    
4897
      rows = dict([(row[name_idx], row) for row in nresult.payload])
4898

    
4899
      for name in utils.NiceSort(rows.keys()):
4900
        row = rows[name]
4901

    
4902
        out = []
4903

    
4904
        for field in self.op.output_fields:
4905
          if field == constants.SF_NODE:
4906
            val = node
4907
          elif field == constants.SF_TYPE:
4908
            val = self.op.storage_type
4909
          elif field in field_idx:
4910
            val = row[field_idx[field]]
4911
          else:
4912
            raise errors.ParameterError(field)
4913

    
4914
          out.append(val)
4915

    
4916
        result.append(out)
4917

    
4918
    return result
4919

    
4920

    
4921
class _InstanceQuery(_QueryBase):
4922
  FIELDS = query.INSTANCE_FIELDS
4923

    
4924
  def ExpandNames(self, lu):
4925
    lu.needed_locks = {}
4926
    lu.share_locks = _ShareAll()
4927

    
4928
    if self.names:
4929
      self.wanted = _GetWantedInstances(lu, self.names)
4930
    else:
4931
      self.wanted = locking.ALL_SET
4932

    
4933
    self.do_locking = (self.use_locking and
4934
                       query.IQ_LIVE in self.requested_data)
4935
    if self.do_locking:
4936
      lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
4937
      lu.needed_locks[locking.LEVEL_NODEGROUP] = []
4938
      lu.needed_locks[locking.LEVEL_NODE] = []
4939
      lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4940

    
4941
    self.do_grouplocks = (self.do_locking and
4942
                          query.IQ_NODES in self.requested_data)
4943

    
4944
  def DeclareLocks(self, lu, level):
4945
    if self.do_locking:
4946
      if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:
4947
        assert not lu.needed_locks[locking.LEVEL_NODEGROUP]
4948

    
4949
        # Lock all groups used by instances optimistically; this requires going
4950
        # via the node before it's locked, requiring verification later on
4951
        lu.needed_locks[locking.LEVEL_NODEGROUP] = \
4952
          set(group_uuid
4953
              for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
4954
              for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
4955
      elif level == locking.LEVEL_NODE:
4956
        lu._LockInstancesNodes() # pylint: disable=W0212
4957

    
4958
  @staticmethod
4959
  def _CheckGroupLocks(lu):
4960
    owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))
4961
    owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))
4962

    
4963
    # Check if node groups for locked instances are still correct
4964
    for instance_name in owned_instances:
4965
      _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
4966

    
4967
  def _GetQueryData(self, lu):
4968
    """Computes the list of instances and their attributes.
4969

4970
    """
4971
    if self.do_grouplocks:
4972
      self._CheckGroupLocks(lu)
4973

    
4974
    cluster = lu.cfg.GetClusterInfo()
4975
    all_info = lu.cfg.GetAllInstancesInfo()
4976

    
4977
    instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
4978

    
4979
    instance_list = [all_info[name] for name in instance_names]
4980
    nodes = frozenset(itertools.chain(*(inst.all_nodes
4981
                                        for inst in instance_list)))
4982
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
4983
    bad_nodes = []
4984
    offline_nodes = []
4985
    wrongnode_inst = set()
4986

    
4987
    # Gather data as requested
4988
    if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
4989
      live_data = {}
4990
      node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
4991
      for name in nodes:
4992
        result = node_data[name]
4993
        if result.offline:
4994
          # offline nodes will be in both lists
4995
          assert result.fail_msg
4996
          offline_nodes.append(name)
4997
        if result.fail_msg:
4998
          bad_nodes.append(name)
4999
        elif result.payload:
5000
          for inst in result.payload:
5001
            if inst in all_info:
5002
              if all_info[inst].primary_node == name:
5003
                live_data.update(result.payload)
5004
              else:
5005
                wrongnode_inst.add(inst)
5006
            else:
5007
              # orphan instance; we don't list it here as we don't
5008
              # handle this case yet in the output of instance listing
5009
              logging.warning("Orphan instance '%s' found on node %s",
5010
                              inst, name)
5011
        # else no instance is alive
5012
    else:
5013
      live_data = {}
5014

    
5015
    if query.IQ_DISKUSAGE in self.requested_data:
5016
      disk_usage = dict((inst.name,
5017
                         _ComputeDiskSize(inst.disk_template,
5018
                                          [{constants.IDISK_SIZE: disk.size}
5019
                                           for disk in inst.disks]))
5020
                        for inst in instance_list)
5021
    else:
5022
      disk_usage = None
5023

    
5024
    if query.IQ_CONSOLE in self.requested_data:
5025
      consinfo = {}
5026
      for inst in instance_list:
5027
        if inst.name in live_data:
5028
          # Instance is running
5029
          consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
5030
        else:
5031
          consinfo[inst.name] = None
5032
      assert set(consinfo.keys()) == set(instance_names)
5033
    else:
5034
      consinfo = None
5035

    
5036
    if query.IQ_NODES in self.requested_data:
5037
      node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),
5038
                                            instance_list)))
5039
      nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))
5040
      groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))
5041
                    for uuid in set(map(operator.attrgetter("group"),
5042
                                        nodes.values())))
5043
    else:
5044
      nodes = None
5045
      groups = None
5046

    
5047
    return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
5048
                                   disk_usage, offline_nodes, bad_nodes,
5049
                                   live_data, wrongnode_inst, consinfo,
5050
                                   nodes, groups)
5051

    
5052

    
5053
class LUQuery(NoHooksLU):
5054
  """Query for resources/items of a certain kind.
5055

5056
  """
5057
  # pylint: disable=W0142
5058
  REQ_BGL = False
5059

    
5060
  def CheckArguments(self):
5061
    qcls = _GetQueryImplementation(self.op.what)
5062

    
5063
    self.impl = qcls(self.op.qfilter, self.op.fields, self.op.use_locking)
5064

    
5065
  def ExpandNames(self):
5066
    self.impl.ExpandNames(self)
5067

    
5068
  def DeclareLocks(self, level):
5069
    self.impl.DeclareLocks(self, level)
5070

    
5071
  def Exec(self, feedback_fn):
5072
    return self.impl.NewStyleQuery(self)
5073

    
5074

    
5075
class LUQueryFields(NoHooksLU):
5076
  """Query for resources/items of a certain kind.
5077

5078
  """
5079
  # pylint: disable=W0142
5080
  REQ_BGL = False
5081

    
5082
  def CheckArguments(self):
5083
    self.qcls = _GetQueryImplementation(self.op.what)
5084

    
5085
  def ExpandNames(self):
5086
    self.needed_locks = {}
5087

    
5088
  def Exec(self, feedback_fn):
5089
    return query.QueryFields(self.qcls.FIELDS, self.op.fields)
5090

    
5091

    
5092
class LUNodeModifyStorage(NoHooksLU):
5093
  """Logical unit for modifying a storage volume on a node.
5094

5095
  """
5096
  REQ_BGL = False
5097

    
5098
  def CheckArguments(self):
5099
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5100

    
5101
    storage_type = self.op.storage_type
5102

    
5103
    try:
5104
      modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
5105
    except KeyError:
5106
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
5107
                                 " modified" % storage_type,
5108
                                 errors.ECODE_INVAL)
5109

    
5110
    diff = set(self.op.changes.keys()) - modifiable
5111
    if diff:
5112
      raise errors.OpPrereqError("The following fields can not be modified for"
5113
                                 " storage units of type '%s': %r" %
5114
                                 (storage_type, list(diff)),
5115
                                 errors.ECODE_INVAL)
5116

    
5117
  def ExpandNames(self):
5118
    self.needed_locks = {
5119
      locking.LEVEL_NODE: self.op.node_name,
5120
      }
5121

    
5122
  def Exec(self, feedback_fn):
5123
    """Computes the list of nodes and their attributes.
5124

5125
    """
5126
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
5127
    result = self.rpc.call_storage_modify(self.op.node_name,
5128
                                          self.op.storage_type, st_args,
5129
                                          self.op.name, self.op.changes)
5130
    result.Raise("Failed to modify storage unit '%s' on %s" %
5131
                 (self.op.name, self.op.node_name))
5132

    
5133

    
5134
class LUNodeAdd(LogicalUnit):
5135
  """Logical unit for adding node to the cluster.
5136

5137
  """
5138
  HPATH = "node-add"
5139
  HTYPE = constants.HTYPE_NODE
5140
  _NFLAGS = ["master_capable", "vm_capable"]
5141

    
5142
  def CheckArguments(self):
5143
    self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
5144
    # validate/normalize the node name
5145
    self.hostname = netutils.GetHostname(name=self.op.node_name,
5146
                                         family=self.primary_ip_family)
5147
    self.op.node_name = self.hostname.name
5148

    
5149
    if self.op.readd and self.op.node_name == self.cfg.GetMasterNode():
5150
      raise errors.OpPrereqError("Cannot readd the master node",
5151
                                 errors.ECODE_STATE)
5152

    
5153
    if self.op.readd and self.op.group:
5154
      raise errors.OpPrereqError("Cannot pass a node group when a node is"
5155
                                 " being readded", errors.ECODE_INVAL)
5156

    
5157
  def BuildHooksEnv(self):
5158
    """Build hooks env.
5159

5160
    This will run on all nodes before, and on all nodes + the new node after.
5161

5162
    """
5163
    return {
5164
      "OP_TARGET": self.op.node_name,
5165
      "NODE_NAME": self.op.node_name,
5166
      "NODE_PIP": self.op.primary_ip,
5167
      "NODE_SIP": self.op.secondary_ip,
5168
      "MASTER_CAPABLE": str(self.op.master_capable),
5169
      "VM_CAPABLE": str(self.op.vm_capable),
5170
      }
5171

    
5172
  def BuildHooksNodes(self):
5173
    """Build hooks nodes.
5174

5175
    """
5176
    # Exclude added node
5177
    pre_nodes = list(set(self.cfg.GetNodeList()) - set([self.op.node_name]))
5178
    post_nodes = pre_nodes + [self.op.node_name, ]
5179

    
5180
    return (pre_nodes, post_nodes)
5181

    
5182
  def CheckPrereq(self):
5183
    """Check prerequisites.
5184

5185
    This checks:
5186
     - the new node is not already in the config
5187
     - it is resolvable
5188
     - its parameters (single/dual homed) matches the cluster
5189

5190
    Any errors are signaled by raising errors.OpPrereqError.
5191

5192
    """
5193
    cfg = self.cfg
5194
    hostname = self.hostname
5195
    node = hostname.name
5196
    primary_ip = self.op.primary_ip = hostname.ip
5197
    if self.op.secondary_ip is None:
5198
      if self.primary_ip_family == netutils.IP6Address.family:
5199
        raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
5200
                                   " IPv4 address must be given as secondary",
5201
                                   errors.ECODE_INVAL)
5202
      self.op.secondary_ip = primary_ip
5203

    
5204
    secondary_ip = self.op.secondary_ip
5205
    if not netutils.IP4Address.IsValid(secondary_ip):
5206
      raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5207
                                 " address" % secondary_ip, errors.ECODE_INVAL)
5208

    
5209
    node_list = cfg.GetNodeList()
5210
    if not self.op.readd and node in node_list:
5211
      raise errors.OpPrereqError("Node %s is already in the configuration" %
5212
                                 node, errors.ECODE_EXISTS)
5213
    elif self.op.readd and node not in node_list:
5214
      raise errors.OpPrereqError("Node %s is not in the configuration" % node,
5215
                                 errors.ECODE_NOENT)
5216

    
5217
    self.changed_primary_ip = False
5218

    
5219
    for existing_node_name, existing_node in cfg.GetMultiNodeInfo(node_list):
5220
      if self.op.readd and node == existing_node_name:
5221
        if existing_node.secondary_ip != secondary_ip:
5222
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
5223
                                     " address configuration as before",
5224
                                     errors.ECODE_INVAL)
5225
        if existing_node.primary_ip != primary_ip:
5226
          self.changed_primary_ip = True
5227

    
5228
        continue
5229

    
5230
      if (existing_node.primary_ip == primary_ip or
5231
          existing_node.secondary_ip == primary_ip or
5232
          existing_node.primary_ip == secondary_ip or
5233
          existing_node.secondary_ip == secondary_ip):
5234
        raise errors.OpPrereqError("New node ip address(es) conflict with"
5235
                                   " existing node %s" % existing_node.name,
5236
                                   errors.ECODE_NOTUNIQUE)
5237

    
5238
    # After this 'if' block, None is no longer a valid value for the
5239
    # _capable op attributes
5240
    if self.op.readd:
5241
      old_node = self.cfg.GetNodeInfo(node)
5242
      assert old_node is not None, "Can't retrieve locked node %s" % node
5243
      for attr in self._NFLAGS:
5244
        if getattr(self.op, attr) is None:
5245
          setattr(self.op, attr, getattr(old_node, attr))
5246
    else:
5247
      for attr in self._NFLAGS:
5248
        if getattr(self.op, attr) is None:
5249
          setattr(self.op, attr, True)
5250

    
5251
    if self.op.readd and not self.op.vm_capable:
5252
      pri, sec = cfg.GetNodeInstances(node)
5253
      if pri or sec:
5254
        raise errors.OpPrereqError("Node %s being re-added with vm_capable"
5255
                                   " flag set to false, but it already holds"
5256
                                   " instances" % node,
5257
                                   errors.ECODE_STATE)
5258

    
5259
    # check that the type of the node (single versus dual homed) is the
5260
    # same as for the master
5261
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
5262
    master_singlehomed = myself.secondary_ip == myself.primary_ip
5263
    newbie_singlehomed = secondary_ip == primary_ip
5264
    if master_singlehomed != newbie_singlehomed:
5265
      if master_singlehomed:
5266
        raise errors.OpPrereqError("The master has no secondary ip but the"
5267
                                   " new node has one",
5268
                                   errors.ECODE_INVAL)
5269
      else:
5270
        raise errors.OpPrereqError("The master has a secondary ip but the"
5271
                                   " new node doesn't have one",
5272
                                   errors.ECODE_INVAL)
5273

    
5274
    # checks reachability
5275
    if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
5276
      raise errors.OpPrereqError("Node not reachable by ping",
5277
                                 errors.ECODE_ENVIRON)
5278

    
5279
    if not newbie_singlehomed:
5280
      # check reachability from my secondary ip to newbie's secondary ip
5281
      if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
5282
                           source=myself.secondary_ip):
5283
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5284
                                   " based ping to node daemon port",
5285
                                   errors.ECODE_ENVIRON)
5286

    
5287
    if self.op.readd:
5288
      exceptions = [node]
5289
    else:
5290
      exceptions = []
5291

    
5292
    if self.op.master_capable:
5293
      self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
5294
    else:
5295
      self.master_candidate = False
5296

    
5297
    if self.op.readd:
5298
      self.new_node = old_node
5299
    else:
5300
      node_group = cfg.LookupNodeGroup(self.op.group)
5301
      self.new_node = objects.Node(name=node,
5302
                                   primary_ip=primary_ip,
5303
                                   secondary_ip=secondary_ip,
5304
                                   master_candidate=self.master_candidate,
5305
                                   offline=False, drained=False,
5306
                                   group=node_group)
5307

    
5308
    if self.op.ndparams:
5309
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
5310

    
5311
  def Exec(self, feedback_fn):
5312
    """Adds the new node to the cluster.
5313

5314
    """
5315
    new_node = self.new_node
5316
    node = new_node.name
5317

    
5318
    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
5319
      "Not owning BGL"
5320

    
5321
    # We adding a new node so we assume it's powered
5322
    new_node.powered = True
5323

    
5324
    # for re-adds, reset the offline/drained/master-candidate flags;
5325
    # we need to reset here, otherwise offline would prevent RPC calls
5326
    # later in the procedure; this also means that if the re-add
5327
    # fails, we are left with a non-offlined, broken node
5328
    if self.op.readd:
5329
      new_node.drained = new_node.offline = False # pylint: disable=W0201
5330
      self.LogInfo("Readding a node, the offline/drained flags were reset")
5331
      # if we demote the node, we do cleanup later in the procedure
5332
      new_node.master_candidate = self.master_candidate
5333
      if self.changed_primary_ip:
5334
        new_node.primary_ip = self.op.primary_ip
5335

    
5336
    # copy the master/vm_capable flags
5337
    for attr in self._NFLAGS:
5338
      setattr(new_node, attr, getattr(self.op, attr))
5339

    
5340
    # notify the user about any possible mc promotion
5341
    if new_node.master_candidate:
5342
      self.LogInfo("Node will be a master candidate")
5343

    
5344
    if self.op.ndparams:
5345
      new_node.ndparams = self.op.ndparams
5346
    else:
5347
      new_node.ndparams = {}
5348

    
5349
    # check connectivity
5350
    result = self.rpc.call_version([node])[node]
5351
    result.Raise("Can't get version information from node %s" % node)
5352
    if constants.PROTOCOL_VERSION == result.payload:
5353
      logging.info("Communication to node %s fine, sw version %s match",
5354
                   node, result.payload)
5355
    else:
5356
      raise errors.OpExecError("Version mismatch master version %s,"
5357
                               " node version %s" %
5358
                               (constants.PROTOCOL_VERSION, result.payload))
5359

    
5360
    # Add node to our /etc/hosts, and add key to known_hosts
5361
    if self.cfg.GetClusterInfo().modify_etc_hosts:
5362
      master_node = self.cfg.GetMasterNode()
5363
      result = self.rpc.call_etc_hosts_modify(master_node,
5364
                                              constants.ETC_HOSTS_ADD,
5365
                                              self.hostname.name,
5366
                                              self.hostname.ip)
5367
      result.Raise("Can't update hosts file with new host data")
5368

    
5369
    if new_node.secondary_ip != new_node.primary_ip:
5370
      _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
5371
                               False)
5372

    
5373
    node_verify_list = [self.cfg.GetMasterNode()]
5374
    node_verify_param = {
5375
      constants.NV_NODELIST: ([node], {}),
5376
      # TODO: do a node-net-test as well?
5377
    }
5378

    
5379
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
5380
                                       self.cfg.GetClusterName())
5381
    for verifier in node_verify_list:
5382
      result[verifier].Raise("Cannot communicate with node %s" % verifier)
5383
      nl_payload = result[verifier].payload[constants.NV_NODELIST]
5384
      if nl_payload:
5385
        for failed in nl_payload:
5386
          feedback_fn("ssh/hostname verification failed"
5387
                      " (checking from %s): %s" %
5388
                      (verifier, nl_payload[failed]))
5389
        raise errors.OpExecError("ssh/hostname verification failed")
5390

    
5391
    if self.op.readd:
5392
      _RedistributeAncillaryFiles(self)
5393
      self.context.ReaddNode(new_node)
5394
      # make sure we redistribute the config
5395
      self.cfg.Update(new_node, feedback_fn)
5396
      # and make sure the new node will not have old files around
5397
      if not new_node.master_candidate:
5398
        result = self.rpc.call_node_demote_from_mc(new_node.name)
5399
        msg = result.fail_msg
5400
        if msg:
5401
          self.LogWarning("Node failed to demote itself from master"
5402
                          " candidate status: %s" % msg)
5403
    else:
5404
      _RedistributeAncillaryFiles(self, additional_nodes=[node],
5405
                                  additional_vm=self.op.vm_capable)
5406
      self.context.AddNode(new_node, self.proc.GetECId())
5407

    
5408

    
5409
class LUNodeSetParams(LogicalUnit):
5410
  """Modifies the parameters of a node.
5411

5412
  @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
5413
      to the node role (as _ROLE_*)
5414
  @cvar _R2F: a dictionary from node role to tuples of flags
5415
  @cvar _FLAGS: a list of attribute names corresponding to the flags
5416

5417
  """
5418
  HPATH = "node-modify"
5419
  HTYPE = constants.HTYPE_NODE
5420
  REQ_BGL = False
5421
  (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
5422
  _F2R = {
5423
    (True, False, False): _ROLE_CANDIDATE,
5424
    (False, True, False): _ROLE_DRAINED,
5425
    (False, False, True): _ROLE_OFFLINE,
5426
    (False, False, False): _ROLE_REGULAR,
5427
    }
5428
  _R2F = dict((v, k) for k, v in _F2R.items())
5429
  _FLAGS = ["master_candidate", "drained", "offline"]
5430

    
5431
  def CheckArguments(self):
5432
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5433
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
5434
                self.op.master_capable, self.op.vm_capable,
5435
                self.op.secondary_ip, self.op.ndparams, self.op.hv_state,
5436
                self.op.disk_state]
5437
    if all_mods.count(None) == len(all_mods):
5438
      raise errors.OpPrereqError("Please pass at least one modification",
5439
                                 errors.ECODE_INVAL)
5440
    if all_mods.count(True) > 1:
5441
      raise errors.OpPrereqError("Can't set the node into more than one"
5442
                                 " state at the same time",
5443
                                 errors.ECODE_INVAL)
5444

    
5445
    # Boolean value that tells us whether we might be demoting from MC
5446
    self.might_demote = (self.op.master_candidate == False or
5447
                         self.op.offline == True or
5448
                         self.op.drained == True or
5449
                         self.op.master_capable == False)
5450

    
5451
    if self.op.secondary_ip:
5452
      if not netutils.IP4Address.IsValid(self.op.secondary_ip):
5453
        raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5454
                                   " address" % self.op.secondary_ip,
5455
                                   errors.ECODE_INVAL)
5456

    
5457
    self.lock_all = self.op.auto_promote and self.might_demote
5458
    self.lock_instances = self.op.secondary_ip is not None
5459

    
5460
  def _InstanceFilter(self, instance):
5461
    """Filter for getting affected instances.
5462

5463
    """
5464
    return (instance.disk_template in constants.DTS_INT_MIRROR and
5465
            self.op.node_name in instance.all_nodes)
5466

    
5467
  def ExpandNames(self):
5468
    if self.lock_all:
5469
      self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
5470
    else:
5471
      self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
5472

    
5473
    # Since modifying a node can have severe effects on currently running
5474
    # operations the resource lock is at least acquired in shared mode
5475
    self.needed_locks[locking.LEVEL_NODE_RES] = \
5476
      self.needed_locks[locking.LEVEL_NODE]
5477

    
5478
    # Get node resource and instance locks in shared mode; they are not used
5479
    # for anything but read-only access
5480
    self.share_locks[locking.LEVEL_NODE_RES] = 1
5481
    self.share_locks[locking.LEVEL_INSTANCE] = 1
5482

    
5483
    if self.lock_instances:
5484
      self.needed_locks[locking.LEVEL_INSTANCE] = \
5485
        frozenset(self.cfg.GetInstancesInfoByFilter(self._InstanceFilter))
5486

    
5487
  def BuildHooksEnv(self):
5488
    """Build hooks env.
5489

5490
    This runs on the master node.
5491

5492
    """
5493
    return {
5494
      "OP_TARGET": self.op.node_name,
5495
      "MASTER_CANDIDATE": str(self.op.master_candidate),
5496
      "OFFLINE": str(self.op.offline),
5497
      "DRAINED": str(self.op.drained),
5498
      "MASTER_CAPABLE": str(self.op.master_capable),
5499
      "VM_CAPABLE": str(self.op.vm_capable),
5500
      }
5501

    
5502
  def BuildHooksNodes(self):
5503
    """Build hooks nodes.
5504

5505
    """
5506
    nl = [self.cfg.GetMasterNode(), self.op.node_name]
5507
    return (nl, nl)
5508

    
5509
  def CheckPrereq(self):
5510
    """Check prerequisites.
5511

5512
    This only checks the instance list against the existing names.
5513

5514
    """
5515
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
5516

    
5517
    if self.lock_instances:
5518
      affected_instances = \
5519
        self.cfg.GetInstancesInfoByFilter(self._InstanceFilter)
5520

    
5521
      # Verify instance locks
5522
      owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
5523
      wanted_instances = frozenset(affected_instances.keys())
5524
      if wanted_instances - owned_instances:
5525
        raise errors.OpPrereqError("Instances affected by changing node %s's"
5526
                                   " secondary IP address have changed since"
5527
                                   " locks were acquired, wanted '%s', have"
5528
                                   " '%s'; retry the operation" %
5529
                                   (self.op.node_name,
5530
                                    utils.CommaJoin(wanted_instances),
5531
                                    utils.CommaJoin(owned_instances)),
5532
                                   errors.ECODE_STATE)
5533
    else:
5534
      affected_instances = None
5535

    
5536
    if (self.op.master_candidate is not None or
5537
        self.op.drained is not None or
5538
        self.op.offline is not None):
5539
      # we can't change the master's node flags
5540
      if self.op.node_name == self.cfg.GetMasterNode():
5541
        raise errors.OpPrereqError("The master role can be changed"
5542
                                   " only via master-failover",
5543
                                   errors.ECODE_INVAL)
5544

    
5545
    if self.op.master_candidate and not node.master_capable:
5546
      raise errors.OpPrereqError("Node %s is not master capable, cannot make"
5547
                                 " it a master candidate" % node.name,
5548
                                 errors.ECODE_STATE)
5549

    
5550
    if self.op.vm_capable == False:
5551
      (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
5552
      if ipri or isec:
5553
        raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
5554
                                   " the vm_capable flag" % node.name,
5555
                                   errors.ECODE_STATE)
5556

    
5557
    if node.master_candidate and self.might_demote and not self.lock_all:
5558
      assert not self.op.auto_promote, "auto_promote set but lock_all not"
5559
      # check if after removing the current node, we're missing master
5560
      # candidates
5561
      (mc_remaining, mc_should, _) = \
5562
          self.cfg.GetMasterCandidateStats(exceptions=[node.name])
5563
      if mc_remaining < mc_should:
5564
        raise errors.OpPrereqError("Not enough master candidates, please"
5565
                                   " pass auto promote option to allow"
5566
                                   " promotion", errors.ECODE_STATE)
5567

    
5568
    self.old_flags = old_flags = (node.master_candidate,
5569
                                  node.drained, node.offline)
5570
    assert old_flags in self._F2R, "Un-handled old flags %s" % str(old_flags)
5571
    self.old_role = old_role = self._F2R[old_flags]
5572

    
5573
    # Check for ineffective changes
5574
    for attr in self._FLAGS:
5575
      if (getattr(self.op, attr) == False and getattr(node, attr) == False):
5576
        self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
5577
        setattr(self.op, attr, None)
5578

    
5579
    # Past this point, any flag change to False means a transition
5580
    # away from the respective state, as only real changes are kept
5581

    
5582
    # TODO: We might query the real power state if it supports OOB
5583
    if _SupportsOob(self.cfg, node):
5584
      if self.op.offline is False and not (node.powered or
5585
                                           self.op.powered == True):
5586
        raise errors.OpPrereqError(("Node %s needs to be turned on before its"
5587
                                    " offline status can be reset") %
5588
                                   self.op.node_name)
5589
    elif self.op.powered is not None:
5590
      raise errors.OpPrereqError(("Unable to change powered state for node %s"
5591
                                  " as it does not support out-of-band"
5592
                                  " handling") % self.op.node_name)
5593

    
5594
    # If we're being deofflined/drained, we'll MC ourself if needed
5595
    if (self.op.drained == False or self.op.offline == False or
5596
        (self.op.master_capable and not node.master_capable)):
5597
      if _DecideSelfPromotion(self):
5598
        self.op.master_candidate = True
5599
        self.LogInfo("Auto-promoting node to master candidate")
5600

    
5601
    # If we're no longer master capable, we'll demote ourselves from MC
5602
    if self.op.master_capable == False and node.master_candidate:
5603
      self.LogInfo("Demoting from master candidate")
5604
      self.op.master_candidate = False
5605

    
5606
    # Compute new role
5607
    assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
5608
    if self.op.master_candidate:
5609
      new_role = self._ROLE_CANDIDATE
5610
    elif self.op.drained:
5611
      new_role = self._ROLE_DRAINED
5612
    elif self.op.offline:
5613
      new_role = self._ROLE_OFFLINE
5614
    elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
5615
      # False is still in new flags, which means we're un-setting (the
5616
      # only) True flag
5617
      new_role = self._ROLE_REGULAR
5618
    else: # no new flags, nothing, keep old role
5619
      new_role = old_role
5620

    
5621
    self.new_role = new_role
5622

    
5623
    if old_role == self._ROLE_OFFLINE and new_role != old_role:
5624
      # Trying to transition out of offline status
5625
      # TODO: Use standard RPC runner, but make sure it works when the node is
5626
      # still marked offline
5627
      result = rpc.BootstrapRunner().call_version([node.name])[node.name]
5628
      if result.fail_msg:
5629
        raise errors.OpPrereqError("Node %s is being de-offlined but fails"
5630
                                   " to report its version: %s" %
5631
                                   (node.name, result.fail_msg),
5632
                                   errors.ECODE_STATE)
5633
      else:
5634
        self.LogWarning("Transitioning node from offline to online state"
5635
                        " without using re-add. Please make sure the node"
5636
                        " is healthy!")
5637

    
5638
    if self.op.secondary_ip:
5639
      # Ok even without locking, because this can't be changed by any LU
5640
      master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
5641
      master_singlehomed = master.secondary_ip == master.primary_ip
5642
      if master_singlehomed and self.op.secondary_ip:
5643
        raise errors.OpPrereqError("Cannot change the secondary ip on a single"
5644
                                   " homed cluster", errors.ECODE_INVAL)
5645

    
5646
      assert not (frozenset(affected_instances) -
5647
                  self.owned_locks(locking.LEVEL_INSTANCE))
5648

    
5649
      if node.offline:
5650
        if affected_instances:
5651
          raise errors.OpPrereqError("Cannot change secondary IP address:"
5652
                                     " offline node has instances (%s)"
5653
                                     " configured to use it" %
5654
                                     utils.CommaJoin(affected_instances.keys()))
5655
      else:
5656
        # On online nodes, check that no instances are running, and that
5657
        # the node has the new ip and we can reach it.
5658
        for instance in affected_instances.values():
5659
          _CheckInstanceState(self, instance, INSTANCE_DOWN,
5660
                              msg="cannot change secondary ip")
5661

    
5662
        _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
5663
        if master.name != node.name:
5664
          # check reachability from master secondary ip to new secondary ip
5665
          if not netutils.TcpPing(self.op.secondary_ip,
5666
                                  constants.DEFAULT_NODED_PORT,
5667
                                  source=master.secondary_ip):
5668
            raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5669
                                       " based ping to node daemon port",
5670
                                       errors.ECODE_ENVIRON)
5671

    
5672
    if self.op.ndparams:
5673
      new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
5674
      utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
5675
      self.new_ndparams = new_ndparams
5676

    
5677
    if self.op.hv_state:
5678
      self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
5679
                                                 self.node.hv_state_static)
5680

    
5681
    if self.op.disk_state:
5682
      self.new_disk_state = \
5683
        _MergeAndVerifyDiskState(self.op.disk_state,
5684
                                 self.node.disk_state_static)
5685

    
5686
  def Exec(self, feedback_fn):
5687
    """Modifies a node.
5688

5689
    """
5690
    node = self.node
5691
    old_role = self.old_role
5692
    new_role = self.new_role
5693

    
5694
    result = []
5695

    
5696
    if self.op.ndparams:
5697
      node.ndparams = self.new_ndparams
5698

    
5699
    if self.op.powered is not None:
5700
      node.powered = self.op.powered
5701

    
5702
    if self.op.hv_state:
5703
      node.hv_state_static = self.new_hv_state
5704

    
5705
    if self.op.disk_state:
5706
      node.disk_state_static = self.new_disk_state
5707

    
5708
    for attr in ["master_capable", "vm_capable"]:
5709
      val = getattr(self.op, attr)
5710
      if val is not None:
5711
        setattr(node, attr, val)
5712
        result.append((attr, str(val)))
5713

    
5714
    if new_role != old_role:
5715
      # Tell the node to demote itself, if no longer MC and not offline
5716
      if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
5717
        msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
5718
        if msg:
5719
          self.LogWarning("Node failed to demote itself: %s", msg)
5720

    
5721
      new_flags = self._R2F[new_role]
5722
      for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
5723
        if of != nf:
5724
          result.append((desc, str(nf)))
5725
      (node.master_candidate, node.drained, node.offline) = new_flags
5726

    
5727
      # we locked all nodes, we adjust the CP before updating this node
5728
      if self.lock_all:
5729
        _AdjustCandidatePool(self, [node.name])
5730

    
5731
    if self.op.secondary_ip:
5732
      node.secondary_ip = self.op.secondary_ip
5733
      result.append(("secondary_ip", self.op.secondary_ip))
5734

    
5735
    # this will trigger configuration file update, if needed
5736
    self.cfg.Update(node, feedback_fn)
5737

    
5738
    # this will trigger job queue propagation or cleanup if the mc
5739
    # flag changed
5740
    if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
5741
      self.context.ReaddNode(node)
5742

    
5743
    return result
5744

    
5745

    
5746
class LUNodePowercycle(NoHooksLU):
5747
  """Powercycles a node.
5748

5749
  """
5750
  REQ_BGL = False
5751

    
5752
  def CheckArguments(self):
5753
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5754
    if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
5755
      raise errors.OpPrereqError("The node is the master and the force"
5756
                                 " parameter was not set",
5757
                                 errors.ECODE_INVAL)
5758

    
5759
  def ExpandNames(self):
5760
    """Locking for PowercycleNode.
5761

5762
    This is a last-resort option and shouldn't block on other
5763
    jobs. Therefore, we grab no locks.
5764

5765
    """
5766
    self.needed_locks = {}
5767

    
5768
  def Exec(self, feedback_fn):
5769
    """Reboots a node.
5770

5771
    """
5772
    result = self.rpc.call_node_powercycle(self.op.node_name,
5773
                                           self.cfg.GetHypervisorType())
5774
    result.Raise("Failed to schedule the reboot")
5775
    return result.payload
5776

    
5777

    
5778
class LUClusterQuery(NoHooksLU):
5779
  """Query cluster configuration.
5780

5781
  """
5782
  REQ_BGL = False
5783

    
5784
  def ExpandNames(self):
5785
    self.needed_locks = {}
5786

    
5787
  def Exec(self, feedback_fn):
5788
    """Return cluster config.
5789

5790
    """
5791
    cluster = self.cfg.GetClusterInfo()
5792
    os_hvp = {}
5793

    
5794
    # Filter just for enabled hypervisors
5795
    for os_name, hv_dict in cluster.os_hvp.items():
5796
      os_hvp[os_name] = {}
5797
      for hv_name, hv_params in hv_dict.items():
5798
        if hv_name in cluster.enabled_hypervisors:
5799
          os_hvp[os_name][hv_name] = hv_params
5800

    
5801
    # Convert ip_family to ip_version
5802
    primary_ip_version = constants.IP4_VERSION
5803
    if cluster.primary_ip_family == netutils.IP6Address.family:
5804
      primary_ip_version = constants.IP6_VERSION
5805

    
5806
    result = {
5807
      "software_version": constants.RELEASE_VERSION,
5808
      "protocol_version": constants.PROTOCOL_VERSION,
5809
      "config_version": constants.CONFIG_VERSION,
5810
      "os_api_version": max(constants.OS_API_VERSIONS),
5811
      "export_version": constants.EXPORT_VERSION,
5812
      "architecture": (platform.architecture()[0], platform.machine()),
5813
      "name": cluster.cluster_name,
5814
      "master": cluster.master_node,
5815
      "default_hypervisor": cluster.primary_hypervisor,
5816
      "enabled_hypervisors": cluster.enabled_hypervisors,
5817
      "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
5818
                        for hypervisor_name in cluster.enabled_hypervisors]),
5819
      "os_hvp": os_hvp,
5820
      "beparams": cluster.beparams,
5821
      "osparams": cluster.osparams,
5822
      "nicparams": cluster.nicparams,
5823
      "ndparams": cluster.ndparams,
5824
      "candidate_pool_size": cluster.candidate_pool_size,
5825
      "master_netdev": cluster.master_netdev,
5826
      "master_netmask": cluster.master_netmask,
5827
      "use_external_mip_script": cluster.use_external_mip_script,
5828
      "volume_group_name": cluster.volume_group_name,
5829
      "drbd_usermode_helper": cluster.drbd_usermode_helper,
5830
      "file_storage_dir": cluster.file_storage_dir,
5831
      "shared_file_storage_dir": cluster.shared_file_storage_dir,
5832
      "maintain_node_health": cluster.maintain_node_health,
5833
      "ctime": cluster.ctime,
5834
      "mtime": cluster.mtime,
5835
      "uuid": cluster.uuid,
5836
      "tags": list(cluster.GetTags()),
5837
      "uid_pool": cluster.uid_pool,
5838
      "default_iallocator": cluster.default_iallocator,
5839
      "reserved_lvs": cluster.reserved_lvs,
5840
      "primary_ip_version": primary_ip_version,
5841
      "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
5842
      "hidden_os": cluster.hidden_os,
5843
      "blacklisted_os": cluster.blacklisted_os,
5844
      }
5845

    
5846
    return result
5847

    
5848

    
5849
class LUClusterConfigQuery(NoHooksLU):
5850
  """Return configuration values.
5851

5852
  """
5853
  REQ_BGL = False
5854
  _FIELDS_DYNAMIC = utils.FieldSet()
5855
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag",
5856
                                  "watcher_pause", "volume_group_name")
5857

    
5858
  def CheckArguments(self):
5859
    _CheckOutputFields(static=self._FIELDS_STATIC,
5860
                       dynamic=self._FIELDS_DYNAMIC,
5861
                       selected=self.op.output_fields)
5862

    
5863
  def ExpandNames(self):
5864
    self.needed_locks = {}
5865

    
5866
  def Exec(self, feedback_fn):
5867
    """Dump a representation of the cluster config to the standard output.
5868

5869
    """
5870
    values = []
5871
    for field in self.op.output_fields:
5872
      if field == "cluster_name":
5873
        entry = self.cfg.GetClusterName()
5874
      elif field == "master_node":
5875
        entry = self.cfg.GetMasterNode()
5876
      elif field == "drain_flag":
5877
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
5878
      elif field == "watcher_pause":
5879
        entry = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
5880
      elif field == "volume_group_name":
5881
        entry = self.cfg.GetVGName()
5882
      else:
5883
        raise errors.ParameterError(field)
5884
      values.append(entry)
5885
    return values
5886

    
5887

    
5888
class LUInstanceActivateDisks(NoHooksLU):
5889
  """Bring up an instance's disks.
5890

5891
  """
5892
  REQ_BGL = False
5893

    
5894
  def ExpandNames(self):
5895
    self._ExpandAndLockInstance()
5896
    self.needed_locks[locking.LEVEL_NODE] = []
5897
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5898

    
5899
  def DeclareLocks(self, level):
5900
    if level == locking.LEVEL_NODE:
5901
      self._LockInstancesNodes()
5902

    
5903
  def CheckPrereq(self):
5904
    """Check prerequisites.
5905

5906
    This checks that the instance is in the cluster.
5907

5908
    """
5909
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5910
    assert self.instance is not None, \
5911
      "Cannot retrieve locked instance %s" % self.op.instance_name
5912
    _CheckNodeOnline(self, self.instance.primary_node)
5913

    
5914
  def Exec(self, feedback_fn):
5915
    """Activate the disks.
5916

5917
    """
5918
    disks_ok, disks_info = \
5919
              _AssembleInstanceDisks(self, self.instance,
5920
                                     ignore_size=self.op.ignore_size)
5921
    if not disks_ok:
5922
      raise errors.OpExecError("Cannot activate block devices")
5923

    
5924
    return disks_info
5925

    
5926

    
5927
def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
5928
                           ignore_size=False):
5929
  """Prepare the block devices for an instance.
5930

5931
  This sets up the block devices on all nodes.
5932

5933
  @type lu: L{LogicalUnit}
5934
  @param lu: the logical unit on whose behalf we execute
5935
  @type instance: L{objects.Instance}
5936
  @param instance: the instance for whose disks we assemble
5937
  @type disks: list of L{objects.Disk} or None
5938
  @param disks: which disks to assemble (or all, if None)
5939
  @type ignore_secondaries: boolean
5940
  @param ignore_secondaries: if true, errors on secondary nodes
5941
      won't result in an error return from the function
5942
  @type ignore_size: boolean
5943
  @param ignore_size: if true, the current known size of the disk
5944
      will not be used during the disk activation, useful for cases
5945
      when the size is wrong
5946
  @return: False if the operation failed, otherwise a list of
5947
      (host, instance_visible_name, node_visible_name)
5948
      with the mapping from node devices to instance devices
5949

5950
  """
5951
  device_info = []
5952
  disks_ok = True
5953
  iname = instance.name
5954
  disks = _ExpandCheckDisks(instance, disks)
5955

    
5956
  # With the two passes mechanism we try to reduce the window of
5957
  # opportunity for the race condition of switching DRBD to primary
5958
  # before handshaking occured, but we do not eliminate it
5959

    
5960
  # The proper fix would be to wait (with some limits) until the
5961
  # connection has been made and drbd transitions from WFConnection
5962
  # into any other network-connected state (Connected, SyncTarget,
5963
  # SyncSource, etc.)
5964

    
5965
  # 1st pass, assemble on all nodes in secondary mode
5966
  for idx, inst_disk in enumerate(disks):
5967
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
5968
      if ignore_size:
5969
        node_disk = node_disk.Copy()
5970
        node_disk.UnsetSize()
5971
      lu.cfg.SetDiskID(node_disk, node)
5972
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False, idx)
5973
      msg = result.fail_msg
5974
      if msg:
5975
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
5976
                           " (is_primary=False, pass=1): %s",
5977
                           inst_disk.iv_name, node, msg)
5978
        if not ignore_secondaries:
5979
          disks_ok = False
5980

    
5981
  # FIXME: race condition on drbd migration to primary
5982

    
5983
  # 2nd pass, do only the primary node
5984
  for idx, inst_disk in enumerate(disks):
5985
    dev_path = None
5986

    
5987
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
5988
      if node != instance.primary_node:
5989
        continue
5990
      if ignore_size:
5991
        node_disk = node_disk.Copy()
5992
        node_disk.UnsetSize()
5993
      lu.cfg.SetDiskID(node_disk, node)
5994
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True, idx)
5995
      msg = result.fail_msg
5996
      if msg:
5997
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
5998
                           " (is_primary=True, pass=2): %s",
5999
                           inst_disk.iv_name, node, msg)
6000
        disks_ok = False
6001
      else:
6002
        dev_path = result.payload
6003

    
6004
    device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
6005

    
6006
  # leave the disks configured for the primary node
6007
  # this is a workaround that would be fixed better by
6008
  # improving the logical/physical id handling
6009
  for disk in disks:
6010
    lu.cfg.SetDiskID(disk, instance.primary_node)
6011

    
6012
  return disks_ok, device_info
6013

    
6014

    
6015
def _StartInstanceDisks(lu, instance, force):
6016
  """Start the disks of an instance.
6017

6018
  """
6019
  disks_ok, _ = _AssembleInstanceDisks(lu, instance,
6020
                                           ignore_secondaries=force)
6021
  if not disks_ok:
6022
    _ShutdownInstanceDisks(lu, instance)
6023
    if force is not None and not force:
6024
      lu.proc.LogWarning("", hint="If the message above refers to a"
6025
                         " secondary node,"
6026
                         " you can retry the operation using '--force'.")
6027
    raise errors.OpExecError("Disk consistency error")
6028

    
6029

    
6030
class LUInstanceDeactivateDisks(NoHooksLU):
6031
  """Shutdown an instance's disks.
6032

6033
  """
6034
  REQ_BGL = False
6035

    
6036
  def ExpandNames(self):
6037
    self._ExpandAndLockInstance()
6038
    self.needed_locks[locking.LEVEL_NODE] = []
6039
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6040

    
6041
  def DeclareLocks(self, level):
6042
    if level == locking.LEVEL_NODE:
6043
      self._LockInstancesNodes()
6044

    
6045
  def CheckPrereq(self):
6046
    """Check prerequisites.
6047

6048
    This checks that the instance is in the cluster.
6049

6050
    """
6051
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6052
    assert self.instance is not None, \
6053
      "Cannot retrieve locked instance %s" % self.op.instance_name
6054

    
6055
  def Exec(self, feedback_fn):
6056
    """Deactivate the disks
6057

6058
    """
6059
    instance = self.instance
6060
    if self.op.force:
6061
      _ShutdownInstanceDisks(self, instance)
6062
    else:
6063
      _SafeShutdownInstanceDisks(self, instance)
6064

    
6065

    
6066
def _SafeShutdownInstanceDisks(lu, instance, disks=None):
6067
  """Shutdown block devices of an instance.
6068

6069
  This function checks if an instance is running, before calling
6070
  _ShutdownInstanceDisks.
6071

6072
  """
6073
  _CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")
6074
  _ShutdownInstanceDisks(lu, instance, disks=disks)
6075

    
6076

    
6077
def _ExpandCheckDisks(instance, disks):
6078
  """Return the instance disks selected by the disks list
6079

6080
  @type disks: list of L{objects.Disk} or None
6081
  @param disks: selected disks
6082
  @rtype: list of L{objects.Disk}
6083
  @return: selected instance disks to act on
6084

6085
  """
6086
  if disks is None:
6087
    return instance.disks
6088
  else:
6089
    if not set(disks).issubset(instance.disks):
6090
      raise errors.ProgrammerError("Can only act on disks belonging to the"
6091
                                   " target instance")
6092
    return disks
6093

    
6094

    
6095
def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
6096
  """Shutdown block devices of an instance.
6097

6098
  This does the shutdown on all nodes of the instance.
6099

6100
  If the ignore_primary is false, errors on the primary node are
6101
  ignored.
6102

6103
  """
6104
  all_result = True
6105
  disks = _ExpandCheckDisks(instance, disks)
6106

    
6107
  for disk in disks:
6108
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
6109
      lu.cfg.SetDiskID(top_disk, node)
6110
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
6111
      msg = result.fail_msg
6112
      if msg:
6113
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
6114
                      disk.iv_name, node, msg)
6115
        if ((node == instance.primary_node and not ignore_primary) or
6116
            (node != instance.primary_node and not result.offline)):
6117
          all_result = False
6118
  return all_result
6119

    
6120

    
6121
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
6122
  """Checks if a node has enough free memory.
6123

6124
  This function check if a given node has the needed amount of free
6125
  memory. In case the node has less memory or we cannot get the
6126
  information from the node, this function raise an OpPrereqError
6127
  exception.
6128

6129
  @type lu: C{LogicalUnit}
6130
  @param lu: a logical unit from which we get configuration data
6131
  @type node: C{str}
6132
  @param node: the node to check
6133
  @type reason: C{str}
6134
  @param reason: string to use in the error message
6135
  @type requested: C{int}
6136
  @param requested: the amount of memory in MiB to check for
6137
  @type hypervisor_name: C{str}
6138
  @param hypervisor_name: the hypervisor to ask for memory stats
6139
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
6140
      we cannot check the node
6141

6142
  """
6143
  nodeinfo = lu.rpc.call_node_info([node], None, [hypervisor_name])
6144
  nodeinfo[node].Raise("Can't get data from node %s" % node,
6145
                       prereq=True, ecode=errors.ECODE_ENVIRON)
6146
  (_, _, (hv_info, )) = nodeinfo[node].payload
6147

    
6148
  free_mem = hv_info.get("memory_free", None)
6149
  if not isinstance(free_mem, int):
6150
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
6151
                               " was '%s'" % (node, free_mem),
6152
                               errors.ECODE_ENVIRON)
6153
  if requested > free_mem:
6154
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
6155
                               " needed %s MiB, available %s MiB" %
6156
                               (node, reason, requested, free_mem),
6157
                               errors.ECODE_NORES)
6158

    
6159

    
6160
def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
6161
  """Checks if nodes have enough free disk space in the all VGs.
6162

6163
  This function check if all given nodes have the needed amount of
6164
  free disk. In case any node has less disk or we cannot get the
6165
  information from the node, this function raise an OpPrereqError
6166
  exception.
6167

6168
  @type lu: C{LogicalUnit}
6169
  @param lu: a logical unit from which we get configuration data
6170
  @type nodenames: C{list}
6171
  @param nodenames: the list of node names to check
6172
  @type req_sizes: C{dict}
6173
  @param req_sizes: the hash of vg and corresponding amount of disk in
6174
      MiB to check for
6175
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
6176
      or we cannot check the node
6177

6178
  """
6179
  for vg, req_size in req_sizes.items():
6180
    _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
6181

    
6182

    
6183
def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
6184
  """Checks if nodes have enough free disk space in the specified VG.
6185

6186
  This function check if all given nodes have the needed amount of
6187
  free disk. In case any node has less disk or we cannot get the
6188
  information from the node, this function raise an OpPrereqError
6189
  exception.
6190

6191
  @type lu: C{LogicalUnit}
6192
  @param lu: a logical unit from which we get configuration data
6193
  @type nodenames: C{list}
6194
  @param nodenames: the list of node names to check
6195
  @type vg: C{str}
6196
  @param vg: the volume group to check
6197
  @type requested: C{int}
6198
  @param requested: the amount of disk in MiB to check for
6199
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
6200
      or we cannot check the node
6201

6202
  """
6203
  nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None)
6204
  for node in nodenames:
6205
    info = nodeinfo[node]
6206
    info.Raise("Cannot get current information from node %s" % node,
6207
               prereq=True, ecode=errors.ECODE_ENVIRON)
6208
    (_, (vg_info, ), _) = info.payload
6209
    vg_free = vg_info.get("vg_free", None)
6210
    if not isinstance(vg_free, int):
6211
      raise errors.OpPrereqError("Can't compute free disk space on node"
6212
                                 " %s for vg %s, result was '%s'" %
6213
                                 (node, vg, vg_free), errors.ECODE_ENVIRON)
6214
    if requested > vg_free:
6215
      raise errors.OpPrereqError("Not enough disk space on target node %s"
6216
                                 " vg %s: required %d MiB, available %d MiB" %
6217
                                 (node, vg, requested, vg_free),
6218
                                 errors.ECODE_NORES)
6219

    
6220

    
6221
def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):
6222
  """Checks if nodes have enough physical CPUs
6223

6224
  This function checks if all given nodes have the needed number of
6225
  physical CPUs. In case any node has less CPUs or we cannot get the
6226
  information from the node, this function raises an OpPrereqError
6227
  exception.
6228

6229
  @type lu: C{LogicalUnit}
6230
  @param lu: a logical unit from which we get configuration data
6231
  @type nodenames: C{list}
6232
  @param nodenames: the list of node names to check
6233
  @type requested: C{int}
6234
  @param requested: the minimum acceptable number of physical CPUs
6235
  @raise errors.OpPrereqError: if the node doesn't have enough CPUs,
6236
      or we cannot check the node
6237

6238
  """
6239
  nodeinfo = lu.rpc.call_node_info(nodenames, None, [hypervisor_name])
6240
  for node in nodenames:
6241
    info = nodeinfo[node]
6242
    info.Raise("Cannot get current information from node %s" % node,
6243
               prereq=True, ecode=errors.ECODE_ENVIRON)
6244
    (_, _, (hv_info, )) = info.payload
6245
    num_cpus = hv_info.get("cpu_total", None)
6246
    if not isinstance(num_cpus, int):
6247
      raise errors.OpPrereqError("Can't compute the number of physical CPUs"
6248
                                 " on node %s, result was '%s'" %
6249
                                 (node, num_cpus), errors.ECODE_ENVIRON)
6250
    if requested > num_cpus:
6251
      raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "
6252
                                 "required" % (node, num_cpus, requested),
6253
                                 errors.ECODE_NORES)
6254

    
6255

    
6256
class LUInstanceStartup(LogicalUnit):
6257
  """Starts an instance.
6258

6259
  """
6260
  HPATH = "instance-start"
6261
  HTYPE = constants.HTYPE_INSTANCE
6262
  REQ_BGL = False
6263

    
6264
  def CheckArguments(self):
6265
    # extra beparams
6266
    if self.op.beparams:
6267
      # fill the beparams dict
6268
      objects.UpgradeBeParams(self.op.beparams)
6269
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
6270

    
6271
  def ExpandNames(self):
6272
    self._ExpandAndLockInstance()
6273

    
6274
  def BuildHooksEnv(self):
6275
    """Build hooks env.
6276

6277
    This runs on master, primary and secondary nodes of the instance.
6278

6279
    """
6280
    env = {
6281
      "FORCE": self.op.force,
6282
      }
6283

    
6284
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6285

    
6286
    return env
6287

    
6288
  def BuildHooksNodes(self):
6289
    """Build hooks nodes.
6290

6291
    """
6292
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6293
    return (nl, nl)
6294

    
6295
  def CheckPrereq(self):
6296
    """Check prerequisites.
6297

6298
    This checks that the instance is in the cluster.
6299

6300
    """
6301
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6302
    assert self.instance is not None, \
6303
      "Cannot retrieve locked instance %s" % self.op.instance_name
6304

    
6305
    # extra hvparams
6306
    if self.op.hvparams:
6307
      # check hypervisor parameter syntax (locally)
6308
      cluster = self.cfg.GetClusterInfo()
6309
      utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
6310
      filled_hvp = cluster.FillHV(instance)
6311
      filled_hvp.update(self.op.hvparams)
6312
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
6313
      hv_type.CheckParameterSyntax(filled_hvp)
6314
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
6315

    
6316
    _CheckInstanceState(self, instance, INSTANCE_ONLINE)
6317

    
6318
    self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
6319

    
6320
    if self.primary_offline and self.op.ignore_offline_nodes:
6321
      self.proc.LogWarning("Ignoring offline primary node")
6322

    
6323
      if self.op.hvparams or self.op.beparams:
6324
        self.proc.LogWarning("Overridden parameters are ignored")
6325
    else:
6326
      _CheckNodeOnline(self, instance.primary_node)
6327

    
6328
      bep = self.cfg.GetClusterInfo().FillBE(instance)
6329

    
6330
      # check bridges existence
6331
      _CheckInstanceBridgesExist(self, instance)
6332

    
6333
      remote_info = self.rpc.call_instance_info(instance.primary_node,
6334
                                                instance.name,
6335
                                                instance.hypervisor)
6336
      remote_info.Raise("Error checking node %s" % instance.primary_node,
6337
                        prereq=True, ecode=errors.ECODE_ENVIRON)
6338
      if not remote_info.payload: # not running already
6339
        _CheckNodeFreeMemory(self, instance.primary_node,
6340
                             "starting instance %s" % instance.name,
6341
                             bep[constants.BE_MAXMEM], instance.hypervisor)
6342

    
6343
  def Exec(self, feedback_fn):
6344
    """Start the instance.
6345

6346
    """
6347
    instance = self.instance
6348
    force = self.op.force
6349

    
6350
    if not self.op.no_remember:
6351
      self.cfg.MarkInstanceUp(instance.name)
6352

    
6353
    if self.primary_offline:
6354
      assert self.op.ignore_offline_nodes
6355
      self.proc.LogInfo("Primary node offline, marked instance as started")
6356
    else:
6357
      node_current = instance.primary_node
6358

    
6359
      _StartInstanceDisks(self, instance, force)
6360

    
6361
      result = \
6362
        self.rpc.call_instance_start(node_current,
6363
                                     (instance, self.op.hvparams,
6364
                                      self.op.beparams),
6365
                                     self.op.startup_paused)
6366
      msg = result.fail_msg
6367
      if msg:
6368
        _ShutdownInstanceDisks(self, instance)
6369
        raise errors.OpExecError("Could not start instance: %s" % msg)
6370

    
6371

    
6372
class LUInstanceReboot(LogicalUnit):
6373
  """Reboot an instance.
6374

6375
  """
6376
  HPATH = "instance-reboot"
6377
  HTYPE = constants.HTYPE_INSTANCE
6378
  REQ_BGL = False
6379

    
6380
  def ExpandNames(self):
6381
    self._ExpandAndLockInstance()
6382

    
6383
  def BuildHooksEnv(self):
6384
    """Build hooks env.
6385

6386
    This runs on master, primary and secondary nodes of the instance.
6387

6388
    """
6389
    env = {
6390
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
6391
      "REBOOT_TYPE": self.op.reboot_type,
6392
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6393
      }
6394

    
6395
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6396

    
6397
    return env
6398

    
6399
  def BuildHooksNodes(self):
6400
    """Build hooks nodes.
6401

6402
    """
6403
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6404
    return (nl, nl)
6405

    
6406
  def CheckPrereq(self):
6407
    """Check prerequisites.
6408

6409
    This checks that the instance is in the cluster.
6410

6411
    """
6412
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6413
    assert self.instance is not None, \
6414
      "Cannot retrieve locked instance %s" % self.op.instance_name
6415
    _CheckInstanceState(self, instance, INSTANCE_ONLINE)
6416
    _CheckNodeOnline(self, instance.primary_node)
6417

    
6418
    # check bridges existence
6419
    _CheckInstanceBridgesExist(self, instance)
6420

    
6421
  def Exec(self, feedback_fn):
6422
    """Reboot the instance.
6423

6424
    """
6425
    instance = self.instance
6426
    ignore_secondaries = self.op.ignore_secondaries
6427
    reboot_type = self.op.reboot_type
6428

    
6429
    remote_info = self.rpc.call_instance_info(instance.primary_node,
6430
                                              instance.name,
6431
                                              instance.hypervisor)
6432
    remote_info.Raise("Error checking node %s" % instance.primary_node)
6433
    instance_running = bool(remote_info.payload)
6434

    
6435
    node_current = instance.primary_node
6436

    
6437
    if instance_running and reboot_type in [constants.INSTANCE_REBOOT_SOFT,
6438
                                            constants.INSTANCE_REBOOT_HARD]:
6439
      for disk in instance.disks:
6440
        self.cfg.SetDiskID(disk, node_current)
6441
      result = self.rpc.call_instance_reboot(node_current, instance,
6442
                                             reboot_type,
6443
                                             self.op.shutdown_timeout)
6444
      result.Raise("Could not reboot instance")
6445
    else:
6446
      if instance_running:
6447
        result = self.rpc.call_instance_shutdown(node_current, instance,
6448
                                                 self.op.shutdown_timeout)
6449
        result.Raise("Could not shutdown instance for full reboot")
6450
        _ShutdownInstanceDisks(self, instance)
6451
      else:
6452
        self.LogInfo("Instance %s was already stopped, starting now",
6453
                     instance.name)
6454
      _StartInstanceDisks(self, instance, ignore_secondaries)
6455
      result = self.rpc.call_instance_start(node_current,
6456
                                            (instance, None, None), False)
6457
      msg = result.fail_msg
6458
      if msg:
6459
        _ShutdownInstanceDisks(self, instance)
6460
        raise errors.OpExecError("Could not start instance for"
6461
                                 " full reboot: %s" % msg)
6462

    
6463
    self.cfg.MarkInstanceUp(instance.name)
6464

    
6465

    
6466
class LUInstanceShutdown(LogicalUnit):
6467
  """Shutdown an instance.
6468

6469
  """
6470
  HPATH = "instance-stop"
6471
  HTYPE = constants.HTYPE_INSTANCE
6472
  REQ_BGL = False
6473

    
6474
  def ExpandNames(self):
6475
    self._ExpandAndLockInstance()
6476

    
6477
  def BuildHooksEnv(self):
6478
    """Build hooks env.
6479

6480
    This runs on master, primary and secondary nodes of the instance.
6481

6482
    """
6483
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6484
    env["TIMEOUT"] = self.op.timeout
6485
    return env
6486

    
6487
  def BuildHooksNodes(self):
6488
    """Build hooks nodes.
6489

6490
    """
6491
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6492
    return (nl, nl)
6493

    
6494
  def CheckPrereq(self):
6495
    """Check prerequisites.
6496

6497
    This checks that the instance is in the cluster.
6498

6499
    """
6500
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6501
    assert self.instance is not None, \
6502
      "Cannot retrieve locked instance %s" % self.op.instance_name
6503

    
6504
    _CheckInstanceState(self, self.instance, INSTANCE_ONLINE)
6505

    
6506
    self.primary_offline = \
6507
      self.cfg.GetNodeInfo(self.instance.primary_node).offline
6508

    
6509
    if self.primary_offline and self.op.ignore_offline_nodes:
6510
      self.proc.LogWarning("Ignoring offline primary node")
6511
    else:
6512
      _CheckNodeOnline(self, self.instance.primary_node)
6513

    
6514
  def Exec(self, feedback_fn):
6515
    """Shutdown the instance.
6516

6517
    """
6518
    instance = self.instance
6519
    node_current = instance.primary_node
6520
    timeout = self.op.timeout
6521

    
6522
    if not self.op.no_remember:
6523
      self.cfg.MarkInstanceDown(instance.name)
6524

    
6525
    if self.primary_offline:
6526
      assert self.op.ignore_offline_nodes
6527
      self.proc.LogInfo("Primary node offline, marked instance as stopped")
6528
    else:
6529
      result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
6530
      msg = result.fail_msg
6531
      if msg:
6532
        self.proc.LogWarning("Could not shutdown instance: %s" % msg)
6533

    
6534
      _ShutdownInstanceDisks(self, instance)
6535

    
6536

    
6537
class LUInstanceReinstall(LogicalUnit):
6538
  """Reinstall an instance.
6539

6540
  """
6541
  HPATH = "instance-reinstall"
6542
  HTYPE = constants.HTYPE_INSTANCE
6543
  REQ_BGL = False
6544

    
6545
  def ExpandNames(self):
6546
    self._ExpandAndLockInstance()
6547

    
6548
  def BuildHooksEnv(self):
6549
    """Build hooks env.
6550

6551
    This runs on master, primary and secondary nodes of the instance.
6552

6553
    """
6554
    return _BuildInstanceHookEnvByObject(self, self.instance)
6555

    
6556
  def BuildHooksNodes(self):
6557
    """Build hooks nodes.
6558

6559
    """
6560
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6561
    return (nl, nl)
6562

    
6563
  def CheckPrereq(self):
6564
    """Check prerequisites.
6565

6566
    This checks that the instance is in the cluster and is not running.
6567

6568
    """
6569
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6570
    assert instance is not None, \
6571
      "Cannot retrieve locked instance %s" % self.op.instance_name
6572
    _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
6573
                     " offline, cannot reinstall")
6574
    for node in instance.secondary_nodes:
6575
      _CheckNodeOnline(self, node, "Instance secondary node offline,"
6576
                       " cannot reinstall")
6577

    
6578
    if instance.disk_template == constants.DT_DISKLESS:
6579
      raise errors.OpPrereqError("Instance '%s' has no disks" %
6580
                                 self.op.instance_name,
6581
                                 errors.ECODE_INVAL)
6582
    _CheckInstanceState(self, instance, INSTANCE_DOWN, msg="cannot reinstall")
6583

    
6584
    if self.op.os_type is not None:
6585
      # OS verification
6586
      pnode = _ExpandNodeName(self.cfg, instance.primary_node)
6587
      _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
6588
      instance_os = self.op.os_type
6589
    else:
6590
      instance_os = instance.os
6591

    
6592
    nodelist = list(instance.all_nodes)
6593

    
6594
    if self.op.osparams:
6595
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
6596
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
6597
      self.os_inst = i_osdict # the new dict (without defaults)
6598
    else:
6599
      self.os_inst = None
6600

    
6601
    self.instance = instance
6602

    
6603
  def Exec(self, feedback_fn):
6604
    """Reinstall the instance.
6605

6606
    """
6607
    inst = self.instance
6608

    
6609
    if self.op.os_type is not None:
6610
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
6611
      inst.os = self.op.os_type
6612
      # Write to configuration
6613
      self.cfg.Update(inst, feedback_fn)
6614

    
6615
    _StartInstanceDisks(self, inst, None)
6616
    try:
6617
      feedback_fn("Running the instance OS create scripts...")
6618
      # FIXME: pass debug option from opcode to backend
6619
      result = self.rpc.call_instance_os_add(inst.primary_node,
6620
                                             (inst, self.os_inst), True,
6621
                                             self.op.debug_level)
6622
      result.Raise("Could not install OS for instance %s on node %s" %
6623
                   (inst.name, inst.primary_node))
6624
    finally:
6625
      _ShutdownInstanceDisks(self, inst)
6626

    
6627

    
6628
class LUInstanceRecreateDisks(LogicalUnit):
6629
  """Recreate an instance's missing disks.
6630

6631
  """
6632
  HPATH = "instance-recreate-disks"
6633
  HTYPE = constants.HTYPE_INSTANCE
6634
  REQ_BGL = False
6635

    
6636
  def CheckArguments(self):
6637
    # normalise the disk list
6638
    self.op.disks = sorted(frozenset(self.op.disks))
6639

    
6640
  def ExpandNames(self):
6641
    self._ExpandAndLockInstance()
6642
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6643
    if self.op.nodes:
6644
      self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
6645
      self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
6646
    else:
6647
      self.needed_locks[locking.LEVEL_NODE] = []
6648

    
6649
  def DeclareLocks(self, level):
6650
    if level == locking.LEVEL_NODE:
6651
      # if we replace the nodes, we only need to lock the old primary,
6652
      # otherwise we need to lock all nodes for disk re-creation
6653
      primary_only = bool(self.op.nodes)
6654
      self._LockInstancesNodes(primary_only=primary_only)
6655
    elif level == locking.LEVEL_NODE_RES:
6656
      # Copy node locks
6657
      self.needed_locks[locking.LEVEL_NODE_RES] = \
6658
        self.needed_locks[locking.LEVEL_NODE][:]
6659

    
6660
  def BuildHooksEnv(self):
6661
    """Build hooks env.
6662

6663
    This runs on master, primary and secondary nodes of the instance.
6664

6665
    """
6666
    return _BuildInstanceHookEnvByObject(self, self.instance)
6667

    
6668
  def BuildHooksNodes(self):
6669
    """Build hooks nodes.
6670

6671
    """
6672
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6673
    return (nl, nl)
6674

    
6675
  def CheckPrereq(self):
6676
    """Check prerequisites.
6677

6678
    This checks that the instance is in the cluster and is not running.
6679

6680
    """
6681
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6682
    assert instance is not None, \
6683
      "Cannot retrieve locked instance %s" % self.op.instance_name
6684
    if self.op.nodes:
6685
      if len(self.op.nodes) != len(instance.all_nodes):
6686
        raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
6687
                                   " %d replacement nodes were specified" %
6688
                                   (instance.name, len(instance.all_nodes),
6689
                                    len(self.op.nodes)),
6690
                                   errors.ECODE_INVAL)
6691
      assert instance.disk_template != constants.DT_DRBD8 or \
6692
          len(self.op.nodes) == 2
6693
      assert instance.disk_template != constants.DT_PLAIN or \
6694
          len(self.op.nodes) == 1
6695
      primary_node = self.op.nodes[0]
6696
    else:
6697
      primary_node = instance.primary_node
6698
    _CheckNodeOnline(self, primary_node)
6699

    
6700
    if instance.disk_template == constants.DT_DISKLESS:
6701
      raise errors.OpPrereqError("Instance '%s' has no disks" %
6702
                                 self.op.instance_name, errors.ECODE_INVAL)
6703
    # if we replace nodes *and* the old primary is offline, we don't
6704
    # check
6705
    assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE)
6706
    assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE_RES)
6707
    old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
6708
    if not (self.op.nodes and old_pnode.offline):
6709
      _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
6710
                          msg="cannot recreate disks")
6711

    
6712
    if not self.op.disks:
6713
      self.op.disks = range(len(instance.disks))
6714
    else:
6715
      for idx in self.op.disks:
6716
        if idx >= len(instance.disks):
6717
          raise errors.OpPrereqError("Invalid disk index '%s'" % idx,
6718
                                     errors.ECODE_INVAL)
6719
    if self.op.disks != range(len(instance.disks)) and self.op.nodes:
6720
      raise errors.OpPrereqError("Can't recreate disks partially and"
6721
                                 " change the nodes at the same time",
6722
                                 errors.ECODE_INVAL)
6723
    self.instance = instance
6724

    
6725
  def Exec(self, feedback_fn):
6726
    """Recreate the disks.
6727

6728
    """
6729
    instance = self.instance
6730

    
6731
    assert (self.owned_locks(locking.LEVEL_NODE) ==
6732
            self.owned_locks(locking.LEVEL_NODE_RES))
6733

    
6734
    to_skip = []
6735
    mods = [] # keeps track of needed logical_id changes
6736

    
6737
    for idx, disk in enumerate(instance.disks):
6738
      if idx not in self.op.disks: # disk idx has not been passed in
6739
        to_skip.append(idx)
6740
        continue
6741
      # update secondaries for disks, if needed
6742
      if self.op.nodes:
6743
        if disk.dev_type == constants.LD_DRBD8:
6744
          # need to update the nodes and minors
6745
          assert len(self.op.nodes) == 2
6746
          assert len(disk.logical_id) == 6 # otherwise disk internals
6747
                                           # have changed
6748
          (_, _, old_port, _, _, old_secret) = disk.logical_id
6749
          new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
6750
          new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
6751
                    new_minors[0], new_minors[1], old_secret)
6752
          assert len(disk.logical_id) == len(new_id)
6753
          mods.append((idx, new_id))
6754

    
6755
    # now that we have passed all asserts above, we can apply the mods
6756
    # in a single run (to avoid partial changes)
6757
    for idx, new_id in mods:
6758
      instance.disks[idx].logical_id = new_id
6759

    
6760
    # change primary node, if needed
6761
    if self.op.nodes:
6762
      instance.primary_node = self.op.nodes[0]
6763
      self.LogWarning("Changing the instance's nodes, you will have to"
6764
                      " remove any disks left on the older nodes manually")
6765

    
6766
    if self.op.nodes:
6767
      self.cfg.Update(instance, feedback_fn)
6768

    
6769
    _CreateDisks(self, instance, to_skip=to_skip)
6770

    
6771

    
6772
class LUInstanceRename(LogicalUnit):
6773
  """Rename an instance.
6774

6775
  """
6776
  HPATH = "instance-rename"
6777
  HTYPE = constants.HTYPE_INSTANCE
6778

    
6779
  def CheckArguments(self):
6780
    """Check arguments.
6781

6782
    """
6783
    if self.op.ip_check and not self.op.name_check:
6784
      # TODO: make the ip check more flexible and not depend on the name check
6785
      raise errors.OpPrereqError("IP address check requires a name check",
6786
                                 errors.ECODE_INVAL)
6787

    
6788
  def BuildHooksEnv(self):
6789
    """Build hooks env.
6790

6791
    This runs on master, primary and secondary nodes of the instance.
6792

6793
    """
6794
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6795
    env["INSTANCE_NEW_NAME"] = self.op.new_name
6796
    return env
6797

    
6798
  def BuildHooksNodes(self):
6799
    """Build hooks nodes.
6800

6801
    """
6802
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6803
    return (nl, nl)
6804

    
6805
  def CheckPrereq(self):
6806
    """Check prerequisites.
6807

6808
    This checks that the instance is in the cluster and is not running.
6809

6810
    """
6811
    self.op.instance_name = _ExpandInstanceName(self.cfg,
6812
                                                self.op.instance_name)
6813
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6814
    assert instance is not None
6815
    _CheckNodeOnline(self, instance.primary_node)
6816
    _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
6817
                        msg="cannot rename")
6818
    self.instance = instance
6819

    
6820
    new_name = self.op.new_name
6821
    if self.op.name_check:
6822
      hostname = netutils.GetHostname(name=new_name)
6823
      if hostname.name != new_name:
6824
        self.LogInfo("Resolved given name '%s' to '%s'", new_name,
6825
                     hostname.name)
6826
      if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
6827
        raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
6828
                                    " same as given hostname '%s'") %
6829
                                    (hostname.name, self.op.new_name),
6830
                                    errors.ECODE_INVAL)
6831
      new_name = self.op.new_name = hostname.name
6832
      if (self.op.ip_check and
6833
          netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
6834
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
6835
                                   (hostname.ip, new_name),
6836
                                   errors.ECODE_NOTUNIQUE)
6837

    
6838
    instance_list = self.cfg.GetInstanceList()
6839
    if new_name in instance_list and new_name != instance.name:
6840
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
6841
                                 new_name, errors.ECODE_EXISTS)
6842

    
6843
  def Exec(self, feedback_fn):
6844
    """Rename the instance.
6845

6846
    """
6847
    inst = self.instance
6848
    old_name = inst.name
6849

    
6850
    rename_file_storage = False
6851
    if (inst.disk_template in constants.DTS_FILEBASED and
6852
        self.op.new_name != inst.name):
6853
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
6854
      rename_file_storage = True
6855

    
6856
    self.cfg.RenameInstance(inst.name, self.op.new_name)
6857
    # Change the instance lock. This is definitely safe while we hold the BGL.
6858
    # Otherwise the new lock would have to be added in acquired mode.
6859
    assert self.REQ_BGL
6860
    self.glm.remove(locking.LEVEL_INSTANCE, old_name)
6861
    self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
6862

    
6863
    # re-read the instance from the configuration after rename
6864
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
6865

    
6866
    if rename_file_storage:
6867
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
6868
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
6869
                                                     old_file_storage_dir,
6870
                                                     new_file_storage_dir)
6871
      result.Raise("Could not rename on node %s directory '%s' to '%s'"
6872
                   " (but the instance has been renamed in Ganeti)" %
6873
                   (inst.primary_node, old_file_storage_dir,
6874
                    new_file_storage_dir))
6875

    
6876
    _StartInstanceDisks(self, inst, None)
6877
    try:
6878
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
6879
                                                 old_name, self.op.debug_level)
6880
      msg = result.fail_msg
6881
      if msg:
6882
        msg = ("Could not run OS rename script for instance %s on node %s"
6883
               " (but the instance has been renamed in Ganeti): %s" %
6884
               (inst.name, inst.primary_node, msg))
6885
        self.proc.LogWarning(msg)
6886
    finally:
6887
      _ShutdownInstanceDisks(self, inst)
6888

    
6889
    return inst.name
6890

    
6891

    
6892
class LUInstanceRemove(LogicalUnit):
6893
  """Remove an instance.
6894

6895
  """
6896
  HPATH = "instance-remove"
6897
  HTYPE = constants.HTYPE_INSTANCE
6898
  REQ_BGL = False
6899

    
6900
  def ExpandNames(self):
6901
    self._ExpandAndLockInstance()
6902
    self.needed_locks[locking.LEVEL_NODE] = []
6903
    self.needed_locks[locking.LEVEL_NODE_RES] = []
6904
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6905

    
6906
  def DeclareLocks(self, level):
6907
    if level == locking.LEVEL_NODE:
6908
      self._LockInstancesNodes()
6909
    elif level == locking.LEVEL_NODE_RES:
6910
      # Copy node locks
6911
      self.needed_locks[locking.LEVEL_NODE_RES] = \
6912
        self.needed_locks[locking.LEVEL_NODE][:]
6913

    
6914
  def BuildHooksEnv(self):
6915
    """Build hooks env.
6916

6917
    This runs on master, primary and secondary nodes of the instance.
6918

6919
    """
6920
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6921
    env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
6922
    return env
6923

    
6924
  def BuildHooksNodes(self):
6925
    """Build hooks nodes.
6926

6927
    """
6928
    nl = [self.cfg.GetMasterNode()]
6929
    nl_post = list(self.instance.all_nodes) + nl
6930
    return (nl, nl_post)
6931

    
6932
  def CheckPrereq(self):
6933
    """Check prerequisites.
6934

6935
    This checks that the instance is in the cluster.
6936

6937
    """
6938
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6939
    assert self.instance is not None, \
6940
      "Cannot retrieve locked instance %s" % self.op.instance_name
6941

    
6942
  def Exec(self, feedback_fn):
6943
    """Remove the instance.
6944

6945
    """
6946
    instance = self.instance
6947
    logging.info("Shutting down instance %s on node %s",
6948
                 instance.name, instance.primary_node)
6949

    
6950
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
6951
                                             self.op.shutdown_timeout)
6952
    msg = result.fail_msg
6953
    if msg:
6954
      if self.op.ignore_failures:
6955
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
6956
      else:
6957
        raise errors.OpExecError("Could not shutdown instance %s on"
6958
                                 " node %s: %s" %
6959
                                 (instance.name, instance.primary_node, msg))
6960

    
6961
    assert (self.owned_locks(locking.LEVEL_NODE) ==
6962
            self.owned_locks(locking.LEVEL_NODE_RES))
6963
    assert not (set(instance.all_nodes) -
6964
                self.owned_locks(locking.LEVEL_NODE)), \
6965
      "Not owning correct locks"
6966

    
6967
    _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
6968

    
6969

    
6970
def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
6971
  """Utility function to remove an instance.
6972

6973
  """
6974
  logging.info("Removing block devices for instance %s", instance.name)
6975

    
6976
  if not _RemoveDisks(lu, instance):
6977
    if not ignore_failures:
6978
      raise errors.OpExecError("Can't remove instance's disks")
6979
    feedback_fn("Warning: can't remove instance's disks")
6980

    
6981
  logging.info("Removing instance %s out of cluster config", instance.name)
6982

    
6983
  lu.cfg.RemoveInstance(instance.name)
6984

    
6985
  assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
6986
    "Instance lock removal conflict"
6987

    
6988
  # Remove lock for the instance
6989
  lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
6990

    
6991

    
6992
class LUInstanceQuery(NoHooksLU):
6993
  """Logical unit for querying instances.
6994

6995
  """
6996
  # pylint: disable=W0142
6997
  REQ_BGL = False
6998

    
6999
  def CheckArguments(self):
7000
    self.iq = _InstanceQuery(qlang.MakeSimpleFilter("name", self.op.names),
7001
                             self.op.output_fields, self.op.use_locking)
7002

    
7003
  def ExpandNames(self):
7004
    self.iq.ExpandNames(self)
7005

    
7006
  def DeclareLocks(self, level):
7007
    self.iq.DeclareLocks(self, level)
7008

    
7009
  def Exec(self, feedback_fn):
7010
    return self.iq.OldStyleQuery(self)
7011

    
7012

    
7013
class LUInstanceFailover(LogicalUnit):
7014
  """Failover an instance.
7015

7016
  """
7017
  HPATH = "instance-failover"
7018
  HTYPE = constants.HTYPE_INSTANCE
7019
  REQ_BGL = False
7020

    
7021
  def CheckArguments(self):
7022
    """Check the arguments.
7023

7024
    """
7025
    self.iallocator = getattr(self.op, "iallocator", None)
7026
    self.target_node = getattr(self.op, "target_node", None)
7027

    
7028
  def ExpandNames(self):
7029
    self._ExpandAndLockInstance()
7030

    
7031
    if self.op.target_node is not None:
7032
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7033

    
7034
    self.needed_locks[locking.LEVEL_NODE] = []
7035
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7036

    
7037
    ignore_consistency = self.op.ignore_consistency
7038
    shutdown_timeout = self.op.shutdown_timeout
7039
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
7040
                                       cleanup=False,
7041
                                       failover=True,
7042
                                       ignore_consistency=ignore_consistency,
7043
                                       shutdown_timeout=shutdown_timeout)
7044
    self.tasklets = [self._migrater]
7045

    
7046
  def DeclareLocks(self, level):
7047
    if level == locking.LEVEL_NODE:
7048
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
7049
      if instance.disk_template in constants.DTS_EXT_MIRROR:
7050
        if self.op.target_node is None:
7051
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7052
        else:
7053
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
7054
                                                   self.op.target_node]
7055
        del self.recalculate_locks[locking.LEVEL_NODE]
7056
      else:
7057
        self._LockInstancesNodes()
7058

    
7059
  def BuildHooksEnv(self):
7060
    """Build hooks env.
7061

7062
    This runs on master, primary and secondary nodes of the instance.
7063

7064
    """
7065
    instance = self._migrater.instance
7066
    source_node = instance.primary_node
7067
    target_node = self.op.target_node
7068
    env = {
7069
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
7070
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
7071
      "OLD_PRIMARY": source_node,
7072
      "NEW_PRIMARY": target_node,
7073
      }
7074

    
7075
    if instance.disk_template in constants.DTS_INT_MIRROR:
7076
      env["OLD_SECONDARY"] = instance.secondary_nodes[0]
7077
      env["NEW_SECONDARY"] = source_node
7078
    else:
7079
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = ""
7080

    
7081
    env.update(_BuildInstanceHookEnvByObject(self, instance))
7082

    
7083
    return env
7084

    
7085
  def BuildHooksNodes(self):
7086
    """Build hooks nodes.
7087

7088
    """
7089
    instance = self._migrater.instance
7090
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
7091
    return (nl, nl + [instance.primary_node])
7092

    
7093

    
7094
class LUInstanceMigrate(LogicalUnit):
7095
  """Migrate an instance.
7096

7097
  This is migration without shutting down, compared to the failover,
7098
  which is done with shutdown.
7099

7100
  """
7101
  HPATH = "instance-migrate"
7102
  HTYPE = constants.HTYPE_INSTANCE
7103
  REQ_BGL = False
7104

    
7105
  def ExpandNames(self):
7106
    self._ExpandAndLockInstance()
7107

    
7108
    if self.op.target_node is not None:
7109
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7110

    
7111
    self.needed_locks[locking.LEVEL_NODE] = []
7112
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7113

    
7114
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
7115
                                       cleanup=self.op.cleanup,
7116
                                       failover=False,
7117
                                       fallback=self.op.allow_failover)
7118
    self.tasklets = [self._migrater]
7119

    
7120
  def DeclareLocks(self, level):
7121
    if level == locking.LEVEL_NODE:
7122
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
7123
      if instance.disk_template in constants.DTS_EXT_MIRROR:
7124
        if self.op.target_node is None:
7125
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7126
        else:
7127
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
7128
                                                   self.op.target_node]
7129
        del self.recalculate_locks[locking.LEVEL_NODE]
7130
      else:
7131
        self._LockInstancesNodes()
7132

    
7133
  def BuildHooksEnv(self):
7134
    """Build hooks env.
7135

7136
    This runs on master, primary and secondary nodes of the instance.
7137

7138
    """
7139
    instance = self._migrater.instance
7140
    source_node = instance.primary_node
7141
    target_node = self.op.target_node
7142
    env = _BuildInstanceHookEnvByObject(self, instance)
7143
    env.update({
7144
      "MIGRATE_LIVE": self._migrater.live,
7145
      "MIGRATE_CLEANUP": self.op.cleanup,
7146
      "OLD_PRIMARY": source_node,
7147
      "NEW_PRIMARY": target_node,
7148
      })
7149

    
7150
    if instance.disk_template in constants.DTS_INT_MIRROR:
7151
      env["OLD_SECONDARY"] = target_node
7152
      env["NEW_SECONDARY"] = source_node
7153
    else:
7154
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
7155

    
7156
    return env
7157

    
7158
  def BuildHooksNodes(self):
7159
    """Build hooks nodes.
7160

7161
    """
7162
    instance = self._migrater.instance
7163
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
7164
    return (nl, nl + [instance.primary_node])
7165

    
7166

    
7167
class LUInstanceMove(LogicalUnit):
7168
  """Move an instance by data-copying.
7169

7170
  """
7171
  HPATH = "instance-move"
7172
  HTYPE = constants.HTYPE_INSTANCE
7173
  REQ_BGL = False
7174

    
7175
  def ExpandNames(self):
7176
    self._ExpandAndLockInstance()
7177
    target_node = _ExpandNodeName(self.cfg, self.op.target_node)
7178
    self.op.target_node = target_node
7179
    self.needed_locks[locking.LEVEL_NODE] = [target_node]
7180
    self.needed_locks[locking.LEVEL_NODE_RES] = []
7181
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
7182

    
7183
  def DeclareLocks(self, level):
7184
    if level == locking.LEVEL_NODE:
7185
      self._LockInstancesNodes(primary_only=True)
7186
    elif level == locking.LEVEL_NODE_RES:
7187
      # Copy node locks
7188
      self.needed_locks[locking.LEVEL_NODE_RES] = \
7189
        self.needed_locks[locking.LEVEL_NODE][:]
7190

    
7191
  def BuildHooksEnv(self):
7192
    """Build hooks env.
7193

7194
    This runs on master, primary and secondary nodes of the instance.
7195

7196
    """
7197
    env = {
7198
      "TARGET_NODE": self.op.target_node,
7199
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
7200
      }
7201
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
7202
    return env
7203

    
7204
  def BuildHooksNodes(self):
7205
    """Build hooks nodes.
7206

7207
    """
7208
    nl = [
7209
      self.cfg.GetMasterNode(),
7210
      self.instance.primary_node,
7211
      self.op.target_node,
7212
      ]
7213
    return (nl, nl)
7214

    
7215
  def CheckPrereq(self):
7216
    """Check prerequisites.
7217

7218
    This checks that the instance is in the cluster.
7219

7220
    """
7221
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7222
    assert self.instance is not None, \
7223
      "Cannot retrieve locked instance %s" % self.op.instance_name
7224

    
7225
    node = self.cfg.GetNodeInfo(self.op.target_node)
7226
    assert node is not None, \
7227
      "Cannot retrieve locked node %s" % self.op.target_node
7228

    
7229
    self.target_node = target_node = node.name
7230

    
7231
    if target_node == instance.primary_node:
7232
      raise errors.OpPrereqError("Instance %s is already on the node %s" %
7233
                                 (instance.name, target_node),
7234
                                 errors.ECODE_STATE)
7235

    
7236
    bep = self.cfg.GetClusterInfo().FillBE(instance)
7237

    
7238
    for idx, dsk in enumerate(instance.disks):
7239
      if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
7240
        raise errors.OpPrereqError("Instance disk %d has a complex layout,"
7241
                                   " cannot copy" % idx, errors.ECODE_STATE)
7242

    
7243
    _CheckNodeOnline(self, target_node)
7244
    _CheckNodeNotDrained(self, target_node)
7245
    _CheckNodeVmCapable(self, target_node)
7246

    
7247
    if instance.admin_state == constants.ADMINST_UP:
7248
      # check memory requirements on the secondary node
7249
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
7250
                           instance.name, bep[constants.BE_MAXMEM],
7251
                           instance.hypervisor)
7252
    else:
7253
      self.LogInfo("Not checking memory on the secondary node as"
7254
                   " instance will not be started")
7255

    
7256
    # check bridge existance
7257
    _CheckInstanceBridgesExist(self, instance, node=target_node)
7258

    
7259
  def Exec(self, feedback_fn):
7260
    """Move an instance.
7261

7262
    The move is done by shutting it down on its present node, copying
7263
    the data over (slow) and starting it on the new node.
7264

7265
    """
7266
    instance = self.instance
7267

    
7268
    source_node = instance.primary_node
7269
    target_node = self.target_node
7270

    
7271
    self.LogInfo("Shutting down instance %s on source node %s",
7272
                 instance.name, source_node)
7273

    
7274
    assert (self.owned_locks(locking.LEVEL_NODE) ==
7275
            self.owned_locks(locking.LEVEL_NODE_RES))
7276

    
7277
    result = self.rpc.call_instance_shutdown(source_node, instance,
7278
                                             self.op.shutdown_timeout)
7279
    msg = result.fail_msg
7280
    if msg:
7281
      if self.op.ignore_consistency:
7282
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
7283
                             " Proceeding anyway. Please make sure node"
7284
                             " %s is down. Error details: %s",
7285
                             instance.name, source_node, source_node, msg)
7286
      else:
7287
        raise errors.OpExecError("Could not shutdown instance %s on"
7288
                                 " node %s: %s" %
7289
                                 (instance.name, source_node, msg))
7290

    
7291
    # create the target disks
7292
    try:
7293
      _CreateDisks(self, instance, target_node=target_node)
7294
    except errors.OpExecError:
7295
      self.LogWarning("Device creation failed, reverting...")
7296
      try:
7297
        _RemoveDisks(self, instance, target_node=target_node)
7298
      finally:
7299
        self.cfg.ReleaseDRBDMinors(instance.name)
7300
        raise
7301

    
7302
    cluster_name = self.cfg.GetClusterInfo().cluster_name
7303

    
7304
    errs = []
7305
    # activate, get path, copy the data over
7306
    for idx, disk in enumerate(instance.disks):
7307
      self.LogInfo("Copying data for disk %d", idx)
7308
      result = self.rpc.call_blockdev_assemble(target_node, disk,
7309
                                               instance.name, True, idx)
7310
      if result.fail_msg:
7311
        self.LogWarning("Can't assemble newly created disk %d: %s",
7312
                        idx, result.fail_msg)
7313
        errs.append(result.fail_msg)
7314
        break
7315
      dev_path = result.payload
7316
      result = self.rpc.call_blockdev_export(source_node, disk,
7317
                                             target_node, dev_path,
7318
                                             cluster_name)
7319
      if result.fail_msg:
7320
        self.LogWarning("Can't copy data over for disk %d: %s",
7321
                        idx, result.fail_msg)
7322
        errs.append(result.fail_msg)
7323
        break
7324

    
7325
    if errs:
7326
      self.LogWarning("Some disks failed to copy, aborting")
7327
      try:
7328
        _RemoveDisks(self, instance, target_node=target_node)
7329
      finally:
7330
        self.cfg.ReleaseDRBDMinors(instance.name)
7331
        raise errors.OpExecError("Errors during disk copy: %s" %
7332
                                 (",".join(errs),))
7333

    
7334
    instance.primary_node = target_node
7335
    self.cfg.Update(instance, feedback_fn)
7336

    
7337
    self.LogInfo("Removing the disks on the original node")
7338
    _RemoveDisks(self, instance, target_node=source_node)
7339

    
7340
    # Only start the instance if it's marked as up
7341
    if instance.admin_state == constants.ADMINST_UP:
7342
      self.LogInfo("Starting instance %s on node %s",
7343
                   instance.name, target_node)
7344

    
7345
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
7346
                                           ignore_secondaries=True)
7347
      if not disks_ok:
7348
        _ShutdownInstanceDisks(self, instance)
7349
        raise errors.OpExecError("Can't activate the instance's disks")
7350

    
7351
      result = self.rpc.call_instance_start(target_node,
7352
                                            (instance, None, None), False)
7353
      msg = result.fail_msg
7354
      if msg:
7355
        _ShutdownInstanceDisks(self, instance)
7356
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
7357
                                 (instance.name, target_node, msg))
7358

    
7359

    
7360
class LUNodeMigrate(LogicalUnit):
7361
  """Migrate all instances from a node.
7362

7363
  """
7364
  HPATH = "node-migrate"
7365
  HTYPE = constants.HTYPE_NODE
7366
  REQ_BGL = False
7367

    
7368
  def CheckArguments(self):
7369
    pass
7370

    
7371
  def ExpandNames(self):
7372
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
7373

    
7374
    self.share_locks = _ShareAll()
7375
    self.needed_locks = {
7376
      locking.LEVEL_NODE: [self.op.node_name],
7377
      }
7378

    
7379
  def BuildHooksEnv(self):
7380
    """Build hooks env.
7381

7382
    This runs on the master, the primary and all the secondaries.
7383

7384
    """
7385
    return {
7386
      "NODE_NAME": self.op.node_name,
7387
      }
7388

    
7389
  def BuildHooksNodes(self):
7390
    """Build hooks nodes.
7391

7392
    """
7393
    nl = [self.cfg.GetMasterNode()]
7394
    return (nl, nl)
7395

    
7396
  def CheckPrereq(self):
7397
    pass
7398

    
7399
  def Exec(self, feedback_fn):
7400
    # Prepare jobs for migration instances
7401
    jobs = [
7402
      [opcodes.OpInstanceMigrate(instance_name=inst.name,
7403
                                 mode=self.op.mode,
7404
                                 live=self.op.live,
7405
                                 iallocator=self.op.iallocator,
7406
                                 target_node=self.op.target_node)]
7407
      for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)
7408
      ]
7409

    
7410
    # TODO: Run iallocator in this opcode and pass correct placement options to
7411
    # OpInstanceMigrate. Since other jobs can modify the cluster between
7412
    # running the iallocator and the actual migration, a good consistency model
7413
    # will have to be found.
7414

    
7415
    assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
7416
            frozenset([self.op.node_name]))
7417

    
7418
    return ResultWithJobs(jobs)
7419

    
7420

    
7421
class TLMigrateInstance(Tasklet):
7422
  """Tasklet class for instance migration.
7423

7424
  @type live: boolean
7425
  @ivar live: whether the migration will be done live or non-live;
7426
      this variable is initalized only after CheckPrereq has run
7427
  @type cleanup: boolean
7428
  @ivar cleanup: Wheater we cleanup from a failed migration
7429
  @type iallocator: string
7430
  @ivar iallocator: The iallocator used to determine target_node
7431
  @type target_node: string
7432
  @ivar target_node: If given, the target_node to reallocate the instance to
7433
  @type failover: boolean
7434
  @ivar failover: Whether operation results in failover or migration
7435
  @type fallback: boolean
7436
  @ivar fallback: Whether fallback to failover is allowed if migration not
7437
                  possible
7438
  @type ignore_consistency: boolean
7439
  @ivar ignore_consistency: Wheter we should ignore consistency between source
7440
                            and target node
7441
  @type shutdown_timeout: int
7442
  @ivar shutdown_timeout: In case of failover timeout of the shutdown
7443

7444
  """
7445

    
7446
  # Constants
7447
  _MIGRATION_POLL_INTERVAL = 1      # seconds
7448
  _MIGRATION_FEEDBACK_INTERVAL = 10 # seconds
7449

    
7450
  def __init__(self, lu, instance_name, cleanup=False,
7451
               failover=False, fallback=False,
7452
               ignore_consistency=False,
7453
               shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT):
7454
    """Initializes this class.
7455

7456
    """
7457
    Tasklet.__init__(self, lu)
7458

    
7459
    # Parameters
7460
    self.instance_name = instance_name
7461
    self.cleanup = cleanup
7462
    self.live = False # will be overridden later
7463
    self.failover = failover
7464
    self.fallback = fallback
7465
    self.ignore_consistency = ignore_consistency
7466
    self.shutdown_timeout = shutdown_timeout
7467

    
7468
  def CheckPrereq(self):
7469
    """Check prerequisites.
7470

7471
    This checks that the instance is in the cluster.
7472

7473
    """
7474
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
7475
    instance = self.cfg.GetInstanceInfo(instance_name)
7476
    assert instance is not None
7477
    self.instance = instance
7478

    
7479
    if (not self.cleanup and
7480
        not instance.admin_state == constants.ADMINST_UP and
7481
        not self.failover and self.fallback):
7482
      self.lu.LogInfo("Instance is marked down or offline, fallback allowed,"
7483
                      " switching to failover")
7484
      self.failover = True
7485

    
7486
    if instance.disk_template not in constants.DTS_MIRRORED:
7487
      if self.failover:
7488
        text = "failovers"
7489
      else:
7490
        text = "migrations"
7491
      raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
7492
                                 " %s" % (instance.disk_template, text),
7493
                                 errors.ECODE_STATE)
7494

    
7495
    if instance.disk_template in constants.DTS_EXT_MIRROR:
7496
      _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
7497

    
7498
      if self.lu.op.iallocator:
7499
        self._RunAllocator()
7500
      else:
7501
        # We set set self.target_node as it is required by
7502
        # BuildHooksEnv
7503
        self.target_node = self.lu.op.target_node
7504

    
7505
      # self.target_node is already populated, either directly or by the
7506
      # iallocator run
7507
      target_node = self.target_node
7508
      if self.target_node == instance.primary_node:
7509
        raise errors.OpPrereqError("Cannot migrate instance %s"
7510
                                   " to its primary (%s)" %
7511
                                   (instance.name, instance.primary_node))
7512

    
7513
      if len(self.lu.tasklets) == 1:
7514
        # It is safe to release locks only when we're the only tasklet
7515
        # in the LU
7516
        _ReleaseLocks(self.lu, locking.LEVEL_NODE,
7517
                      keep=[instance.primary_node, self.target_node])
7518

    
7519
    else:
7520
      secondary_nodes = instance.secondary_nodes
7521
      if not secondary_nodes:
7522
        raise errors.ConfigurationError("No secondary node but using"
7523
                                        " %s disk template" %
7524
                                        instance.disk_template)
7525
      target_node = secondary_nodes[0]
7526
      if self.lu.op.iallocator or (self.lu.op.target_node and
7527
                                   self.lu.op.target_node != target_node):
7528
        if self.failover:
7529
          text = "failed over"
7530
        else:
7531
          text = "migrated"
7532
        raise errors.OpPrereqError("Instances with disk template %s cannot"
7533
                                   " be %s to arbitrary nodes"
7534
                                   " (neither an iallocator nor a target"
7535
                                   " node can be passed)" %
7536
                                   (instance.disk_template, text),
7537
                                   errors.ECODE_INVAL)
7538

    
7539
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
7540

    
7541
    # check memory requirements on the secondary node
7542
    if not self.failover or instance.admin_state == constants.ADMINST_UP:
7543
      _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
7544
                           instance.name, i_be[constants.BE_MAXMEM],
7545
                           instance.hypervisor)
7546
    else:
7547
      self.lu.LogInfo("Not checking memory on the secondary node as"
7548
                      " instance will not be started")
7549

    
7550
    # check if failover must be forced instead of migration
7551
    if (not self.cleanup and not self.failover and
7552
        i_be[constants.BE_ALWAYS_FAILOVER]):
7553
      if self.fallback:
7554
        self.lu.LogInfo("Instance configured to always failover; fallback"
7555
                        " to failover")
7556
        self.failover = True
7557
      else:
7558
        raise errors.OpPrereqError("This instance has been configured to"
7559
                                   " always failover, please allow failover",
7560
                                   errors.ECODE_STATE)
7561

    
7562
    # check bridge existance
7563
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
7564

    
7565
    if not self.cleanup:
7566
      _CheckNodeNotDrained(self.lu, target_node)
7567
      if not self.failover:
7568
        result = self.rpc.call_instance_migratable(instance.primary_node,
7569
                                                   instance)
7570
        if result.fail_msg and self.fallback:
7571
          self.lu.LogInfo("Can't migrate, instance offline, fallback to"
7572
                          " failover")
7573
          self.failover = True
7574
        else:
7575
          result.Raise("Can't migrate, please use failover",
7576
                       prereq=True, ecode=errors.ECODE_STATE)
7577

    
7578
    assert not (self.failover and self.cleanup)
7579

    
7580
    if not self.failover:
7581
      if self.lu.op.live is not None and self.lu.op.mode is not None:
7582
        raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
7583
                                   " parameters are accepted",
7584
                                   errors.ECODE_INVAL)
7585
      if self.lu.op.live is not None:
7586
        if self.lu.op.live:
7587
          self.lu.op.mode = constants.HT_MIGRATION_LIVE
7588
        else:
7589
          self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
7590
        # reset the 'live' parameter to None so that repeated
7591
        # invocations of CheckPrereq do not raise an exception
7592
        self.lu.op.live = None
7593
      elif self.lu.op.mode is None:
7594
        # read the default value from the hypervisor
7595
        i_hv = self.cfg.GetClusterInfo().FillHV(self.instance,
7596
                                                skip_globals=False)
7597
        self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
7598

    
7599
      self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
7600
    else:
7601
      # Failover is never live
7602
      self.live = False
7603

    
7604
  def _RunAllocator(self):
7605
    """Run the allocator based on input opcode.
7606

7607
    """
7608
    ial = IAllocator(self.cfg, self.rpc,
7609
                     mode=constants.IALLOCATOR_MODE_RELOC,
7610
                     name=self.instance_name,
7611
                     # TODO See why hail breaks with a single node below
7612
                     relocate_from=[self.instance.primary_node,
7613
                                    self.instance.primary_node],
7614
                     )
7615

    
7616
    ial.Run(self.lu.op.iallocator)
7617

    
7618
    if not ial.success:
7619
      raise errors.OpPrereqError("Can't compute nodes using"
7620
                                 " iallocator '%s': %s" %
7621
                                 (self.lu.op.iallocator, ial.info),
7622
                                 errors.ECODE_NORES)
7623
    if len(ial.result) != ial.required_nodes:
7624
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
7625
                                 " of nodes (%s), required %s" %
7626
                                 (self.lu.op.iallocator, len(ial.result),
7627
                                  ial.required_nodes), errors.ECODE_FAULT)
7628
    self.target_node = ial.result[0]
7629
    self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
7630
                 self.instance_name, self.lu.op.iallocator,
7631
                 utils.CommaJoin(ial.result))
7632

    
7633
  def _WaitUntilSync(self):
7634
    """Poll with custom rpc for disk sync.
7635

7636
    This uses our own step-based rpc call.
7637

7638
    """
7639
    self.feedback_fn("* wait until resync is done")
7640
    all_done = False
7641
    while not all_done:
7642
      all_done = True
7643
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
7644
                                            self.nodes_ip,
7645
                                            self.instance.disks)
7646
      min_percent = 100
7647
      for node, nres in result.items():
7648
        nres.Raise("Cannot resync disks on node %s" % node)
7649
        node_done, node_percent = nres.payload
7650
        all_done = all_done and node_done
7651
        if node_percent is not None:
7652
          min_percent = min(min_percent, node_percent)
7653
      if not all_done:
7654
        if min_percent < 100:
7655
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
7656
        time.sleep(2)
7657

    
7658
  def _EnsureSecondary(self, node):
7659
    """Demote a node to secondary.
7660

7661
    """
7662
    self.feedback_fn("* switching node %s to secondary mode" % node)
7663

    
7664
    for dev in self.instance.disks:
7665
      self.cfg.SetDiskID(dev, node)
7666

    
7667
    result = self.rpc.call_blockdev_close(node, self.instance.name,
7668
                                          self.instance.disks)
7669
    result.Raise("Cannot change disk to secondary on node %s" % node)
7670

    
7671
  def _GoStandalone(self):
7672
    """Disconnect from the network.
7673

7674
    """
7675
    self.feedback_fn("* changing into standalone mode")
7676
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
7677
                                               self.instance.disks)
7678
    for node, nres in result.items():
7679
      nres.Raise("Cannot disconnect disks node %s" % node)
7680

    
7681
  def _GoReconnect(self, multimaster):
7682
    """Reconnect to the network.
7683

7684
    """
7685
    if multimaster:
7686
      msg = "dual-master"
7687
    else:
7688
      msg = "single-master"
7689
    self.feedback_fn("* changing disks into %s mode" % msg)
7690
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
7691
                                           self.instance.disks,
7692
                                           self.instance.name, multimaster)
7693
    for node, nres in result.items():
7694
      nres.Raise("Cannot change disks config on node %s" % node)
7695

    
7696
  def _ExecCleanup(self):
7697
    """Try to cleanup after a failed migration.
7698

7699
    The cleanup is done by:
7700
      - check that the instance is running only on one node
7701
        (and update the config if needed)
7702
      - change disks on its secondary node to secondary
7703
      - wait until disks are fully synchronized
7704
      - disconnect from the network
7705
      - change disks into single-master mode
7706
      - wait again until disks are fully synchronized
7707

7708
    """
7709
    instance = self.instance
7710
    target_node = self.target_node
7711
    source_node = self.source_node
7712

    
7713
    # check running on only one node
7714
    self.feedback_fn("* checking where the instance actually runs"
7715
                     " (if this hangs, the hypervisor might be in"
7716
                     " a bad state)")
7717
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
7718
    for node, result in ins_l.items():
7719
      result.Raise("Can't contact node %s" % node)
7720

    
7721
    runningon_source = instance.name in ins_l[source_node].payload
7722
    runningon_target = instance.name in ins_l[target_node].payload
7723

    
7724
    if runningon_source and runningon_target:
7725
      raise errors.OpExecError("Instance seems to be running on two nodes,"
7726
                               " or the hypervisor is confused; you will have"
7727
                               " to ensure manually that it runs only on one"
7728
                               " and restart this operation")
7729

    
7730
    if not (runningon_source or runningon_target):
7731
      raise errors.OpExecError("Instance does not seem to be running at all;"
7732
                               " in this case it's safer to repair by"
7733
                               " running 'gnt-instance stop' to ensure disk"
7734
                               " shutdown, and then restarting it")
7735

    
7736
    if runningon_target:
7737
      # the migration has actually succeeded, we need to update the config
7738
      self.feedback_fn("* instance running on secondary node (%s),"
7739
                       " updating config" % target_node)
7740
      instance.primary_node = target_node
7741
      self.cfg.Update(instance, self.feedback_fn)
7742
      demoted_node = source_node
7743
    else:
7744
      self.feedback_fn("* instance confirmed to be running on its"
7745
                       " primary node (%s)" % source_node)
7746
      demoted_node = target_node
7747

    
7748
    if instance.disk_template in constants.DTS_INT_MIRROR:
7749
      self._EnsureSecondary(demoted_node)
7750
      try:
7751
        self._WaitUntilSync()
7752
      except errors.OpExecError:
7753
        # we ignore here errors, since if the device is standalone, it
7754
        # won't be able to sync
7755
        pass
7756
      self._GoStandalone()
7757
      self._GoReconnect(False)
7758
      self._WaitUntilSync()
7759

    
7760
    self.feedback_fn("* done")
7761

    
7762
  def _RevertDiskStatus(self):
7763
    """Try to revert the disk status after a failed migration.
7764

7765
    """
7766
    target_node = self.target_node
7767
    if self.instance.disk_template in constants.DTS_EXT_MIRROR:
7768
      return
7769

    
7770
    try:
7771
      self._EnsureSecondary(target_node)
7772
      self._GoStandalone()
7773
      self._GoReconnect(False)
7774
      self._WaitUntilSync()
7775
    except errors.OpExecError, err:
7776
      self.lu.LogWarning("Migration failed and I can't reconnect the drives,"
7777
                         " please try to recover the instance manually;"
7778
                         " error '%s'" % str(err))
7779

    
7780
  def _AbortMigration(self):
7781
    """Call the hypervisor code to abort a started migration.
7782

7783
    """
7784
    instance = self.instance
7785
    target_node = self.target_node
7786
    source_node = self.source_node
7787
    migration_info = self.migration_info
7788

    
7789
    abort_result = self.rpc.call_instance_finalize_migration_dst(target_node,
7790
                                                                 instance,
7791
                                                                 migration_info,
7792
                                                                 False)
7793
    abort_msg = abort_result.fail_msg
7794
    if abort_msg:
7795
      logging.error("Aborting migration failed on target node %s: %s",
7796
                    target_node, abort_msg)
7797
      # Don't raise an exception here, as we stil have to try to revert the
7798
      # disk status, even if this step failed.
7799

    
7800
    abort_result = self.rpc.call_instance_finalize_migration_src(source_node,
7801
        instance, False, self.live)
7802
    abort_msg = abort_result.fail_msg
7803
    if abort_msg:
7804
      logging.error("Aborting migration failed on source node %s: %s",
7805
                    source_node, abort_msg)
7806

    
7807
  def _ExecMigration(self):
7808
    """Migrate an instance.
7809

7810
    The migrate is done by:
7811
      - change the disks into dual-master mode
7812
      - wait until disks are fully synchronized again
7813
      - migrate the instance
7814
      - change disks on the new secondary node (the old primary) to secondary
7815
      - wait until disks are fully synchronized
7816
      - change disks into single-master mode
7817

7818
    """
7819
    instance = self.instance
7820
    target_node = self.target_node
7821
    source_node = self.source_node
7822

    
7823
    # Check for hypervisor version mismatch and warn the user.
7824
    nodeinfo = self.rpc.call_node_info([source_node, target_node],
7825
                                       None, [self.instance.hypervisor])
7826
    for ninfo in nodeinfo.values():
7827
      ninfo.Raise("Unable to retrieve node information from node '%s'" %
7828
                  ninfo.node)
7829
    (_, _, (src_info, )) = nodeinfo[source_node].payload
7830
    (_, _, (dst_info, )) = nodeinfo[target_node].payload
7831

    
7832
    if ((constants.HV_NODEINFO_KEY_VERSION in src_info) and
7833
        (constants.HV_NODEINFO_KEY_VERSION in dst_info)):
7834
      src_version = src_info[constants.HV_NODEINFO_KEY_VERSION]
7835
      dst_version = dst_info[constants.HV_NODEINFO_KEY_VERSION]
7836
      if src_version != dst_version:
7837
        self.feedback_fn("* warning: hypervisor version mismatch between"
7838
                         " source (%s) and target (%s) node" %
7839
                         (src_version, dst_version))
7840

    
7841
    self.feedback_fn("* checking disk consistency between source and target")
7842
    for dev in instance.disks:
7843
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
7844
        raise errors.OpExecError("Disk %s is degraded or not fully"
7845
                                 " synchronized on target node,"
7846
                                 " aborting migration" % dev.iv_name)
7847

    
7848
    # First get the migration information from the remote node
7849
    result = self.rpc.call_migration_info(source_node, instance)
7850
    msg = result.fail_msg
7851
    if msg:
7852
      log_err = ("Failed fetching source migration information from %s: %s" %
7853
                 (source_node, msg))
7854
      logging.error(log_err)
7855
      raise errors.OpExecError(log_err)
7856

    
7857
    self.migration_info = migration_info = result.payload
7858

    
7859
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
7860
      # Then switch the disks to master/master mode
7861
      self._EnsureSecondary(target_node)
7862
      self._GoStandalone()
7863
      self._GoReconnect(True)
7864
      self._WaitUntilSync()
7865

    
7866
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
7867
    result = self.rpc.call_accept_instance(target_node,
7868
                                           instance,
7869
                                           migration_info,
7870
                                           self.nodes_ip[target_node])
7871

    
7872
    msg = result.fail_msg
7873
    if msg:
7874
      logging.error("Instance pre-migration failed, trying to revert"
7875
                    " disk status: %s", msg)
7876
      self.feedback_fn("Pre-migration failed, aborting")
7877
      self._AbortMigration()
7878
      self._RevertDiskStatus()
7879
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
7880
                               (instance.name, msg))
7881

    
7882
    self.feedback_fn("* migrating instance to %s" % target_node)
7883
    result = self.rpc.call_instance_migrate(source_node, instance,
7884
                                            self.nodes_ip[target_node],
7885
                                            self.live)
7886
    msg = result.fail_msg
7887
    if msg:
7888
      logging.error("Instance migration failed, trying to revert"
7889
                    " disk status: %s", msg)
7890
      self.feedback_fn("Migration failed, aborting")
7891
      self._AbortMigration()
7892
      self._RevertDiskStatus()
7893
      raise errors.OpExecError("Could not migrate instance %s: %s" %
7894
                               (instance.name, msg))
7895

    
7896
    self.feedback_fn("* starting memory transfer")
7897
    last_feedback = time.time()
7898
    while True:
7899
      result = self.rpc.call_instance_get_migration_status(source_node,
7900
                                                           instance)
7901
      msg = result.fail_msg
7902
      ms = result.payload   # MigrationStatus instance
7903
      if msg or (ms.status in constants.HV_MIGRATION_FAILED_STATUSES):
7904
        logging.error("Instance migration failed, trying to revert"
7905
                      " disk status: %s", msg)
7906
        self.feedback_fn("Migration failed, aborting")
7907
        self._AbortMigration()
7908
        self._RevertDiskStatus()
7909
        raise errors.OpExecError("Could not migrate instance %s: %s" %
7910
                                 (instance.name, msg))
7911

    
7912
      if result.payload.status != constants.HV_MIGRATION_ACTIVE:
7913
        self.feedback_fn("* memory transfer complete")
7914
        break
7915

    
7916
      if (utils.TimeoutExpired(last_feedback,
7917
                               self._MIGRATION_FEEDBACK_INTERVAL) and
7918
          ms.transferred_ram is not None):
7919
        mem_progress = 100 * float(ms.transferred_ram) / float(ms.total_ram)
7920
        self.feedback_fn("* memory transfer progress: %.2f %%" % mem_progress)
7921
        last_feedback = time.time()
7922

    
7923
      time.sleep(self._MIGRATION_POLL_INTERVAL)
7924

    
7925
    result = self.rpc.call_instance_finalize_migration_src(source_node,
7926
                                                           instance,
7927
                                                           True,
7928
                                                           self.live)
7929
    msg = result.fail_msg
7930
    if msg:
7931
      logging.error("Instance migration succeeded, but finalization failed"
7932
                    " on the source node: %s", msg)
7933
      raise errors.OpExecError("Could not finalize instance migration: %s" %
7934
                               msg)
7935

    
7936
    instance.primary_node = target_node
7937

    
7938
    # distribute new instance config to the other nodes
7939
    self.cfg.Update(instance, self.feedback_fn)
7940

    
7941
    result = self.rpc.call_instance_finalize_migration_dst(target_node,
7942
                                                           instance,
7943
                                                           migration_info,
7944
                                                           True)
7945
    msg = result.fail_msg
7946
    if msg:
7947
      logging.error("Instance migration succeeded, but finalization failed"
7948
                    " on the target node: %s", msg)
7949
      raise errors.OpExecError("Could not finalize instance migration: %s" %
7950
                               msg)
7951

    
7952
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
7953
      self._EnsureSecondary(source_node)
7954
      self._WaitUntilSync()
7955
      self._GoStandalone()
7956
      self._GoReconnect(False)
7957
      self._WaitUntilSync()
7958

    
7959
    self.feedback_fn("* done")
7960

    
7961
  def _ExecFailover(self):
7962
    """Failover an instance.
7963

7964
    The failover is done by shutting it down on its present node and
7965
    starting it on the secondary.
7966

7967
    """
7968
    instance = self.instance
7969
    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
7970

    
7971
    source_node = instance.primary_node
7972
    target_node = self.target_node
7973

    
7974
    if instance.admin_state == constants.ADMINST_UP:
7975
      self.feedback_fn("* checking disk consistency between source and target")
7976
      for dev in instance.disks:
7977
        # for drbd, these are drbd over lvm
7978
        if not _CheckDiskConsistency(self.lu, dev, target_node, False):
7979
          if primary_node.offline:
7980
            self.feedback_fn("Node %s is offline, ignoring degraded disk %s on"
7981
                             " target node %s" %
7982
                             (primary_node.name, dev.iv_name, target_node))
7983
          elif not self.ignore_consistency:
7984
            raise errors.OpExecError("Disk %s is degraded on target node,"
7985
                                     " aborting failover" % dev.iv_name)
7986
    else:
7987
      self.feedback_fn("* not checking disk consistency as instance is not"
7988
                       " running")
7989

    
7990
    self.feedback_fn("* shutting down instance on source node")
7991
    logging.info("Shutting down instance %s on node %s",
7992
                 instance.name, source_node)
7993

    
7994
    result = self.rpc.call_instance_shutdown(source_node, instance,
7995
                                             self.shutdown_timeout)
7996
    msg = result.fail_msg
7997
    if msg:
7998
      if self.ignore_consistency or primary_node.offline:
7999
        self.lu.LogWarning("Could not shutdown instance %s on node %s,"
8000
                           " proceeding anyway; please make sure node"
8001
                           " %s is down; error details: %s",
8002
                           instance.name, source_node, source_node, msg)
8003
      else:
8004
        raise errors.OpExecError("Could not shutdown instance %s on"
8005
                                 " node %s: %s" %
8006
                                 (instance.name, source_node, msg))
8007

    
8008
    self.feedback_fn("* deactivating the instance's disks on source node")
8009
    if not _ShutdownInstanceDisks(self.lu, instance, ignore_primary=True):
8010
      raise errors.OpExecError("Can't shut down the instance's disks")
8011

    
8012
    instance.primary_node = target_node
8013
    # distribute new instance config to the other nodes
8014
    self.cfg.Update(instance, self.feedback_fn)
8015

    
8016
    # Only start the instance if it's marked as up
8017
    if instance.admin_state == constants.ADMINST_UP:
8018
      self.feedback_fn("* activating the instance's disks on target node %s" %
8019
                       target_node)
8020
      logging.info("Starting instance %s on node %s",
8021
                   instance.name, target_node)
8022

    
8023
      disks_ok, _ = _AssembleInstanceDisks(self.lu, instance,
8024
                                           ignore_secondaries=True)
8025
      if not disks_ok:
8026
        _ShutdownInstanceDisks(self.lu, instance)
8027
        raise errors.OpExecError("Can't activate the instance's disks")
8028

    
8029
      self.feedback_fn("* starting the instance on the target node %s" %
8030
                       target_node)
8031
      result = self.rpc.call_instance_start(target_node, (instance, None, None),
8032
                                            False)
8033
      msg = result.fail_msg
8034
      if msg:
8035
        _ShutdownInstanceDisks(self.lu, instance)
8036
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
8037
                                 (instance.name, target_node, msg))
8038

    
8039
  def Exec(self, feedback_fn):
8040
    """Perform the migration.
8041

8042
    """
8043
    self.feedback_fn = feedback_fn
8044
    self.source_node = self.instance.primary_node
8045

    
8046
    # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
8047
    if self.instance.disk_template in constants.DTS_INT_MIRROR:
8048
      self.target_node = self.instance.secondary_nodes[0]
8049
      # Otherwise self.target_node has been populated either
8050
      # directly, or through an iallocator.
8051

    
8052
    self.all_nodes = [self.source_node, self.target_node]
8053
    self.nodes_ip = dict((name, node.secondary_ip) for (name, node)
8054
                         in self.cfg.GetMultiNodeInfo(self.all_nodes))
8055

    
8056
    if self.failover:
8057
      feedback_fn("Failover instance %s" % self.instance.name)
8058
      self._ExecFailover()
8059
    else:
8060
      feedback_fn("Migrating instance %s" % self.instance.name)
8061

    
8062
      if self.cleanup:
8063
        return self._ExecCleanup()
8064
      else:
8065
        return self._ExecMigration()
8066

    
8067

    
8068
def _CreateBlockDev(lu, node, instance, device, force_create,
8069
                    info, force_open):
8070
  """Create a tree of block devices on a given node.
8071

8072
  If this device type has to be created on secondaries, create it and
8073
  all its children.
8074

8075
  If not, just recurse to children keeping the same 'force' value.
8076

8077
  @param lu: the lu on whose behalf we execute
8078
  @param node: the node on which to create the device
8079
  @type instance: L{objects.Instance}
8080
  @param instance: the instance which owns the device
8081
  @type device: L{objects.Disk}
8082
  @param device: the device to create
8083
  @type force_create: boolean
8084
  @param force_create: whether to force creation of this device; this
8085
      will be change to True whenever we find a device which has
8086
      CreateOnSecondary() attribute
8087
  @param info: the extra 'metadata' we should attach to the device
8088
      (this will be represented as a LVM tag)
8089
  @type force_open: boolean
8090
  @param force_open: this parameter will be passes to the
8091
      L{backend.BlockdevCreate} function where it specifies
8092
      whether we run on primary or not, and it affects both
8093
      the child assembly and the device own Open() execution
8094

8095
  """
8096
  if device.CreateOnSecondary():
8097
    force_create = True
8098

    
8099
  if device.children:
8100
    for child in device.children:
8101
      _CreateBlockDev(lu, node, instance, child, force_create,
8102
                      info, force_open)
8103

    
8104
  if not force_create:
8105
    return
8106

    
8107
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
8108

    
8109

    
8110
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
8111
  """Create a single block device on a given node.
8112

8113
  This will not recurse over children of the device, so they must be
8114
  created in advance.
8115

8116
  @param lu: the lu on whose behalf we execute
8117
  @param node: the node on which to create the device
8118
  @type instance: L{objects.Instance}
8119
  @param instance: the instance which owns the device
8120
  @type device: L{objects.Disk}
8121
  @param device: the device to create
8122
  @param info: the extra 'metadata' we should attach to the device
8123
      (this will be represented as a LVM tag)
8124
  @type force_open: boolean
8125
  @param force_open: this parameter will be passes to the
8126
      L{backend.BlockdevCreate} function where it specifies
8127
      whether we run on primary or not, and it affects both
8128
      the child assembly and the device own Open() execution
8129

8130
  """
8131
  lu.cfg.SetDiskID(device, node)
8132
  result = lu.rpc.call_blockdev_create(node, device, device.size,
8133
                                       instance.name, force_open, info)
8134
  result.Raise("Can't create block device %s on"
8135
               " node %s for instance %s" % (device, node, instance.name))
8136
  if device.physical_id is None:
8137
    device.physical_id = result.payload
8138

    
8139

    
8140
def _GenerateUniqueNames(lu, exts):
8141
  """Generate a suitable LV name.
8142

8143
  This will generate a logical volume name for the given instance.
8144

8145
  """
8146
  results = []
8147
  for val in exts:
8148
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
8149
    results.append("%s%s" % (new_id, val))
8150
  return results
8151

    
8152

    
8153
def _ComputeLDParams(disk_template, disk_params):
8154
  """Computes Logical Disk parameters from Disk Template parameters.
8155

8156
  @type disk_template: string
8157
  @param disk_template: disk template, one of L{constants.DISK_TEMPLATES}
8158
  @type disk_params: dict
8159
  @param disk_params: disk template parameters; dict(template_name -> parameters
8160
  @rtype: list(dict)
8161
  @return: a list of dicts, one for each node of the disk hierarchy. Each dict
8162
    contains the LD parameters of the node. The tree is flattened in-order.
8163

8164
  """
8165
  if disk_template not in constants.DISK_TEMPLATES:
8166
    raise errors.ProgrammerError("Unknown disk template %s" % disk_template)
8167

    
8168
  result = list()
8169
  dt_params = disk_params[disk_template]
8170
  if disk_template == constants.DT_DRBD8:
8171
    drbd_params = {
8172
      constants.RESYNC_RATE: dt_params[constants.DRBD_RESYNC_RATE],
8173
      constants.BARRIERS: dt_params[constants.DRBD_DISK_BARRIERS],
8174
      constants.NO_META_FLUSH: dt_params[constants.DRBD_META_BARRIERS],
8175
      }
8176

    
8177
    drbd_params = \
8178
      objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_DRBD8],
8179
                       drbd_params)
8180

    
8181
    result.append(drbd_params)
8182

    
8183
    # data LV
8184
    data_params = {
8185
      constants.STRIPES: dt_params[constants.DRBD_DATA_STRIPES],
8186
      }
8187
    data_params = \
8188
      objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
8189
                       data_params)
8190
    result.append(data_params)
8191

    
8192
    # metadata LV
8193
    meta_params = {
8194
      constants.STRIPES: dt_params[constants.DRBD_META_STRIPES],
8195
      }
8196
    meta_params = \
8197
      objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
8198
                       meta_params)
8199
    result.append(meta_params)
8200

    
8201
  elif (disk_template == constants.DT_FILE or
8202
        disk_template == constants.DT_SHARED_FILE):
8203
    result.append(constants.DISK_LD_DEFAULTS[constants.LD_FILE])
8204

    
8205
  elif disk_template == constants.DT_PLAIN:
8206
    params = {
8207
      constants.STRIPES: dt_params[constants.LV_STRIPES],
8208
      }
8209
    params = \
8210
      objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
8211
                       params)
8212
    result.append(params)
8213

    
8214
  elif disk_template == constants.DT_BLOCK:
8215
    result.append(constants.DISK_LD_DEFAULTS[constants.LD_BLOCKDEV])
8216

    
8217
  return result
8218

    
8219

    
8220
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
8221
                         iv_name, p_minor, s_minor, drbd_params, data_params,
8222
                         meta_params):
8223
  """Generate a drbd8 device complete with its children.
8224

8225
  """
8226
  assert len(vgnames) == len(names) == 2
8227
  port = lu.cfg.AllocatePort()
8228
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
8229

    
8230
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
8231
                          logical_id=(vgnames[0], names[0]),
8232
                          params=data_params)
8233
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
8234
                          logical_id=(vgnames[1], names[1]),
8235
                          params=meta_params)
8236
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
8237
                          logical_id=(primary, secondary, port,
8238
                                      p_minor, s_minor,
8239
                                      shared_secret),
8240
                          children=[dev_data, dev_meta],
8241
                          iv_name=iv_name, params=drbd_params)
8242
  return drbd_dev
8243

    
8244

    
8245
def _GenerateDiskTemplate(lu, template_name,
8246
                          instance_name, primary_node,
8247
                          secondary_nodes, disk_info,
8248
                          file_storage_dir, file_driver,
8249
                          base_index, feedback_fn, disk_params):
8250
  """Generate the entire disk layout for a given template type.
8251

8252
  """
8253
  #TODO: compute space requirements
8254

    
8255
  vgname = lu.cfg.GetVGName()
8256
  disk_count = len(disk_info)
8257
  disks = []
8258
  ld_params = _ComputeLDParams(template_name, disk_params)
8259
  if template_name == constants.DT_DISKLESS:
8260
    pass
8261
  elif template_name == constants.DT_PLAIN:
8262
    if len(secondary_nodes) != 0:
8263
      raise errors.ProgrammerError("Wrong template configuration")
8264

    
8265
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
8266
                                      for i in range(disk_count)])
8267
    for idx, disk in enumerate(disk_info):
8268
      disk_index = idx + base_index
8269
      vg = disk.get(constants.IDISK_VG, vgname)
8270
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
8271
      disk_dev = objects.Disk(dev_type=constants.LD_LV,
8272
                              size=disk[constants.IDISK_SIZE],
8273
                              logical_id=(vg, names[idx]),
8274
                              iv_name="disk/%d" % disk_index,
8275
                              mode=disk[constants.IDISK_MODE],
8276
                              params=ld_params[0])
8277
      disks.append(disk_dev)
8278
  elif template_name == constants.DT_DRBD8:
8279
    drbd_params, data_params, meta_params = ld_params
8280
    if len(secondary_nodes) != 1:
8281
      raise errors.ProgrammerError("Wrong template configuration")
8282
    remote_node = secondary_nodes[0]
8283
    minors = lu.cfg.AllocateDRBDMinor(
8284
      [primary_node, remote_node] * len(disk_info), instance_name)
8285

    
8286
    names = []
8287
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
8288
                                               for i in range(disk_count)]):
8289
      names.append(lv_prefix + "_data")
8290
      names.append(lv_prefix + "_meta")
8291
    for idx, disk in enumerate(disk_info):
8292
      disk_index = idx + base_index
8293
      data_vg = disk.get(constants.IDISK_VG, vgname)
8294
      meta_vg = disk.get(constants.IDISK_METAVG, data_vg)
8295
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
8296
                                      disk[constants.IDISK_SIZE],
8297
                                      [data_vg, meta_vg],
8298
                                      names[idx * 2:idx * 2 + 2],
8299
                                      "disk/%d" % disk_index,
8300
                                      minors[idx * 2], minors[idx * 2 + 1],
8301
                                      drbd_params, data_params, meta_params)
8302
      disk_dev.mode = disk[constants.IDISK_MODE]
8303
      disks.append(disk_dev)
8304
  elif template_name == constants.DT_FILE:
8305
    if len(secondary_nodes) != 0:
8306
      raise errors.ProgrammerError("Wrong template configuration")
8307

    
8308
    opcodes.RequireFileStorage()
8309

    
8310
    for idx, disk in enumerate(disk_info):
8311
      disk_index = idx + base_index
8312
      disk_dev = objects.Disk(dev_type=constants.LD_FILE,
8313
                              size=disk[constants.IDISK_SIZE],
8314
                              iv_name="disk/%d" % disk_index,
8315
                              logical_id=(file_driver,
8316
                                          "%s/disk%d" % (file_storage_dir,
8317
                                                         disk_index)),
8318
                              mode=disk[constants.IDISK_MODE],
8319
                              params=ld_params[0])
8320
      disks.append(disk_dev)
8321
  elif template_name == constants.DT_SHARED_FILE:
8322
    if len(secondary_nodes) != 0:
8323
      raise errors.ProgrammerError("Wrong template configuration")
8324

    
8325
    opcodes.RequireSharedFileStorage()
8326

    
8327
    for idx, disk in enumerate(disk_info):
8328
      disk_index = idx + base_index
8329
      disk_dev = objects.Disk(dev_type=constants.LD_FILE,
8330
                              size=disk[constants.IDISK_SIZE],
8331
                              iv_name="disk/%d" % disk_index,
8332
                              logical_id=(file_driver,
8333
                                          "%s/disk%d" % (file_storage_dir,
8334
                                                         disk_index)),
8335
                              mode=disk[constants.IDISK_MODE],
8336
                              params=ld_params[0])
8337
      disks.append(disk_dev)
8338
  elif template_name == constants.DT_BLOCK:
8339
    if len(secondary_nodes) != 0:
8340
      raise errors.ProgrammerError("Wrong template configuration")
8341

    
8342
    for idx, disk in enumerate(disk_info):
8343
      disk_index = idx + base_index
8344
      disk_dev = objects.Disk(dev_type=constants.LD_BLOCKDEV,
8345
                              size=disk[constants.IDISK_SIZE],
8346
                              logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
8347
                                          disk[constants.IDISK_ADOPT]),
8348
                              iv_name="disk/%d" % disk_index,
8349
                              mode=disk[constants.IDISK_MODE],
8350
                              params=ld_params[0])
8351
      disks.append(disk_dev)
8352

    
8353
  else:
8354
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
8355
  return disks
8356

    
8357

    
8358
def _GetInstanceInfoText(instance):
8359
  """Compute that text that should be added to the disk's metadata.
8360

8361
  """
8362
  return "originstname+%s" % instance.name
8363

    
8364

    
8365
def _CalcEta(time_taken, written, total_size):
8366
  """Calculates the ETA based on size written and total size.
8367

8368
  @param time_taken: The time taken so far
8369
  @param written: amount written so far
8370
  @param total_size: The total size of data to be written
8371
  @return: The remaining time in seconds
8372

8373
  """
8374
  avg_time = time_taken / float(written)
8375
  return (total_size - written) * avg_time
8376

    
8377

    
8378
def _WipeDisks(lu, instance):
8379
  """Wipes instance disks.
8380

8381
  @type lu: L{LogicalUnit}
8382
  @param lu: the logical unit on whose behalf we execute
8383
  @type instance: L{objects.Instance}
8384
  @param instance: the instance whose disks we should create
8385
  @return: the success of the wipe
8386

8387
  """
8388
  node = instance.primary_node
8389

    
8390
  for device in instance.disks:
8391
    lu.cfg.SetDiskID(device, node)
8392

    
8393
  logging.info("Pause sync of instance %s disks", instance.name)
8394
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
8395

    
8396
  for idx, success in enumerate(result.payload):
8397
    if not success:
8398
      logging.warn("pause-sync of instance %s for disks %d failed",
8399
                   instance.name, idx)
8400

    
8401
  try:
8402
    for idx, device in enumerate(instance.disks):
8403
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
8404
      # MAX_WIPE_CHUNK at max
8405
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
8406
                            constants.MIN_WIPE_CHUNK_PERCENT)
8407
      # we _must_ make this an int, otherwise rounding errors will
8408
      # occur
8409
      wipe_chunk_size = int(wipe_chunk_size)
8410

    
8411
      lu.LogInfo("* Wiping disk %d", idx)
8412
      logging.info("Wiping disk %d for instance %s, node %s using"
8413
                   " chunk size %s", idx, instance.name, node, wipe_chunk_size)
8414

    
8415
      offset = 0
8416
      size = device.size
8417
      last_output = 0
8418
      start_time = time.time()
8419

    
8420
      while offset < size:
8421
        wipe_size = min(wipe_chunk_size, size - offset)
8422
        logging.debug("Wiping disk %d, offset %s, chunk %s",
8423
                      idx, offset, wipe_size)
8424
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
8425
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
8426
                     (idx, offset, wipe_size))
8427
        now = time.time()
8428
        offset += wipe_size
8429
        if now - last_output >= 60:
8430
          eta = _CalcEta(now - start_time, offset, size)
8431
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
8432
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
8433
          last_output = now
8434
  finally:
8435
    logging.info("Resume sync of instance %s disks", instance.name)
8436

    
8437
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
8438

    
8439
    for idx, success in enumerate(result.payload):
8440
      if not success:
8441
        lu.LogWarning("Resume sync of disk %d failed, please have a"
8442
                      " look at the status and troubleshoot the issue", idx)
8443
        logging.warn("resume-sync of instance %s for disks %d failed",
8444
                     instance.name, idx)
8445

    
8446

    
8447
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
8448
  """Create all disks for an instance.
8449

8450
  This abstracts away some work from AddInstance.
8451

8452
  @type lu: L{LogicalUnit}
8453
  @param lu: the logical unit on whose behalf we execute
8454
  @type instance: L{objects.Instance}
8455
  @param instance: the instance whose disks we should create
8456
  @type to_skip: list
8457
  @param to_skip: list of indices to skip
8458
  @type target_node: string
8459
  @param target_node: if passed, overrides the target node for creation
8460
  @rtype: boolean
8461
  @return: the success of the creation
8462

8463
  """
8464
  info = _GetInstanceInfoText(instance)
8465
  if target_node is None:
8466
    pnode = instance.primary_node
8467
    all_nodes = instance.all_nodes
8468
  else:
8469
    pnode = target_node
8470
    all_nodes = [pnode]
8471

    
8472
  if instance.disk_template in constants.DTS_FILEBASED:
8473
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
8474
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
8475

    
8476
    result.Raise("Failed to create directory '%s' on"
8477
                 " node %s" % (file_storage_dir, pnode))
8478

    
8479
  # Note: this needs to be kept in sync with adding of disks in
8480
  # LUInstanceSetParams
8481
  for idx, device in enumerate(instance.disks):
8482
    if to_skip and idx in to_skip:
8483
      continue
8484
    logging.info("Creating volume %s for instance %s",
8485
                 device.iv_name, instance.name)
8486
    #HARDCODE
8487
    for node in all_nodes:
8488
      f_create = node == pnode
8489
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
8490

    
8491

    
8492
def _RemoveDisks(lu, instance, target_node=None):
8493
  """Remove all disks for an instance.
8494

8495
  This abstracts away some work from `AddInstance()` and
8496
  `RemoveInstance()`. Note that in case some of the devices couldn't
8497
  be removed, the removal will continue with the other ones (compare
8498
  with `_CreateDisks()`).
8499

8500
  @type lu: L{LogicalUnit}
8501
  @param lu: the logical unit on whose behalf we execute
8502
  @type instance: L{objects.Instance}
8503
  @param instance: the instance whose disks we should remove
8504
  @type target_node: string
8505
  @param target_node: used to override the node on which to remove the disks
8506
  @rtype: boolean
8507
  @return: the success of the removal
8508

8509
  """
8510
  logging.info("Removing block devices for instance %s", instance.name)
8511

    
8512
  all_result = True
8513
  for device in instance.disks:
8514
    if target_node:
8515
      edata = [(target_node, device)]
8516
    else:
8517
      edata = device.ComputeNodeTree(instance.primary_node)
8518
    for node, disk in edata:
8519
      lu.cfg.SetDiskID(disk, node)
8520
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
8521
      if msg:
8522
        lu.LogWarning("Could not remove block device %s on node %s,"
8523
                      " continuing anyway: %s", device.iv_name, node, msg)
8524
        all_result = False
8525

    
8526
    # if this is a DRBD disk, return its port to the pool
8527
    if device.dev_type in constants.LDS_DRBD:
8528
      tcp_port = device.logical_id[2]
8529
      lu.cfg.AddTcpUdpPort(tcp_port)
8530

    
8531
  if instance.disk_template == constants.DT_FILE:
8532
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
8533
    if target_node:
8534
      tgt = target_node
8535
    else:
8536
      tgt = instance.primary_node
8537
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
8538
    if result.fail_msg:
8539
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
8540
                    file_storage_dir, instance.primary_node, result.fail_msg)
8541
      all_result = False
8542

    
8543
  return all_result
8544

    
8545

    
8546
def _ComputeDiskSizePerVG(disk_template, disks):
8547
  """Compute disk size requirements in the volume group
8548

8549
  """
8550
  def _compute(disks, payload):
8551
    """Universal algorithm.
8552

8553
    """
8554
    vgs = {}
8555
    for disk in disks:
8556
      vgs[disk[constants.IDISK_VG]] = \
8557
        vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload
8558

    
8559
    return vgs
8560

    
8561
  # Required free disk space as a function of disk and swap space
8562
  req_size_dict = {
8563
    constants.DT_DISKLESS: {},
8564
    constants.DT_PLAIN: _compute(disks, 0),
8565
    # 128 MB are added for drbd metadata for each disk
8566
    constants.DT_DRBD8: _compute(disks, DRBD_META_SIZE),
8567
    constants.DT_FILE: {},
8568
    constants.DT_SHARED_FILE: {},
8569
  }
8570

    
8571
  if disk_template not in req_size_dict:
8572
    raise errors.ProgrammerError("Disk template '%s' size requirement"
8573
                                 " is unknown" % disk_template)
8574

    
8575
  return req_size_dict[disk_template]
8576

    
8577

    
8578
def _ComputeDiskSize(disk_template, disks):
8579
  """Compute disk size requirements in the volume group
8580

8581
  """
8582
  # Required free disk space as a function of disk and swap space
8583
  req_size_dict = {
8584
    constants.DT_DISKLESS: None,
8585
    constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
8586
    # 128 MB are added for drbd metadata for each disk
8587
    constants.DT_DRBD8:
8588
      sum(d[constants.IDISK_SIZE] + DRBD_META_SIZE for d in disks),
8589
    constants.DT_FILE: None,
8590
    constants.DT_SHARED_FILE: 0,
8591
    constants.DT_BLOCK: 0,
8592
  }
8593

    
8594
  if disk_template not in req_size_dict:
8595
    raise errors.ProgrammerError("Disk template '%s' size requirement"
8596
                                 " is unknown" % disk_template)
8597

    
8598
  return req_size_dict[disk_template]
8599

    
8600

    
8601
def _FilterVmNodes(lu, nodenames):
8602
  """Filters out non-vm_capable nodes from a list.
8603

8604
  @type lu: L{LogicalUnit}
8605
  @param lu: the logical unit for which we check
8606
  @type nodenames: list
8607
  @param nodenames: the list of nodes on which we should check
8608
  @rtype: list
8609
  @return: the list of vm-capable nodes
8610

8611
  """
8612
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
8613
  return [name for name in nodenames if name not in vm_nodes]
8614

    
8615

    
8616
def _CheckHVParams(lu, nodenames, hvname, hvparams):
8617
  """Hypervisor parameter validation.
8618

8619
  This function abstract the hypervisor parameter validation to be
8620
  used in both instance create and instance modify.
8621

8622
  @type lu: L{LogicalUnit}
8623
  @param lu: the logical unit for which we check
8624
  @type nodenames: list
8625
  @param nodenames: the list of nodes on which we should check
8626
  @type hvname: string
8627
  @param hvname: the name of the hypervisor we should use
8628
  @type hvparams: dict
8629
  @param hvparams: the parameters which we need to check
8630
  @raise errors.OpPrereqError: if the parameters are not valid
8631

8632
  """
8633
  nodenames = _FilterVmNodes(lu, nodenames)
8634

    
8635
  cluster = lu.cfg.GetClusterInfo()
8636
  hvfull = objects.FillDict(cluster.hvparams.get(hvname, {}), hvparams)
8637

    
8638
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames, hvname, hvfull)
8639
  for node in nodenames:
8640
    info = hvinfo[node]
8641
    if info.offline:
8642
      continue
8643
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
8644

    
8645

    
8646
def _CheckOSParams(lu, required, nodenames, osname, osparams):
8647
  """OS parameters validation.
8648

8649
  @type lu: L{LogicalUnit}
8650
  @param lu: the logical unit for which we check
8651
  @type required: boolean
8652
  @param required: whether the validation should fail if the OS is not
8653
      found
8654
  @type nodenames: list
8655
  @param nodenames: the list of nodes on which we should check
8656
  @type osname: string
8657
  @param osname: the name of the hypervisor we should use
8658
  @type osparams: dict
8659
  @param osparams: the parameters which we need to check
8660
  @raise errors.OpPrereqError: if the parameters are not valid
8661

8662
  """
8663
  nodenames = _FilterVmNodes(lu, nodenames)
8664
  result = lu.rpc.call_os_validate(nodenames, required, osname,
8665
                                   [constants.OS_VALIDATE_PARAMETERS],
8666
                                   osparams)
8667
  for node, nres in result.items():
8668
    # we don't check for offline cases since this should be run only
8669
    # against the master node and/or an instance's nodes
8670
    nres.Raise("OS Parameters validation failed on node %s" % node)
8671
    if not nres.payload:
8672
      lu.LogInfo("OS %s not found on node %s, validation skipped",
8673
                 osname, node)
8674

    
8675

    
8676
class LUInstanceCreate(LogicalUnit):
8677
  """Create an instance.
8678

8679
  """
8680
  HPATH = "instance-add"
8681
  HTYPE = constants.HTYPE_INSTANCE
8682
  REQ_BGL = False
8683

    
8684
  def CheckArguments(self):
8685
    """Check arguments.
8686

8687
    """
8688
    # do not require name_check to ease forward/backward compatibility
8689
    # for tools
8690
    if self.op.no_install and self.op.start:
8691
      self.LogInfo("No-installation mode selected, disabling startup")
8692
      self.op.start = False
8693
    # validate/normalize the instance name
8694
    self.op.instance_name = \
8695
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
8696

    
8697
    if self.op.ip_check and not self.op.name_check:
8698
      # TODO: make the ip check more flexible and not depend on the name check
8699
      raise errors.OpPrereqError("Cannot do IP address check without a name"
8700
                                 " check", errors.ECODE_INVAL)
8701

    
8702
    # check nics' parameter names
8703
    for nic in self.op.nics:
8704
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
8705

    
8706
    # check disks. parameter names and consistent adopt/no-adopt strategy
8707
    has_adopt = has_no_adopt = False
8708
    for disk in self.op.disks:
8709
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
8710
      if constants.IDISK_ADOPT in disk:
8711
        has_adopt = True
8712
      else:
8713
        has_no_adopt = True
8714
    if has_adopt and has_no_adopt:
8715
      raise errors.OpPrereqError("Either all disks are adopted or none is",
8716
                                 errors.ECODE_INVAL)
8717
    if has_adopt:
8718
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
8719
        raise errors.OpPrereqError("Disk adoption is not supported for the"
8720
                                   " '%s' disk template" %
8721
                                   self.op.disk_template,
8722
                                   errors.ECODE_INVAL)
8723
      if self.op.iallocator is not None:
8724
        raise errors.OpPrereqError("Disk adoption not allowed with an"
8725
                                   " iallocator script", errors.ECODE_INVAL)
8726
      if self.op.mode == constants.INSTANCE_IMPORT:
8727
        raise errors.OpPrereqError("Disk adoption not allowed for"
8728
                                   " instance import", errors.ECODE_INVAL)
8729
    else:
8730
      if self.op.disk_template in constants.DTS_MUST_ADOPT:
8731
        raise errors.OpPrereqError("Disk template %s requires disk adoption,"
8732
                                   " but no 'adopt' parameter given" %
8733
                                   self.op.disk_template,
8734
                                   errors.ECODE_INVAL)
8735

    
8736
    self.adopt_disks = has_adopt
8737

    
8738
    # instance name verification
8739
    if self.op.name_check:
8740
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
8741
      self.op.instance_name = self.hostname1.name
8742
      # used in CheckPrereq for ip ping check
8743
      self.check_ip = self.hostname1.ip
8744
    else:
8745
      self.check_ip = None
8746

    
8747
    # file storage checks
8748
    if (self.op.file_driver and
8749
        not self.op.file_driver in constants.FILE_DRIVER):
8750
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
8751
                                 self.op.file_driver, errors.ECODE_INVAL)
8752

    
8753
    if self.op.disk_template == constants.DT_FILE:
8754
      opcodes.RequireFileStorage()
8755
    elif self.op.disk_template == constants.DT_SHARED_FILE:
8756
      opcodes.RequireSharedFileStorage()
8757

    
8758
    ### Node/iallocator related checks
8759
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
8760

    
8761
    if self.op.pnode is not None:
8762
      if self.op.disk_template in constants.DTS_INT_MIRROR:
8763
        if self.op.snode is None:
8764
          raise errors.OpPrereqError("The networked disk templates need"
8765
                                     " a mirror node", errors.ECODE_INVAL)
8766
      elif self.op.snode:
8767
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
8768
                        " template")
8769
        self.op.snode = None
8770

    
8771
    self._cds = _GetClusterDomainSecret()
8772

    
8773
    if self.op.mode == constants.INSTANCE_IMPORT:
8774
      # On import force_variant must be True, because if we forced it at
8775
      # initial install, our only chance when importing it back is that it
8776
      # works again!
8777
      self.op.force_variant = True
8778

    
8779
      if self.op.no_install:
8780
        self.LogInfo("No-installation mode has no effect during import")
8781

    
8782
    elif self.op.mode == constants.INSTANCE_CREATE:
8783
      if self.op.os_type is None:
8784
        raise errors.OpPrereqError("No guest OS specified",
8785
                                   errors.ECODE_INVAL)
8786
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
8787
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
8788
                                   " installation" % self.op.os_type,
8789
                                   errors.ECODE_STATE)
8790
      if self.op.disk_template is None:
8791
        raise errors.OpPrereqError("No disk template specified",
8792
                                   errors.ECODE_INVAL)
8793

    
8794
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
8795
      # Check handshake to ensure both clusters have the same domain secret
8796
      src_handshake = self.op.source_handshake
8797
      if not src_handshake:
8798
        raise errors.OpPrereqError("Missing source handshake",
8799
                                   errors.ECODE_INVAL)
8800

    
8801
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
8802
                                                           src_handshake)
8803
      if errmsg:
8804
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
8805
                                   errors.ECODE_INVAL)
8806

    
8807
      # Load and check source CA
8808
      self.source_x509_ca_pem = self.op.source_x509_ca
8809
      if not self.source_x509_ca_pem:
8810
        raise errors.OpPrereqError("Missing source X509 CA",
8811
                                   errors.ECODE_INVAL)
8812

    
8813
      try:
8814
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
8815
                                                    self._cds)
8816
      except OpenSSL.crypto.Error, err:
8817
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
8818
                                   (err, ), errors.ECODE_INVAL)
8819

    
8820
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
8821
      if errcode is not None:
8822
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
8823
                                   errors.ECODE_INVAL)
8824

    
8825
      self.source_x509_ca = cert
8826

    
8827
      src_instance_name = self.op.source_instance_name
8828
      if not src_instance_name:
8829
        raise errors.OpPrereqError("Missing source instance name",
8830
                                   errors.ECODE_INVAL)
8831

    
8832
      self.source_instance_name = \
8833
          netutils.GetHostname(name=src_instance_name).name
8834

    
8835
    else:
8836
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
8837
                                 self.op.mode, errors.ECODE_INVAL)
8838

    
8839
  def ExpandNames(self):
8840
    """ExpandNames for CreateInstance.
8841

8842
    Figure out the right locks for instance creation.
8843

8844
    """
8845
    self.needed_locks = {}
8846

    
8847
    instance_name = self.op.instance_name
8848
    # this is just a preventive check, but someone might still add this
8849
    # instance in the meantime, and creation will fail at lock-add time
8850
    if instance_name in self.cfg.GetInstanceList():
8851
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
8852
                                 instance_name, errors.ECODE_EXISTS)
8853

    
8854
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
8855

    
8856
    if self.op.iallocator:
8857
      # TODO: Find a solution to not lock all nodes in the cluster, e.g. by
8858
      # specifying a group on instance creation and then selecting nodes from
8859
      # that group
8860
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8861
      self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
8862
    else:
8863
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
8864
      nodelist = [self.op.pnode]
8865
      if self.op.snode is not None:
8866
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
8867
        nodelist.append(self.op.snode)
8868
      self.needed_locks[locking.LEVEL_NODE] = nodelist
8869
      # Lock resources of instance's primary and secondary nodes (copy to
8870
      # prevent accidential modification)
8871
      self.needed_locks[locking.LEVEL_NODE_RES] = list(nodelist)
8872

    
8873
    # in case of import lock the source node too
8874
    if self.op.mode == constants.INSTANCE_IMPORT:
8875
      src_node = self.op.src_node
8876
      src_path = self.op.src_path
8877

    
8878
      if src_path is None:
8879
        self.op.src_path = src_path = self.op.instance_name
8880

    
8881
      if src_node is None:
8882
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8883
        self.op.src_node = None
8884
        if os.path.isabs(src_path):
8885
          raise errors.OpPrereqError("Importing an instance from a path"
8886
                                     " requires a source node option",
8887
                                     errors.ECODE_INVAL)
8888
      else:
8889
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
8890
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
8891
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
8892
        if not os.path.isabs(src_path):
8893
          self.op.src_path = src_path = \
8894
            utils.PathJoin(constants.EXPORT_DIR, src_path)
8895

    
8896
  def _RunAllocator(self):
8897
    """Run the allocator based on input opcode.
8898

8899
    """
8900
    nics = [n.ToDict() for n in self.nics]
8901
    ial = IAllocator(self.cfg, self.rpc,
8902
                     mode=constants.IALLOCATOR_MODE_ALLOC,
8903
                     name=self.op.instance_name,
8904
                     disk_template=self.op.disk_template,
8905
                     tags=self.op.tags,
8906
                     os=self.op.os_type,
8907
                     vcpus=self.be_full[constants.BE_VCPUS],
8908
                     memory=self.be_full[constants.BE_MAXMEM],
8909
                     disks=self.disks,
8910
                     nics=nics,
8911
                     hypervisor=self.op.hypervisor,
8912
                     )
8913

    
8914
    ial.Run(self.op.iallocator)
8915

    
8916
    if not ial.success:
8917
      raise errors.OpPrereqError("Can't compute nodes using"
8918
                                 " iallocator '%s': %s" %
8919
                                 (self.op.iallocator, ial.info),
8920
                                 errors.ECODE_NORES)
8921
    if len(ial.result) != ial.required_nodes:
8922
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8923
                                 " of nodes (%s), required %s" %
8924
                                 (self.op.iallocator, len(ial.result),
8925
                                  ial.required_nodes), errors.ECODE_FAULT)
8926
    self.op.pnode = ial.result[0]
8927
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
8928
                 self.op.instance_name, self.op.iallocator,
8929
                 utils.CommaJoin(ial.result))
8930
    if ial.required_nodes == 2:
8931
      self.op.snode = ial.result[1]
8932

    
8933
  def BuildHooksEnv(self):
8934
    """Build hooks env.
8935

8936
    This runs on master, primary and secondary nodes of the instance.
8937

8938
    """
8939
    env = {
8940
      "ADD_MODE": self.op.mode,
8941
      }
8942
    if self.op.mode == constants.INSTANCE_IMPORT:
8943
      env["SRC_NODE"] = self.op.src_node
8944
      env["SRC_PATH"] = self.op.src_path
8945
      env["SRC_IMAGES"] = self.src_images
8946

    
8947
    env.update(_BuildInstanceHookEnv(
8948
      name=self.op.instance_name,
8949
      primary_node=self.op.pnode,
8950
      secondary_nodes=self.secondaries,
8951
      status=self.op.start,
8952
      os_type=self.op.os_type,
8953
      minmem=self.be_full[constants.BE_MINMEM],
8954
      maxmem=self.be_full[constants.BE_MAXMEM],
8955
      vcpus=self.be_full[constants.BE_VCPUS],
8956
      nics=_NICListToTuple(self, self.nics),
8957
      disk_template=self.op.disk_template,
8958
      disks=[(d[constants.IDISK_SIZE], d[constants.IDISK_MODE])
8959
             for d in self.disks],
8960
      bep=self.be_full,
8961
      hvp=self.hv_full,
8962
      hypervisor_name=self.op.hypervisor,
8963
      tags=self.op.tags,
8964
    ))
8965

    
8966
    return env
8967

    
8968
  def BuildHooksNodes(self):
8969
    """Build hooks nodes.
8970

8971
    """
8972
    nl = [self.cfg.GetMasterNode(), self.op.pnode] + self.secondaries
8973
    return nl, nl
8974

    
8975
  def _ReadExportInfo(self):
8976
    """Reads the export information from disk.
8977

8978
    It will override the opcode source node and path with the actual
8979
    information, if these two were not specified before.
8980

8981
    @return: the export information
8982

8983
    """
8984
    assert self.op.mode == constants.INSTANCE_IMPORT
8985

    
8986
    src_node = self.op.src_node
8987
    src_path = self.op.src_path
8988

    
8989
    if src_node is None:
8990
      locked_nodes = self.owned_locks(locking.LEVEL_NODE)
8991
      exp_list = self.rpc.call_export_list(locked_nodes)
8992
      found = False
8993
      for node in exp_list:
8994
        if exp_list[node].fail_msg:
8995
          continue
8996
        if src_path in exp_list[node].payload:
8997
          found = True
8998
          self.op.src_node = src_node = node
8999
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
9000
                                                       src_path)
9001
          break
9002
      if not found:
9003
        raise errors.OpPrereqError("No export found for relative path %s" %
9004
                                    src_path, errors.ECODE_INVAL)
9005

    
9006
    _CheckNodeOnline(self, src_node)
9007
    result = self.rpc.call_export_info(src_node, src_path)
9008
    result.Raise("No export or invalid export found in dir %s" % src_path)
9009

    
9010
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
9011
    if not export_info.has_section(constants.INISECT_EXP):
9012
      raise errors.ProgrammerError("Corrupted export config",
9013
                                   errors.ECODE_ENVIRON)
9014

    
9015
    ei_version = export_info.get(constants.INISECT_EXP, "version")
9016
    if (int(ei_version) != constants.EXPORT_VERSION):
9017
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
9018
                                 (ei_version, constants.EXPORT_VERSION),
9019
                                 errors.ECODE_ENVIRON)
9020
    return export_info
9021

    
9022
  def _ReadExportParams(self, einfo):
9023
    """Use export parameters as defaults.
9024

9025
    In case the opcode doesn't specify (as in override) some instance
9026
    parameters, then try to use them from the export information, if
9027
    that declares them.
9028

9029
    """
9030
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
9031

    
9032
    if self.op.disk_template is None:
9033
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
9034
        self.op.disk_template = einfo.get(constants.INISECT_INS,
9035
                                          "disk_template")
9036
        if self.op.disk_template not in constants.DISK_TEMPLATES:
9037
          raise errors.OpPrereqError("Disk template specified in configuration"
9038
                                     " file is not one of the allowed values:"
9039
                                     " %s" % " ".join(constants.DISK_TEMPLATES))
9040
      else:
9041
        raise errors.OpPrereqError("No disk template specified and the export"
9042
                                   " is missing the disk_template information",
9043
                                   errors.ECODE_INVAL)
9044

    
9045
    if not self.op.disks:
9046
      disks = []
9047
      # TODO: import the disk iv_name too
9048
      for idx in range(constants.MAX_DISKS):
9049
        if einfo.has_option(constants.INISECT_INS, "disk%d_size" % idx):
9050
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
9051
          disks.append({constants.IDISK_SIZE: disk_sz})
9052
      self.op.disks = disks
9053
      if not disks and self.op.disk_template != constants.DT_DISKLESS:
9054
        raise errors.OpPrereqError("No disk info specified and the export"
9055
                                   " is missing the disk information",
9056
                                   errors.ECODE_INVAL)
9057

    
9058
    if not self.op.nics:
9059
      nics = []
9060
      for idx in range(constants.MAX_NICS):
9061
        if einfo.has_option(constants.INISECT_INS, "nic%d_mac" % idx):
9062
          ndict = {}
9063
          for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
9064
            v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
9065
            ndict[name] = v
9066
          nics.append(ndict)
9067
        else:
9068
          break
9069
      self.op.nics = nics
9070

    
9071
    if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
9072
      self.op.tags = einfo.get(constants.INISECT_INS, "tags").split()
9073

    
9074
    if (self.op.hypervisor is None and
9075
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
9076
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
9077

    
9078
    if einfo.has_section(constants.INISECT_HYP):
9079
      # use the export parameters but do not override the ones
9080
      # specified by the user
9081
      for name, value in einfo.items(constants.INISECT_HYP):
9082
        if name not in self.op.hvparams:
9083
          self.op.hvparams[name] = value
9084

    
9085
    if einfo.has_section(constants.INISECT_BEP):
9086
      # use the parameters, without overriding
9087
      for name, value in einfo.items(constants.INISECT_BEP):
9088
        if name not in self.op.beparams:
9089
          self.op.beparams[name] = value
9090
        # Compatibility for the old "memory" be param
9091
        if name == constants.BE_MEMORY:
9092
          if constants.BE_MAXMEM not in self.op.beparams:
9093
            self.op.beparams[constants.BE_MAXMEM] = value
9094
          if constants.BE_MINMEM not in self.op.beparams:
9095
            self.op.beparams[constants.BE_MINMEM] = value
9096
    else:
9097
      # try to read the parameters old style, from the main section
9098
      for name in constants.BES_PARAMETERS:
9099
        if (name not in self.op.beparams and
9100
            einfo.has_option(constants.INISECT_INS, name)):
9101
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
9102

    
9103
    if einfo.has_section(constants.INISECT_OSP):
9104
      # use the parameters, without overriding
9105
      for name, value in einfo.items(constants.INISECT_OSP):
9106
        if name not in self.op.osparams:
9107
          self.op.osparams[name] = value
9108

    
9109
  def _RevertToDefaults(self, cluster):
9110
    """Revert the instance parameters to the default values.
9111

9112
    """
9113
    # hvparams
9114
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
9115
    for name in self.op.hvparams.keys():
9116
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
9117
        del self.op.hvparams[name]
9118
    # beparams
9119
    be_defs = cluster.SimpleFillBE({})
9120
    for name in self.op.beparams.keys():
9121
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
9122
        del self.op.beparams[name]
9123
    # nic params
9124
    nic_defs = cluster.SimpleFillNIC({})
9125
    for nic in self.op.nics:
9126
      for name in constants.NICS_PARAMETERS:
9127
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
9128
          del nic[name]
9129
    # osparams
9130
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
9131
    for name in self.op.osparams.keys():
9132
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
9133
        del self.op.osparams[name]
9134

    
9135
  def _CalculateFileStorageDir(self):
9136
    """Calculate final instance file storage dir.
9137

9138
    """
9139
    # file storage dir calculation/check
9140
    self.instance_file_storage_dir = None
9141
    if self.op.disk_template in constants.DTS_FILEBASED:
9142
      # build the full file storage dir path
9143
      joinargs = []
9144

    
9145
      if self.op.disk_template == constants.DT_SHARED_FILE:
9146
        get_fsd_fn = self.cfg.GetSharedFileStorageDir
9147
      else:
9148
        get_fsd_fn = self.cfg.GetFileStorageDir
9149

    
9150
      cfg_storagedir = get_fsd_fn()
9151
      if not cfg_storagedir:
9152
        raise errors.OpPrereqError("Cluster file storage dir not defined")
9153
      joinargs.append(cfg_storagedir)
9154

    
9155
      if self.op.file_storage_dir is not None:
9156
        joinargs.append(self.op.file_storage_dir)
9157

    
9158
      joinargs.append(self.op.instance_name)
9159

    
9160
      # pylint: disable=W0142
9161
      self.instance_file_storage_dir = utils.PathJoin(*joinargs)
9162

    
9163
  def CheckPrereq(self):
9164
    """Check prerequisites.
9165

9166
    """
9167
    self._CalculateFileStorageDir()
9168

    
9169
    if self.op.mode == constants.INSTANCE_IMPORT:
9170
      export_info = self._ReadExportInfo()
9171
      self._ReadExportParams(export_info)
9172

    
9173
    if (not self.cfg.GetVGName() and
9174
        self.op.disk_template not in constants.DTS_NOT_LVM):
9175
      raise errors.OpPrereqError("Cluster does not support lvm-based"
9176
                                 " instances", errors.ECODE_STATE)
9177

    
9178
    if (self.op.hypervisor is None or
9179
        self.op.hypervisor == constants.VALUE_AUTO):
9180
      self.op.hypervisor = self.cfg.GetHypervisorType()
9181

    
9182
    cluster = self.cfg.GetClusterInfo()
9183
    enabled_hvs = cluster.enabled_hypervisors
9184
    if self.op.hypervisor not in enabled_hvs:
9185
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
9186
                                 " cluster (%s)" % (self.op.hypervisor,
9187
                                  ",".join(enabled_hvs)),
9188
                                 errors.ECODE_STATE)
9189

    
9190
    # Check tag validity
9191
    for tag in self.op.tags:
9192
      objects.TaggableObject.ValidateTag(tag)
9193

    
9194
    # check hypervisor parameter syntax (locally)
9195
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
9196
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
9197
                                      self.op.hvparams)
9198
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
9199
    hv_type.CheckParameterSyntax(filled_hvp)
9200
    self.hv_full = filled_hvp
9201
    # check that we don't specify global parameters on an instance
9202
    _CheckGlobalHvParams(self.op.hvparams)
9203

    
9204
    # fill and remember the beparams dict
9205
    default_beparams = cluster.beparams[constants.PP_DEFAULT]
9206
    for param, value in self.op.beparams.iteritems():
9207
      if value == constants.VALUE_AUTO:
9208
        self.op.beparams[param] = default_beparams[param]
9209
    objects.UpgradeBeParams(self.op.beparams)
9210
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
9211
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
9212

    
9213
    # build os parameters
9214
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
9215

    
9216
    # now that hvp/bep are in final format, let's reset to defaults,
9217
    # if told to do so
9218
    if self.op.identify_defaults:
9219
      self._RevertToDefaults(cluster)
9220

    
9221
    # NIC buildup
9222
    self.nics = []
9223
    for idx, nic in enumerate(self.op.nics):
9224
      nic_mode_req = nic.get(constants.INIC_MODE, None)
9225
      nic_mode = nic_mode_req
9226
      if nic_mode is None or nic_mode == constants.VALUE_AUTO:
9227
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
9228

    
9229
      # in routed mode, for the first nic, the default ip is 'auto'
9230
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
9231
        default_ip_mode = constants.VALUE_AUTO
9232
      else:
9233
        default_ip_mode = constants.VALUE_NONE
9234

    
9235
      # ip validity checks
9236
      ip = nic.get(constants.INIC_IP, default_ip_mode)
9237
      if ip is None or ip.lower() == constants.VALUE_NONE:
9238
        nic_ip = None
9239
      elif ip.lower() == constants.VALUE_AUTO:
9240
        if not self.op.name_check:
9241
          raise errors.OpPrereqError("IP address set to auto but name checks"
9242
                                     " have been skipped",
9243
                                     errors.ECODE_INVAL)
9244
        nic_ip = self.hostname1.ip
9245
      else:
9246
        if not netutils.IPAddress.IsValid(ip):
9247
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
9248
                                     errors.ECODE_INVAL)
9249
        nic_ip = ip
9250

    
9251
      # TODO: check the ip address for uniqueness
9252
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
9253
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
9254
                                   errors.ECODE_INVAL)
9255

    
9256
      # MAC address verification
9257
      mac = nic.get(constants.INIC_MAC, constants.VALUE_AUTO)
9258
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9259
        mac = utils.NormalizeAndValidateMac(mac)
9260

    
9261
        try:
9262
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
9263
        except errors.ReservationError:
9264
          raise errors.OpPrereqError("MAC address %s already in use"
9265
                                     " in cluster" % mac,
9266
                                     errors.ECODE_NOTUNIQUE)
9267

    
9268
      #  Build nic parameters
9269
      link = nic.get(constants.INIC_LINK, None)
9270
      if link == constants.VALUE_AUTO:
9271
        link = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_LINK]
9272
      nicparams = {}
9273
      if nic_mode_req:
9274
        nicparams[constants.NIC_MODE] = nic_mode
9275
      if link:
9276
        nicparams[constants.NIC_LINK] = link
9277

    
9278
      check_params = cluster.SimpleFillNIC(nicparams)
9279
      objects.NIC.CheckParameterSyntax(check_params)
9280
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
9281

    
9282
    # disk checks/pre-build
9283
    default_vg = self.cfg.GetVGName()
9284
    self.disks = []
9285
    for disk in self.op.disks:
9286
      mode = disk.get(constants.IDISK_MODE, constants.DISK_RDWR)
9287
      if mode not in constants.DISK_ACCESS_SET:
9288
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
9289
                                   mode, errors.ECODE_INVAL)
9290
      size = disk.get(constants.IDISK_SIZE, None)
9291
      if size is None:
9292
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
9293
      try:
9294
        size = int(size)
9295
      except (TypeError, ValueError):
9296
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
9297
                                   errors.ECODE_INVAL)
9298

    
9299
      data_vg = disk.get(constants.IDISK_VG, default_vg)
9300
      new_disk = {
9301
        constants.IDISK_SIZE: size,
9302
        constants.IDISK_MODE: mode,
9303
        constants.IDISK_VG: data_vg,
9304
        constants.IDISK_METAVG: disk.get(constants.IDISK_METAVG, data_vg),
9305
        }
9306
      if constants.IDISK_ADOPT in disk:
9307
        new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
9308
      self.disks.append(new_disk)
9309

    
9310
    if self.op.mode == constants.INSTANCE_IMPORT:
9311
      disk_images = []
9312
      for idx in range(len(self.disks)):
9313
        option = "disk%d_dump" % idx
9314
        if export_info.has_option(constants.INISECT_INS, option):
9315
          # FIXME: are the old os-es, disk sizes, etc. useful?
9316
          export_name = export_info.get(constants.INISECT_INS, option)
9317
          image = utils.PathJoin(self.op.src_path, export_name)
9318
          disk_images.append(image)
9319
        else:
9320
          disk_images.append(False)
9321

    
9322
      self.src_images = disk_images
9323

    
9324
      old_name = export_info.get(constants.INISECT_INS, "name")
9325
      if self.op.instance_name == old_name:
9326
        for idx, nic in enumerate(self.nics):
9327
          if nic.mac == constants.VALUE_AUTO:
9328
            nic_mac_ini = "nic%d_mac" % idx
9329
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
9330

    
9331
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
9332

    
9333
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
9334
    if self.op.ip_check:
9335
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
9336
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
9337
                                   (self.check_ip, self.op.instance_name),
9338
                                   errors.ECODE_NOTUNIQUE)
9339

    
9340
    #### mac address generation
9341
    # By generating here the mac address both the allocator and the hooks get
9342
    # the real final mac address rather than the 'auto' or 'generate' value.
9343
    # There is a race condition between the generation and the instance object
9344
    # creation, which means that we know the mac is valid now, but we're not
9345
    # sure it will be when we actually add the instance. If things go bad
9346
    # adding the instance will abort because of a duplicate mac, and the
9347
    # creation job will fail.
9348
    for nic in self.nics:
9349
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9350
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
9351

    
9352
    #### allocator run
9353

    
9354
    if self.op.iallocator is not None:
9355
      self._RunAllocator()
9356

    
9357
    # Release all unneeded node locks
9358
    _ReleaseLocks(self, locking.LEVEL_NODE,
9359
                  keep=filter(None, [self.op.pnode, self.op.snode,
9360
                                     self.op.src_node]))
9361

    
9362
    #### node related checks
9363

    
9364
    # check primary node
9365
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
9366
    assert self.pnode is not None, \
9367
      "Cannot retrieve locked node %s" % self.op.pnode
9368
    if pnode.offline:
9369
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
9370
                                 pnode.name, errors.ECODE_STATE)
9371
    if pnode.drained:
9372
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
9373
                                 pnode.name, errors.ECODE_STATE)
9374
    if not pnode.vm_capable:
9375
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
9376
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
9377

    
9378
    self.secondaries = []
9379

    
9380
    # mirror node verification
9381
    if self.op.disk_template in constants.DTS_INT_MIRROR:
9382
      if self.op.snode == pnode.name:
9383
        raise errors.OpPrereqError("The secondary node cannot be the"
9384
                                   " primary node", errors.ECODE_INVAL)
9385
      _CheckNodeOnline(self, self.op.snode)
9386
      _CheckNodeNotDrained(self, self.op.snode)
9387
      _CheckNodeVmCapable(self, self.op.snode)
9388
      self.secondaries.append(self.op.snode)
9389

    
9390
      snode = self.cfg.GetNodeInfo(self.op.snode)
9391
      if pnode.group != snode.group:
9392
        self.LogWarning("The primary and secondary nodes are in two"
9393
                        " different node groups; the disk parameters"
9394
                        " from the first disk's node group will be"
9395
                        " used")
9396

    
9397
    nodenames = [pnode.name] + self.secondaries
9398

    
9399
    # disk parameters (not customizable at instance or node level)
9400
    # just use the primary node parameters, ignoring the secondary.
9401
    self.diskparams = self.cfg.GetNodeGroup(pnode.group).diskparams
9402

    
9403
    if not self.adopt_disks:
9404
      # Check lv size requirements, if not adopting
9405
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
9406
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
9407

    
9408
    elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
9409
      all_lvs = set(["%s/%s" % (disk[constants.IDISK_VG],
9410
                                disk[constants.IDISK_ADOPT])
9411
                     for disk in self.disks])
9412
      if len(all_lvs) != len(self.disks):
9413
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
9414
                                   errors.ECODE_INVAL)
9415
      for lv_name in all_lvs:
9416
        try:
9417
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
9418
          # to ReserveLV uses the same syntax
9419
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
9420
        except errors.ReservationError:
9421
          raise errors.OpPrereqError("LV named %s used by another instance" %
9422
                                     lv_name, errors.ECODE_NOTUNIQUE)
9423

    
9424
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
9425
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
9426

    
9427
      node_lvs = self.rpc.call_lv_list([pnode.name],
9428
                                       vg_names.payload.keys())[pnode.name]
9429
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
9430
      node_lvs = node_lvs.payload
9431

    
9432
      delta = all_lvs.difference(node_lvs.keys())
9433
      if delta:
9434
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
9435
                                   utils.CommaJoin(delta),
9436
                                   errors.ECODE_INVAL)
9437
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
9438
      if online_lvs:
9439
        raise errors.OpPrereqError("Online logical volumes found, cannot"
9440
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
9441
                                   errors.ECODE_STATE)
9442
      # update the size of disk based on what is found
9443
      for dsk in self.disks:
9444
        dsk[constants.IDISK_SIZE] = \
9445
          int(float(node_lvs["%s/%s" % (dsk[constants.IDISK_VG],
9446
                                        dsk[constants.IDISK_ADOPT])][0]))
9447

    
9448
    elif self.op.disk_template == constants.DT_BLOCK:
9449
      # Normalize and de-duplicate device paths
9450
      all_disks = set([os.path.abspath(disk[constants.IDISK_ADOPT])
9451
                       for disk in self.disks])
9452
      if len(all_disks) != len(self.disks):
9453
        raise errors.OpPrereqError("Duplicate disk names given for adoption",
9454
                                   errors.ECODE_INVAL)
9455
      baddisks = [d for d in all_disks
9456
                  if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
9457
      if baddisks:
9458
        raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
9459
                                   " cannot be adopted" %
9460
                                   (", ".join(baddisks),
9461
                                    constants.ADOPTABLE_BLOCKDEV_ROOT),
9462
                                   errors.ECODE_INVAL)
9463

    
9464
      node_disks = self.rpc.call_bdev_sizes([pnode.name],
9465
                                            list(all_disks))[pnode.name]
9466
      node_disks.Raise("Cannot get block device information from node %s" %
9467
                       pnode.name)
9468
      node_disks = node_disks.payload
9469
      delta = all_disks.difference(node_disks.keys())
9470
      if delta:
9471
        raise errors.OpPrereqError("Missing block device(s): %s" %
9472
                                   utils.CommaJoin(delta),
9473
                                   errors.ECODE_INVAL)
9474
      for dsk in self.disks:
9475
        dsk[constants.IDISK_SIZE] = \
9476
          int(float(node_disks[dsk[constants.IDISK_ADOPT]]))
9477

    
9478
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
9479

    
9480
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
9481
    # check OS parameters (remotely)
9482
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
9483

    
9484
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
9485

    
9486
    # memory check on primary node
9487
    #TODO(dynmem): use MINMEM for checking
9488
    if self.op.start:
9489
      _CheckNodeFreeMemory(self, self.pnode.name,
9490
                           "creating instance %s" % self.op.instance_name,
9491
                           self.be_full[constants.BE_MAXMEM],
9492
                           self.op.hypervisor)
9493

    
9494
    self.dry_run_result = list(nodenames)
9495

    
9496
  def Exec(self, feedback_fn):
9497
    """Create and add the instance to the cluster.
9498

9499
    """
9500
    instance = self.op.instance_name
9501
    pnode_name = self.pnode.name
9502

    
9503
    assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
9504
                self.owned_locks(locking.LEVEL_NODE)), \
9505
      "Node locks differ from node resource locks"
9506

    
9507
    ht_kind = self.op.hypervisor
9508
    if ht_kind in constants.HTS_REQ_PORT:
9509
      network_port = self.cfg.AllocatePort()
9510
    else:
9511
      network_port = None
9512

    
9513
    disks = _GenerateDiskTemplate(self,
9514
                                  self.op.disk_template,
9515
                                  instance, pnode_name,
9516
                                  self.secondaries,
9517
                                  self.disks,
9518
                                  self.instance_file_storage_dir,
9519
                                  self.op.file_driver,
9520
                                  0,
9521
                                  feedback_fn,
9522
                                  self.diskparams)
9523

    
9524
    iobj = objects.Instance(name=instance, os=self.op.os_type,
9525
                            primary_node=pnode_name,
9526
                            nics=self.nics, disks=disks,
9527
                            disk_template=self.op.disk_template,
9528
                            admin_state=constants.ADMINST_DOWN,
9529
                            network_port=network_port,
9530
                            beparams=self.op.beparams,
9531
                            hvparams=self.op.hvparams,
9532
                            hypervisor=self.op.hypervisor,
9533
                            osparams=self.op.osparams,
9534
                            )
9535

    
9536
    if self.op.tags:
9537
      for tag in self.op.tags:
9538
        iobj.AddTag(tag)
9539

    
9540
    if self.adopt_disks:
9541
      if self.op.disk_template == constants.DT_PLAIN:
9542
        # rename LVs to the newly-generated names; we need to construct
9543
        # 'fake' LV disks with the old data, plus the new unique_id
9544
        tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
9545
        rename_to = []
9546
        for t_dsk, a_dsk in zip(tmp_disks, self.disks):
9547
          rename_to.append(t_dsk.logical_id)
9548
          t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
9549
          self.cfg.SetDiskID(t_dsk, pnode_name)
9550
        result = self.rpc.call_blockdev_rename(pnode_name,
9551
                                               zip(tmp_disks, rename_to))
9552
        result.Raise("Failed to rename adoped LVs")
9553
    else:
9554
      feedback_fn("* creating instance disks...")
9555
      try:
9556
        _CreateDisks(self, iobj)
9557
      except errors.OpExecError:
9558
        self.LogWarning("Device creation failed, reverting...")
9559
        try:
9560
          _RemoveDisks(self, iobj)
9561
        finally:
9562
          self.cfg.ReleaseDRBDMinors(instance)
9563
          raise
9564

    
9565
    feedback_fn("adding instance %s to cluster config" % instance)
9566

    
9567
    self.cfg.AddInstance(iobj, self.proc.GetECId())
9568

    
9569
    # Declare that we don't want to remove the instance lock anymore, as we've
9570
    # added the instance to the config
9571
    del self.remove_locks[locking.LEVEL_INSTANCE]
9572

    
9573
    if self.op.mode == constants.INSTANCE_IMPORT:
9574
      # Release unused nodes
9575
      _ReleaseLocks(self, locking.LEVEL_NODE, keep=[self.op.src_node])
9576
    else:
9577
      # Release all nodes
9578
      _ReleaseLocks(self, locking.LEVEL_NODE)
9579

    
9580
    disk_abort = False
9581
    if not self.adopt_disks and self.cfg.GetClusterInfo().prealloc_wipe_disks:
9582
      feedback_fn("* wiping instance disks...")
9583
      try:
9584
        _WipeDisks(self, iobj)
9585
      except errors.OpExecError, err:
9586
        logging.exception("Wiping disks failed")
9587
        self.LogWarning("Wiping instance disks failed (%s)", err)
9588
        disk_abort = True
9589

    
9590
    if disk_abort:
9591
      # Something is already wrong with the disks, don't do anything else
9592
      pass
9593
    elif self.op.wait_for_sync:
9594
      disk_abort = not _WaitForSync(self, iobj)
9595
    elif iobj.disk_template in constants.DTS_INT_MIRROR:
9596
      # make sure the disks are not degraded (still sync-ing is ok)
9597
      feedback_fn("* checking mirrors status")
9598
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
9599
    else:
9600
      disk_abort = False
9601

    
9602
    if disk_abort:
9603
      _RemoveDisks(self, iobj)
9604
      self.cfg.RemoveInstance(iobj.name)
9605
      # Make sure the instance lock gets removed
9606
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
9607
      raise errors.OpExecError("There are some degraded disks for"
9608
                               " this instance")
9609

    
9610
    # Release all node resource locks
9611
    _ReleaseLocks(self, locking.LEVEL_NODE_RES)
9612

    
9613
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
9614
      if self.op.mode == constants.INSTANCE_CREATE:
9615
        if not self.op.no_install:
9616
          pause_sync = (iobj.disk_template in constants.DTS_INT_MIRROR and
9617
                        not self.op.wait_for_sync)
9618
          if pause_sync:
9619
            feedback_fn("* pausing disk sync to install instance OS")
9620
            result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
9621
                                                              iobj.disks, True)
9622
            for idx, success in enumerate(result.payload):
9623
              if not success:
9624
                logging.warn("pause-sync of instance %s for disk %d failed",
9625
                             instance, idx)
9626

    
9627
          feedback_fn("* running the instance OS create scripts...")
9628
          # FIXME: pass debug option from opcode to backend
9629
          os_add_result = \
9630
            self.rpc.call_instance_os_add(pnode_name, (iobj, None), False,
9631
                                          self.op.debug_level)
9632
          if pause_sync:
9633
            feedback_fn("* resuming disk sync")
9634
            result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
9635
                                                              iobj.disks, False)
9636
            for idx, success in enumerate(result.payload):
9637
              if not success:
9638
                logging.warn("resume-sync of instance %s for disk %d failed",
9639
                             instance, idx)
9640

    
9641
          os_add_result.Raise("Could not add os for instance %s"
9642
                              " on node %s" % (instance, pnode_name))
9643

    
9644
      elif self.op.mode == constants.INSTANCE_IMPORT:
9645
        feedback_fn("* running the instance OS import scripts...")
9646

    
9647
        transfers = []
9648

    
9649
        for idx, image in enumerate(self.src_images):
9650
          if not image:
9651
            continue
9652

    
9653
          # FIXME: pass debug option from opcode to backend
9654
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
9655
                                             constants.IEIO_FILE, (image, ),
9656
                                             constants.IEIO_SCRIPT,
9657
                                             (iobj.disks[idx], idx),
9658
                                             None)
9659
          transfers.append(dt)
9660

    
9661
        import_result = \
9662
          masterd.instance.TransferInstanceData(self, feedback_fn,
9663
                                                self.op.src_node, pnode_name,
9664
                                                self.pnode.secondary_ip,
9665
                                                iobj, transfers)
9666
        if not compat.all(import_result):
9667
          self.LogWarning("Some disks for instance %s on node %s were not"
9668
                          " imported successfully" % (instance, pnode_name))
9669

    
9670
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
9671
        feedback_fn("* preparing remote import...")
9672
        # The source cluster will stop the instance before attempting to make a
9673
        # connection. In some cases stopping an instance can take a long time,
9674
        # hence the shutdown timeout is added to the connection timeout.
9675
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
9676
                           self.op.source_shutdown_timeout)
9677
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
9678

    
9679
        assert iobj.primary_node == self.pnode.name
9680
        disk_results = \
9681
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
9682
                                        self.source_x509_ca,
9683
                                        self._cds, timeouts)
9684
        if not compat.all(disk_results):
9685
          # TODO: Should the instance still be started, even if some disks
9686
          # failed to import (valid for local imports, too)?
9687
          self.LogWarning("Some disks for instance %s on node %s were not"
9688
                          " imported successfully" % (instance, pnode_name))
9689

    
9690
        # Run rename script on newly imported instance
9691
        assert iobj.name == instance
9692
        feedback_fn("Running rename script for %s" % instance)
9693
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
9694
                                                   self.source_instance_name,
9695
                                                   self.op.debug_level)
9696
        if result.fail_msg:
9697
          self.LogWarning("Failed to run rename script for %s on node"
9698
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
9699

    
9700
      else:
9701
        # also checked in the prereq part
9702
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
9703
                                     % self.op.mode)
9704

    
9705
    assert not self.owned_locks(locking.LEVEL_NODE_RES)
9706

    
9707
    if self.op.start:
9708
      iobj.admin_state = constants.ADMINST_UP
9709
      self.cfg.Update(iobj, feedback_fn)
9710
      logging.info("Starting instance %s on node %s", instance, pnode_name)
9711
      feedback_fn("* starting instance...")
9712
      result = self.rpc.call_instance_start(pnode_name, (iobj, None, None),
9713
                                            False)
9714
      result.Raise("Could not start instance")
9715

    
9716
    return list(iobj.all_nodes)
9717

    
9718

    
9719
class LUInstanceConsole(NoHooksLU):
9720
  """Connect to an instance's console.
9721

9722
  This is somewhat special in that it returns the command line that
9723
  you need to run on the master node in order to connect to the
9724
  console.
9725

9726
  """
9727
  REQ_BGL = False
9728

    
9729
  def ExpandNames(self):
9730
    self.share_locks = _ShareAll()
9731
    self._ExpandAndLockInstance()
9732

    
9733
  def CheckPrereq(self):
9734
    """Check prerequisites.
9735

9736
    This checks that the instance is in the cluster.
9737

9738
    """
9739
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9740
    assert self.instance is not None, \
9741
      "Cannot retrieve locked instance %s" % self.op.instance_name
9742
    _CheckNodeOnline(self, self.instance.primary_node)
9743

    
9744
  def Exec(self, feedback_fn):
9745
    """Connect to the console of an instance
9746

9747
    """
9748
    instance = self.instance
9749
    node = instance.primary_node
9750

    
9751
    node_insts = self.rpc.call_instance_list([node],
9752
                                             [instance.hypervisor])[node]
9753
    node_insts.Raise("Can't get node information from %s" % node)
9754

    
9755
    if instance.name not in node_insts.payload:
9756
      if instance.admin_state == constants.ADMINST_UP:
9757
        state = constants.INSTST_ERRORDOWN
9758
      elif instance.admin_state == constants.ADMINST_DOWN:
9759
        state = constants.INSTST_ADMINDOWN
9760
      else:
9761
        state = constants.INSTST_ADMINOFFLINE
9762
      raise errors.OpExecError("Instance %s is not running (state %s)" %
9763
                               (instance.name, state))
9764

    
9765
    logging.debug("Connecting to console of %s on %s", instance.name, node)
9766

    
9767
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
9768

    
9769

    
9770
def _GetInstanceConsole(cluster, instance):
9771
  """Returns console information for an instance.
9772

9773
  @type cluster: L{objects.Cluster}
9774
  @type instance: L{objects.Instance}
9775
  @rtype: dict
9776

9777
  """
9778
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
9779
  # beparams and hvparams are passed separately, to avoid editing the
9780
  # instance and then saving the defaults in the instance itself.
9781
  hvparams = cluster.FillHV(instance)
9782
  beparams = cluster.FillBE(instance)
9783
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
9784

    
9785
  assert console.instance == instance.name
9786
  assert console.Validate()
9787

    
9788
  return console.ToDict()
9789

    
9790

    
9791
class LUInstanceReplaceDisks(LogicalUnit):
9792
  """Replace the disks of an instance.
9793

9794
  """
9795
  HPATH = "mirrors-replace"
9796
  HTYPE = constants.HTYPE_INSTANCE
9797
  REQ_BGL = False
9798

    
9799
  def CheckArguments(self):
9800
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
9801
                                  self.op.iallocator)
9802

    
9803
  def ExpandNames(self):
9804
    self._ExpandAndLockInstance()
9805

    
9806
    assert locking.LEVEL_NODE not in self.needed_locks
9807
    assert locking.LEVEL_NODE_RES not in self.needed_locks
9808
    assert locking.LEVEL_NODEGROUP not in self.needed_locks
9809

    
9810
    assert self.op.iallocator is None or self.op.remote_node is None, \
9811
      "Conflicting options"
9812

    
9813
    if self.op.remote_node is not None:
9814
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9815

    
9816
      # Warning: do not remove the locking of the new secondary here
9817
      # unless DRBD8.AddChildren is changed to work in parallel;
9818
      # currently it doesn't since parallel invocations of
9819
      # FindUnusedMinor will conflict
9820
      self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
9821
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
9822
    else:
9823
      self.needed_locks[locking.LEVEL_NODE] = []
9824
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9825

    
9826
      if self.op.iallocator is not None:
9827
        # iallocator will select a new node in the same group
9828
        self.needed_locks[locking.LEVEL_NODEGROUP] = []
9829

    
9830
    self.needed_locks[locking.LEVEL_NODE_RES] = []
9831

    
9832
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
9833
                                   self.op.iallocator, self.op.remote_node,
9834
                                   self.op.disks, False, self.op.early_release)
9835

    
9836
    self.tasklets = [self.replacer]
9837

    
9838
  def DeclareLocks(self, level):
9839
    if level == locking.LEVEL_NODEGROUP:
9840
      assert self.op.remote_node is None
9841
      assert self.op.iallocator is not None
9842
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
9843

    
9844
      self.share_locks[locking.LEVEL_NODEGROUP] = 1
9845
      # Lock all groups used by instance optimistically; this requires going
9846
      # via the node before it's locked, requiring verification later on
9847
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
9848
        self.cfg.GetInstanceNodeGroups(self.op.instance_name)
9849

    
9850
    elif level == locking.LEVEL_NODE:
9851
      if self.op.iallocator is not None:
9852
        assert self.op.remote_node is None
9853
        assert not self.needed_locks[locking.LEVEL_NODE]
9854

    
9855
        # Lock member nodes of all locked groups
9856
        self.needed_locks[locking.LEVEL_NODE] = [node_name
9857
          for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
9858
          for node_name in self.cfg.GetNodeGroup(group_uuid).members]
9859
      else:
9860
        self._LockInstancesNodes()
9861
    elif level == locking.LEVEL_NODE_RES:
9862
      # Reuse node locks
9863
      self.needed_locks[locking.LEVEL_NODE_RES] = \
9864
        self.needed_locks[locking.LEVEL_NODE]
9865

    
9866
  def BuildHooksEnv(self):
9867
    """Build hooks env.
9868

9869
    This runs on the master, the primary and all the secondaries.
9870

9871
    """
9872
    instance = self.replacer.instance
9873
    env = {
9874
      "MODE": self.op.mode,
9875
      "NEW_SECONDARY": self.op.remote_node,
9876
      "OLD_SECONDARY": instance.secondary_nodes[0],
9877
      }
9878
    env.update(_BuildInstanceHookEnvByObject(self, instance))
9879
    return env
9880

    
9881
  def BuildHooksNodes(self):
9882
    """Build hooks nodes.
9883

9884
    """
9885
    instance = self.replacer.instance
9886
    nl = [
9887
      self.cfg.GetMasterNode(),
9888
      instance.primary_node,
9889
      ]
9890
    if self.op.remote_node is not None:
9891
      nl.append(self.op.remote_node)
9892
    return nl, nl
9893

    
9894
  def CheckPrereq(self):
9895
    """Check prerequisites.
9896

9897
    """
9898
    assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
9899
            self.op.iallocator is None)
9900

    
9901
    # Verify if node group locks are still correct
9902
    owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
9903
    if owned_groups:
9904
      _CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
9905

    
9906
    return LogicalUnit.CheckPrereq(self)
9907

    
9908

    
9909
class TLReplaceDisks(Tasklet):
9910
  """Replaces disks for an instance.
9911

9912
  Note: Locking is not within the scope of this class.
9913

9914
  """
9915
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
9916
               disks, delay_iallocator, early_release):
9917
    """Initializes this class.
9918

9919
    """
9920
    Tasklet.__init__(self, lu)
9921

    
9922
    # Parameters
9923
    self.instance_name = instance_name
9924
    self.mode = mode
9925
    self.iallocator_name = iallocator_name
9926
    self.remote_node = remote_node
9927
    self.disks = disks
9928
    self.delay_iallocator = delay_iallocator
9929
    self.early_release = early_release
9930

    
9931
    # Runtime data
9932
    self.instance = None
9933
    self.new_node = None
9934
    self.target_node = None
9935
    self.other_node = None
9936
    self.remote_node_info = None
9937
    self.node_secondary_ip = None
9938

    
9939
  @staticmethod
9940
  def CheckArguments(mode, remote_node, iallocator):
9941
    """Helper function for users of this class.
9942

9943
    """
9944
    # check for valid parameter combination
9945
    if mode == constants.REPLACE_DISK_CHG:
9946
      if remote_node is None and iallocator is None:
9947
        raise errors.OpPrereqError("When changing the secondary either an"
9948
                                   " iallocator script must be used or the"
9949
                                   " new node given", errors.ECODE_INVAL)
9950

    
9951
      if remote_node is not None and iallocator is not None:
9952
        raise errors.OpPrereqError("Give either the iallocator or the new"
9953
                                   " secondary, not both", errors.ECODE_INVAL)
9954

    
9955
    elif remote_node is not None or iallocator is not None:
9956
      # Not replacing the secondary
9957
      raise errors.OpPrereqError("The iallocator and new node options can"
9958
                                 " only be used when changing the"
9959
                                 " secondary node", errors.ECODE_INVAL)
9960

    
9961
  @staticmethod
9962
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
9963
    """Compute a new secondary node using an IAllocator.
9964

9965
    """
9966
    ial = IAllocator(lu.cfg, lu.rpc,
9967
                     mode=constants.IALLOCATOR_MODE_RELOC,
9968
                     name=instance_name,
9969
                     relocate_from=list(relocate_from))
9970

    
9971
    ial.Run(iallocator_name)
9972

    
9973
    if not ial.success:
9974
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
9975
                                 " %s" % (iallocator_name, ial.info),
9976
                                 errors.ECODE_NORES)
9977

    
9978
    if len(ial.result) != ial.required_nodes:
9979
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
9980
                                 " of nodes (%s), required %s" %
9981
                                 (iallocator_name,
9982
                                  len(ial.result), ial.required_nodes),
9983
                                 errors.ECODE_FAULT)
9984

    
9985
    remote_node_name = ial.result[0]
9986

    
9987
    lu.LogInfo("Selected new secondary for instance '%s': %s",
9988
               instance_name, remote_node_name)
9989

    
9990
    return remote_node_name
9991

    
9992
  def _FindFaultyDisks(self, node_name):
9993
    """Wrapper for L{_FindFaultyInstanceDisks}.
9994

9995
    """
9996
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
9997
                                    node_name, True)
9998

    
9999
  def _CheckDisksActivated(self, instance):
10000
    """Checks if the instance disks are activated.
10001

10002
    @param instance: The instance to check disks
10003
    @return: True if they are activated, False otherwise
10004

10005
    """
10006
    nodes = instance.all_nodes
10007

    
10008
    for idx, dev in enumerate(instance.disks):
10009
      for node in nodes:
10010
        self.lu.LogInfo("Checking disk/%d on %s", idx, node)
10011
        self.cfg.SetDiskID(dev, node)
10012

    
10013
        result = self.rpc.call_blockdev_find(node, dev)
10014

    
10015
        if result.offline:
10016
          continue
10017
        elif result.fail_msg or not result.payload:
10018
          return False
10019

    
10020
    return True
10021

    
10022
  def CheckPrereq(self):
10023
    """Check prerequisites.
10024

10025
    This checks that the instance is in the cluster.
10026

10027
    """
10028
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
10029
    assert instance is not None, \
10030
      "Cannot retrieve locked instance %s" % self.instance_name
10031

    
10032
    if instance.disk_template != constants.DT_DRBD8:
10033
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
10034
                                 " instances", errors.ECODE_INVAL)
10035

    
10036
    if len(instance.secondary_nodes) != 1:
10037
      raise errors.OpPrereqError("The instance has a strange layout,"
10038
                                 " expected one secondary but found %d" %
10039
                                 len(instance.secondary_nodes),
10040
                                 errors.ECODE_FAULT)
10041

    
10042
    if not self.delay_iallocator:
10043
      self._CheckPrereq2()
10044

    
10045
  def _CheckPrereq2(self):
10046
    """Check prerequisites, second part.
10047

10048
    This function should always be part of CheckPrereq. It was separated and is
10049
    now called from Exec because during node evacuation iallocator was only
10050
    called with an unmodified cluster model, not taking planned changes into
10051
    account.
10052

10053
    """
10054
    instance = self.instance
10055
    secondary_node = instance.secondary_nodes[0]
10056

    
10057
    if self.iallocator_name is None:
10058
      remote_node = self.remote_node
10059
    else:
10060
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
10061
                                       instance.name, instance.secondary_nodes)
10062

    
10063
    if remote_node is None:
10064
      self.remote_node_info = None
10065
    else:
10066
      assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
10067
             "Remote node '%s' is not locked" % remote_node
10068

    
10069
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
10070
      assert self.remote_node_info is not None, \
10071
        "Cannot retrieve locked node %s" % remote_node
10072

    
10073
    if remote_node == self.instance.primary_node:
10074
      raise errors.OpPrereqError("The specified node is the primary node of"
10075
                                 " the instance", errors.ECODE_INVAL)
10076

    
10077
    if remote_node == secondary_node:
10078
      raise errors.OpPrereqError("The specified node is already the"
10079
                                 " secondary node of the instance",
10080
                                 errors.ECODE_INVAL)
10081

    
10082
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
10083
                                    constants.REPLACE_DISK_CHG):
10084
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
10085
                                 errors.ECODE_INVAL)
10086

    
10087
    if self.mode == constants.REPLACE_DISK_AUTO:
10088
      if not self._CheckDisksActivated(instance):
10089
        raise errors.OpPrereqError("Please run activate-disks on instance %s"
10090
                                   " first" % self.instance_name,
10091
                                   errors.ECODE_STATE)
10092
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
10093
      faulty_secondary = self._FindFaultyDisks(secondary_node)
10094

    
10095
      if faulty_primary and faulty_secondary:
10096
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
10097
                                   " one node and can not be repaired"
10098
                                   " automatically" % self.instance_name,
10099
                                   errors.ECODE_STATE)
10100

    
10101
      if faulty_primary:
10102
        self.disks = faulty_primary
10103
        self.target_node = instance.primary_node
10104
        self.other_node = secondary_node
10105
        check_nodes = [self.target_node, self.other_node]
10106
      elif faulty_secondary:
10107
        self.disks = faulty_secondary
10108
        self.target_node = secondary_node
10109
        self.other_node = instance.primary_node
10110
        check_nodes = [self.target_node, self.other_node]
10111
      else:
10112
        self.disks = []
10113
        check_nodes = []
10114

    
10115
    else:
10116
      # Non-automatic modes
10117
      if self.mode == constants.REPLACE_DISK_PRI:
10118
        self.target_node = instance.primary_node
10119
        self.other_node = secondary_node
10120
        check_nodes = [self.target_node, self.other_node]
10121

    
10122
      elif self.mode == constants.REPLACE_DISK_SEC:
10123
        self.target_node = secondary_node
10124
        self.other_node = instance.primary_node
10125
        check_nodes = [self.target_node, self.other_node]
10126

    
10127
      elif self.mode == constants.REPLACE_DISK_CHG:
10128
        self.new_node = remote_node
10129
        self.other_node = instance.primary_node
10130
        self.target_node = secondary_node
10131
        check_nodes = [self.new_node, self.other_node]
10132

    
10133
        _CheckNodeNotDrained(self.lu, remote_node)
10134
        _CheckNodeVmCapable(self.lu, remote_node)
10135

    
10136
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
10137
        assert old_node_info is not None
10138
        if old_node_info.offline and not self.early_release:
10139
          # doesn't make sense to delay the release
10140
          self.early_release = True
10141
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
10142
                          " early-release mode", secondary_node)
10143

    
10144
      else:
10145
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
10146
                                     self.mode)
10147

    
10148
      # If not specified all disks should be replaced
10149
      if not self.disks:
10150
        self.disks = range(len(self.instance.disks))
10151

    
10152
    # TODO: compute disk parameters
10153
    primary_node_info = self.cfg.GetNodeInfo(instance.primary_node)
10154
    secondary_node_info = self.cfg.GetNodeInfo(secondary_node)
10155
    if primary_node_info.group != secondary_node_info.group:
10156
      self.lu.LogInfo("The instance primary and secondary nodes are in two"
10157
                      " different node groups; the disk parameters of the"
10158
                      " primary node's group will be applied.")
10159

    
10160
    self.diskparams = self.cfg.GetNodeGroup(primary_node_info.group).diskparams
10161

    
10162
    for node in check_nodes:
10163
      _CheckNodeOnline(self.lu, node)
10164

    
10165
    touched_nodes = frozenset(node_name for node_name in [self.new_node,
10166
                                                          self.other_node,
10167
                                                          self.target_node]
10168
                              if node_name is not None)
10169

    
10170
    # Release unneeded node and node resource locks
10171
    _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
10172
    _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes)
10173

    
10174
    # Release any owned node group
10175
    if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
10176
      _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
10177

    
10178
    # Check whether disks are valid
10179
    for disk_idx in self.disks:
10180
      instance.FindDisk(disk_idx)
10181

    
10182
    # Get secondary node IP addresses
10183
    self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
10184
                                  in self.cfg.GetMultiNodeInfo(touched_nodes))
10185

    
10186
  def Exec(self, feedback_fn):
10187
    """Execute disk replacement.
10188

10189
    This dispatches the disk replacement to the appropriate handler.
10190

10191
    """
10192
    if self.delay_iallocator:
10193
      self._CheckPrereq2()
10194

    
10195
    if __debug__:
10196
      # Verify owned locks before starting operation
10197
      owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
10198
      assert set(owned_nodes) == set(self.node_secondary_ip), \
10199
          ("Incorrect node locks, owning %s, expected %s" %
10200
           (owned_nodes, self.node_secondary_ip.keys()))
10201
      assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
10202
              self.lu.owned_locks(locking.LEVEL_NODE_RES))
10203

    
10204
      owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
10205
      assert list(owned_instances) == [self.instance_name], \
10206
          "Instance '%s' not locked" % self.instance_name
10207

    
10208
      assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
10209
          "Should not own any node group lock at this point"
10210

    
10211
    if not self.disks:
10212
      feedback_fn("No disks need replacement")
10213
      return
10214

    
10215
    feedback_fn("Replacing disk(s) %s for %s" %
10216
                (utils.CommaJoin(self.disks), self.instance.name))
10217

    
10218
    activate_disks = (self.instance.admin_state != constants.ADMINST_UP)
10219

    
10220
    # Activate the instance disks if we're replacing them on a down instance
10221
    if activate_disks:
10222
      _StartInstanceDisks(self.lu, self.instance, True)
10223

    
10224
    try:
10225
      # Should we replace the secondary node?
10226
      if self.new_node is not None:
10227
        fn = self._ExecDrbd8Secondary
10228
      else:
10229
        fn = self._ExecDrbd8DiskOnly
10230

    
10231
      result = fn(feedback_fn)
10232
    finally:
10233
      # Deactivate the instance disks if we're replacing them on a
10234
      # down instance
10235
      if activate_disks:
10236
        _SafeShutdownInstanceDisks(self.lu, self.instance)
10237

    
10238
    assert not self.lu.owned_locks(locking.LEVEL_NODE)
10239

    
10240
    if __debug__:
10241
      # Verify owned locks
10242
      owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE_RES)
10243
      nodes = frozenset(self.node_secondary_ip)
10244
      assert ((self.early_release and not owned_nodes) or
10245
              (not self.early_release and not (set(owned_nodes) - nodes))), \
10246
        ("Not owning the correct locks, early_release=%s, owned=%r,"
10247
         " nodes=%r" % (self.early_release, owned_nodes, nodes))
10248

    
10249
    return result
10250

    
10251
  def _CheckVolumeGroup(self, nodes):
10252
    self.lu.LogInfo("Checking volume groups")
10253

    
10254
    vgname = self.cfg.GetVGName()
10255

    
10256
    # Make sure volume group exists on all involved nodes
10257
    results = self.rpc.call_vg_list(nodes)
10258
    if not results:
10259
      raise errors.OpExecError("Can't list volume groups on the nodes")
10260

    
10261
    for node in nodes:
10262
      res = results[node]
10263
      res.Raise("Error checking node %s" % node)
10264
      if vgname not in res.payload:
10265
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
10266
                                 (vgname, node))
10267

    
10268
  def _CheckDisksExistence(self, nodes):
10269
    # Check disk existence
10270
    for idx, dev in enumerate(self.instance.disks):
10271
      if idx not in self.disks:
10272
        continue
10273

    
10274
      for node in nodes:
10275
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
10276
        self.cfg.SetDiskID(dev, node)
10277

    
10278
        result = self.rpc.call_blockdev_find(node, dev)
10279

    
10280
        msg = result.fail_msg
10281
        if msg or not result.payload:
10282
          if not msg:
10283
            msg = "disk not found"
10284
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
10285
                                   (idx, node, msg))
10286

    
10287
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
10288
    for idx, dev in enumerate(self.instance.disks):
10289
      if idx not in self.disks:
10290
        continue
10291

    
10292
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
10293
                      (idx, node_name))
10294

    
10295
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
10296
                                   ldisk=ldisk):
10297
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
10298
                                 " replace disks for instance %s" %
10299
                                 (node_name, self.instance.name))
10300

    
10301
  def _CreateNewStorage(self, node_name):
10302
    """Create new storage on the primary or secondary node.
10303

10304
    This is only used for same-node replaces, not for changing the
10305
    secondary node, hence we don't want to modify the existing disk.
10306

10307
    """
10308
    iv_names = {}
10309

    
10310
    for idx, dev in enumerate(self.instance.disks):
10311
      if idx not in self.disks:
10312
        continue
10313

    
10314
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
10315

    
10316
      self.cfg.SetDiskID(dev, node_name)
10317

    
10318
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
10319
      names = _GenerateUniqueNames(self.lu, lv_names)
10320

    
10321
      _, data_p, meta_p = _ComputeLDParams(constants.DT_DRBD8, self.diskparams)
10322

    
10323
      vg_data = dev.children[0].logical_id[0]
10324
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
10325
                             logical_id=(vg_data, names[0]), params=data_p)
10326
      vg_meta = dev.children[1].logical_id[0]
10327
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
10328
                             logical_id=(vg_meta, names[1]), params=meta_p)
10329

    
10330
      new_lvs = [lv_data, lv_meta]
10331
      old_lvs = [child.Copy() for child in dev.children]
10332
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
10333

    
10334
      # we pass force_create=True to force the LVM creation
10335
      for new_lv in new_lvs:
10336
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
10337
                        _GetInstanceInfoText(self.instance), False)
10338

    
10339
    return iv_names
10340

    
10341
  def _CheckDevices(self, node_name, iv_names):
10342
    for name, (dev, _, _) in iv_names.iteritems():
10343
      self.cfg.SetDiskID(dev, node_name)
10344

    
10345
      result = self.rpc.call_blockdev_find(node_name, dev)
10346

    
10347
      msg = result.fail_msg
10348
      if msg or not result.payload:
10349
        if not msg:
10350
          msg = "disk not found"
10351
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
10352
                                 (name, msg))
10353

    
10354
      if result.payload.is_degraded:
10355
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
10356

    
10357
  def _RemoveOldStorage(self, node_name, iv_names):
10358
    for name, (_, old_lvs, _) in iv_names.iteritems():
10359
      self.lu.LogInfo("Remove logical volumes for %s" % name)
10360

    
10361
      for lv in old_lvs:
10362
        self.cfg.SetDiskID(lv, node_name)
10363

    
10364
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
10365
        if msg:
10366
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
10367
                             hint="remove unused LVs manually")
10368

    
10369
  def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
10370
    """Replace a disk on the primary or secondary for DRBD 8.
10371

10372
    The algorithm for replace is quite complicated:
10373

10374
      1. for each disk to be replaced:
10375

10376
        1. create new LVs on the target node with unique names
10377
        1. detach old LVs from the drbd device
10378
        1. rename old LVs to name_replaced.<time_t>
10379
        1. rename new LVs to old LVs
10380
        1. attach the new LVs (with the old names now) to the drbd device
10381

10382
      1. wait for sync across all devices
10383

10384
      1. for each modified disk:
10385

10386
        1. remove old LVs (which have the name name_replaces.<time_t>)
10387

10388
    Failures are not very well handled.
10389

10390
    """
10391
    steps_total = 6
10392

    
10393
    # Step: check device activation
10394
    self.lu.LogStep(1, steps_total, "Check device existence")
10395
    self._CheckDisksExistence([self.other_node, self.target_node])
10396
    self._CheckVolumeGroup([self.target_node, self.other_node])
10397

    
10398
    # Step: check other node consistency
10399
    self.lu.LogStep(2, steps_total, "Check peer consistency")
10400
    self._CheckDisksConsistency(self.other_node,
10401
                                self.other_node == self.instance.primary_node,
10402
                                False)
10403

    
10404
    # Step: create new storage
10405
    self.lu.LogStep(3, steps_total, "Allocate new storage")
10406
    iv_names = self._CreateNewStorage(self.target_node)
10407

    
10408
    # Step: for each lv, detach+rename*2+attach
10409
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
10410
    for dev, old_lvs, new_lvs in iv_names.itervalues():
10411
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
10412

    
10413
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
10414
                                                     old_lvs)
10415
      result.Raise("Can't detach drbd from local storage on node"
10416
                   " %s for device %s" % (self.target_node, dev.iv_name))
10417
      #dev.children = []
10418
      #cfg.Update(instance)
10419

    
10420
      # ok, we created the new LVs, so now we know we have the needed
10421
      # storage; as such, we proceed on the target node to rename
10422
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
10423
      # using the assumption that logical_id == physical_id (which in
10424
      # turn is the unique_id on that node)
10425

    
10426
      # FIXME(iustin): use a better name for the replaced LVs
10427
      temp_suffix = int(time.time())
10428
      ren_fn = lambda d, suff: (d.physical_id[0],
10429
                                d.physical_id[1] + "_replaced-%s" % suff)
10430

    
10431
      # Build the rename list based on what LVs exist on the node
10432
      rename_old_to_new = []
10433
      for to_ren in old_lvs:
10434
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
10435
        if not result.fail_msg and result.payload:
10436
          # device exists
10437
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
10438

    
10439
      self.lu.LogInfo("Renaming the old LVs on the target node")
10440
      result = self.rpc.call_blockdev_rename(self.target_node,
10441
                                             rename_old_to_new)
10442
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
10443

    
10444
      # Now we rename the new LVs to the old LVs
10445
      self.lu.LogInfo("Renaming the new LVs on the target node")
10446
      rename_new_to_old = [(new, old.physical_id)
10447
                           for old, new in zip(old_lvs, new_lvs)]
10448
      result = self.rpc.call_blockdev_rename(self.target_node,
10449
                                             rename_new_to_old)
10450
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
10451

    
10452
      # Intermediate steps of in memory modifications
10453
      for old, new in zip(old_lvs, new_lvs):
10454
        new.logical_id = old.logical_id
10455
        self.cfg.SetDiskID(new, self.target_node)
10456

    
10457
      # We need to modify old_lvs so that removal later removes the
10458
      # right LVs, not the newly added ones; note that old_lvs is a
10459
      # copy here
10460
      for disk in old_lvs:
10461
        disk.logical_id = ren_fn(disk, temp_suffix)
10462
        self.cfg.SetDiskID(disk, self.target_node)
10463

    
10464
      # Now that the new lvs have the old name, we can add them to the device
10465
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
10466
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
10467
                                                  new_lvs)
10468
      msg = result.fail_msg
10469
      if msg:
10470
        for new_lv in new_lvs:
10471
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
10472
                                               new_lv).fail_msg
10473
          if msg2:
10474
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
10475
                               hint=("cleanup manually the unused logical"
10476
                                     "volumes"))
10477
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
10478

    
10479
    cstep = itertools.count(5)
10480

    
10481
    if self.early_release:
10482
      self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
10483
      self._RemoveOldStorage(self.target_node, iv_names)
10484
      # TODO: Check if releasing locks early still makes sense
10485
      _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
10486
    else:
10487
      # Release all resource locks except those used by the instance
10488
      _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
10489
                    keep=self.node_secondary_ip.keys())
10490

    
10491
    # Release all node locks while waiting for sync
10492
    _ReleaseLocks(self.lu, locking.LEVEL_NODE)
10493

    
10494
    # TODO: Can the instance lock be downgraded here? Take the optional disk
10495
    # shutdown in the caller into consideration.
10496

    
10497
    # Wait for sync
10498
    # This can fail as the old devices are degraded and _WaitForSync
10499
    # does a combined result over all disks, so we don't check its return value
10500
    self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
10501
    _WaitForSync(self.lu, self.instance)
10502

    
10503
    # Check all devices manually
10504
    self._CheckDevices(self.instance.primary_node, iv_names)
10505

    
10506
    # Step: remove old storage
10507
    if not self.early_release:
10508
      self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
10509
      self._RemoveOldStorage(self.target_node, iv_names)
10510

    
10511
  def _ExecDrbd8Secondary(self, feedback_fn):
10512
    """Replace the secondary node for DRBD 8.
10513

10514
    The algorithm for replace is quite complicated:
10515
      - for all disks of the instance:
10516
        - create new LVs on the new node with same names
10517
        - shutdown the drbd device on the old secondary
10518
        - disconnect the drbd network on the primary
10519
        - create the drbd device on the new secondary
10520
        - network attach the drbd on the primary, using an artifice:
10521
          the drbd code for Attach() will connect to the network if it
10522
          finds a device which is connected to the good local disks but
10523
          not network enabled
10524
      - wait for sync across all devices
10525
      - remove all disks from the old secondary
10526

10527
    Failures are not very well handled.
10528

10529
    """
10530
    steps_total = 6
10531

    
10532
    pnode = self.instance.primary_node
10533

    
10534
    # Step: check device activation
10535
    self.lu.LogStep(1, steps_total, "Check device existence")
10536
    self._CheckDisksExistence([self.instance.primary_node])
10537
    self._CheckVolumeGroup([self.instance.primary_node])
10538

    
10539
    # Step: check other node consistency
10540
    self.lu.LogStep(2, steps_total, "Check peer consistency")
10541
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
10542

    
10543
    # Step: create new storage
10544
    self.lu.LogStep(3, steps_total, "Allocate new storage")
10545
    for idx, dev in enumerate(self.instance.disks):
10546
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
10547
                      (self.new_node, idx))
10548
      # we pass force_create=True to force LVM creation
10549
      for new_lv in dev.children:
10550
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
10551
                        _GetInstanceInfoText(self.instance), False)
10552

    
10553
    # Step 4: dbrd minors and drbd setups changes
10554
    # after this, we must manually remove the drbd minors on both the
10555
    # error and the success paths
10556
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
10557
    minors = self.cfg.AllocateDRBDMinor([self.new_node
10558
                                         for dev in self.instance.disks],
10559
                                        self.instance.name)
10560
    logging.debug("Allocated minors %r", minors)
10561

    
10562
    iv_names = {}
10563
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
10564
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
10565
                      (self.new_node, idx))
10566
      # create new devices on new_node; note that we create two IDs:
10567
      # one without port, so the drbd will be activated without
10568
      # networking information on the new node at this stage, and one
10569
      # with network, for the latter activation in step 4
10570
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
10571
      if self.instance.primary_node == o_node1:
10572
        p_minor = o_minor1
10573
      else:
10574
        assert self.instance.primary_node == o_node2, "Three-node instance?"
10575
        p_minor = o_minor2
10576

    
10577
      new_alone_id = (self.instance.primary_node, self.new_node, None,
10578
                      p_minor, new_minor, o_secret)
10579
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
10580
                    p_minor, new_minor, o_secret)
10581

    
10582
      iv_names[idx] = (dev, dev.children, new_net_id)
10583
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
10584
                    new_net_id)
10585
      drbd_params, _, _ = _ComputeLDParams(constants.DT_DRBD8, self.diskparams)
10586
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
10587
                              logical_id=new_alone_id,
10588
                              children=dev.children,
10589
                              size=dev.size,
10590
                              params=drbd_params)
10591
      try:
10592
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
10593
                              _GetInstanceInfoText(self.instance), False)
10594
      except errors.GenericError:
10595
        self.cfg.ReleaseDRBDMinors(self.instance.name)
10596
        raise
10597

    
10598
    # We have new devices, shutdown the drbd on the old secondary
10599
    for idx, dev in enumerate(self.instance.disks):
10600
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
10601
      self.cfg.SetDiskID(dev, self.target_node)
10602
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
10603
      if msg:
10604
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
10605
                           "node: %s" % (idx, msg),
10606
                           hint=("Please cleanup this device manually as"
10607
                                 " soon as possible"))
10608

    
10609
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
10610
    result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
10611
                                               self.instance.disks)[pnode]
10612

    
10613
    msg = result.fail_msg
10614
    if msg:
10615
      # detaches didn't succeed (unlikely)
10616
      self.cfg.ReleaseDRBDMinors(self.instance.name)
10617
      raise errors.OpExecError("Can't detach the disks from the network on"
10618
                               " old node: %s" % (msg,))
10619

    
10620
    # if we managed to detach at least one, we update all the disks of
10621
    # the instance to point to the new secondary
10622
    self.lu.LogInfo("Updating instance configuration")
10623
    for dev, _, new_logical_id in iv_names.itervalues():
10624
      dev.logical_id = new_logical_id
10625
      self.cfg.SetDiskID(dev, self.instance.primary_node)
10626

    
10627
    self.cfg.Update(self.instance, feedback_fn)
10628

    
10629
    # Release all node locks (the configuration has been updated)
10630
    _ReleaseLocks(self.lu, locking.LEVEL_NODE)
10631

    
10632
    # and now perform the drbd attach
10633
    self.lu.LogInfo("Attaching primary drbds to new secondary"
10634
                    " (standalone => connected)")
10635
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
10636
                                            self.new_node],
10637
                                           self.node_secondary_ip,
10638
                                           self.instance.disks,
10639
                                           self.instance.name,
10640
                                           False)
10641
    for to_node, to_result in result.items():
10642
      msg = to_result.fail_msg
10643
      if msg:
10644
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
10645
                           to_node, msg,
10646
                           hint=("please do a gnt-instance info to see the"
10647
                                 " status of disks"))
10648

    
10649
    cstep = itertools.count(5)
10650

    
10651
    if self.early_release:
10652
      self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
10653
      self._RemoveOldStorage(self.target_node, iv_names)
10654
      # TODO: Check if releasing locks early still makes sense
10655
      _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
10656
    else:
10657
      # Release all resource locks except those used by the instance
10658
      _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
10659
                    keep=self.node_secondary_ip.keys())
10660

    
10661
    # TODO: Can the instance lock be downgraded here? Take the optional disk
10662
    # shutdown in the caller into consideration.
10663

    
10664
    # Wait for sync
10665
    # This can fail as the old devices are degraded and _WaitForSync
10666
    # does a combined result over all disks, so we don't check its return value
10667
    self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
10668
    _WaitForSync(self.lu, self.instance)
10669

    
10670
    # Check all devices manually
10671
    self._CheckDevices(self.instance.primary_node, iv_names)
10672

    
10673
    # Step: remove old storage
10674
    if not self.early_release:
10675
      self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
10676
      self._RemoveOldStorage(self.target_node, iv_names)
10677

    
10678

    
10679
class LURepairNodeStorage(NoHooksLU):
10680
  """Repairs the volume group on a node.
10681

10682
  """
10683
  REQ_BGL = False
10684

    
10685
  def CheckArguments(self):
10686
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
10687

    
10688
    storage_type = self.op.storage_type
10689

    
10690
    if (constants.SO_FIX_CONSISTENCY not in
10691
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
10692
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
10693
                                 " repaired" % storage_type,
10694
                                 errors.ECODE_INVAL)
10695

    
10696
  def ExpandNames(self):
10697
    self.needed_locks = {
10698
      locking.LEVEL_NODE: [self.op.node_name],
10699
      }
10700

    
10701
  def _CheckFaultyDisks(self, instance, node_name):
10702
    """Ensure faulty disks abort the opcode or at least warn."""
10703
    try:
10704
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
10705
                                  node_name, True):
10706
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
10707
                                   " node '%s'" % (instance.name, node_name),
10708
                                   errors.ECODE_STATE)
10709
    except errors.OpPrereqError, err:
10710
      if self.op.ignore_consistency:
10711
        self.proc.LogWarning(str(err.args[0]))
10712
      else:
10713
        raise
10714

    
10715
  def CheckPrereq(self):
10716
    """Check prerequisites.
10717

10718
    """
10719
    # Check whether any instance on this node has faulty disks
10720
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
10721
      if inst.admin_state != constants.ADMINST_UP:
10722
        continue
10723
      check_nodes = set(inst.all_nodes)
10724
      check_nodes.discard(self.op.node_name)
10725
      for inst_node_name in check_nodes:
10726
        self._CheckFaultyDisks(inst, inst_node_name)
10727

    
10728
  def Exec(self, feedback_fn):
10729
    feedback_fn("Repairing storage unit '%s' on %s ..." %
10730
                (self.op.name, self.op.node_name))
10731

    
10732
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
10733
    result = self.rpc.call_storage_execute(self.op.node_name,
10734
                                           self.op.storage_type, st_args,
10735
                                           self.op.name,
10736
                                           constants.SO_FIX_CONSISTENCY)
10737
    result.Raise("Failed to repair storage unit '%s' on %s" %
10738
                 (self.op.name, self.op.node_name))
10739

    
10740

    
10741
class LUNodeEvacuate(NoHooksLU):
10742
  """Evacuates instances off a list of nodes.
10743

10744
  """
10745
  REQ_BGL = False
10746

    
10747
  _MODE2IALLOCATOR = {
10748
    constants.NODE_EVAC_PRI: constants.IALLOCATOR_NEVAC_PRI,
10749
    constants.NODE_EVAC_SEC: constants.IALLOCATOR_NEVAC_SEC,
10750
    constants.NODE_EVAC_ALL: constants.IALLOCATOR_NEVAC_ALL,
10751
    }
10752
  assert frozenset(_MODE2IALLOCATOR.keys()) == constants.NODE_EVAC_MODES
10753
  assert (frozenset(_MODE2IALLOCATOR.values()) ==
10754
          constants.IALLOCATOR_NEVAC_MODES)
10755

    
10756
  def CheckArguments(self):
10757
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
10758

    
10759
  def ExpandNames(self):
10760
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
10761

    
10762
    if self.op.remote_node is not None:
10763
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
10764
      assert self.op.remote_node
10765

    
10766
      if self.op.remote_node == self.op.node_name:
10767
        raise errors.OpPrereqError("Can not use evacuated node as a new"
10768
                                   " secondary node", errors.ECODE_INVAL)
10769

    
10770
      if self.op.mode != constants.NODE_EVAC_SEC:
10771
        raise errors.OpPrereqError("Without the use of an iallocator only"
10772
                                   " secondary instances can be evacuated",
10773
                                   errors.ECODE_INVAL)
10774

    
10775
    # Declare locks
10776
    self.share_locks = _ShareAll()
10777
    self.needed_locks = {
10778
      locking.LEVEL_INSTANCE: [],
10779
      locking.LEVEL_NODEGROUP: [],
10780
      locking.LEVEL_NODE: [],
10781
      }
10782

    
10783
    # Determine nodes (via group) optimistically, needs verification once locks
10784
    # have been acquired
10785
    self.lock_nodes = self._DetermineNodes()
10786

    
10787
  def _DetermineNodes(self):
10788
    """Gets the list of nodes to operate on.
10789

10790
    """
10791
    if self.op.remote_node is None:
10792
      # Iallocator will choose any node(s) in the same group
10793
      group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
10794
    else:
10795
      group_nodes = frozenset([self.op.remote_node])
10796

    
10797
    # Determine nodes to be locked
10798
    return set([self.op.node_name]) | group_nodes
10799

    
10800
  def _DetermineInstances(self):
10801
    """Builds list of instances to operate on.
10802

10803
    """
10804
    assert self.op.mode in constants.NODE_EVAC_MODES
10805

    
10806
    if self.op.mode == constants.NODE_EVAC_PRI:
10807
      # Primary instances only
10808
      inst_fn = _GetNodePrimaryInstances
10809
      assert self.op.remote_node is None, \
10810
        "Evacuating primary instances requires iallocator"
10811
    elif self.op.mode == constants.NODE_EVAC_SEC:
10812
      # Secondary instances only
10813
      inst_fn = _GetNodeSecondaryInstances
10814
    else:
10815
      # All instances
10816
      assert self.op.mode == constants.NODE_EVAC_ALL
10817
      inst_fn = _GetNodeInstances
10818
      # TODO: In 2.6, change the iallocator interface to take an evacuation mode
10819
      # per instance
10820
      raise errors.OpPrereqError("Due to an issue with the iallocator"
10821
                                 " interface it is not possible to evacuate"
10822
                                 " all instances at once; specify explicitly"
10823
                                 " whether to evacuate primary or secondary"
10824
                                 " instances",
10825
                                 errors.ECODE_INVAL)
10826

    
10827
    return inst_fn(self.cfg, self.op.node_name)
10828

    
10829
  def DeclareLocks(self, level):
10830
    if level == locking.LEVEL_INSTANCE:
10831
      # Lock instances optimistically, needs verification once node and group
10832
      # locks have been acquired
10833
      self.needed_locks[locking.LEVEL_INSTANCE] = \
10834
        set(i.name for i in self._DetermineInstances())
10835

    
10836
    elif level == locking.LEVEL_NODEGROUP:
10837
      # Lock node groups for all potential target nodes optimistically, needs
10838
      # verification once nodes have been acquired
10839
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
10840
        self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
10841

    
10842
    elif level == locking.LEVEL_NODE:
10843
      self.needed_locks[locking.LEVEL_NODE] = self.lock_nodes
10844

    
10845
  def CheckPrereq(self):
10846
    # Verify locks
10847
    owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
10848
    owned_nodes = self.owned_locks(locking.LEVEL_NODE)
10849
    owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
10850

    
10851
    need_nodes = self._DetermineNodes()
10852

    
10853
    if not owned_nodes.issuperset(need_nodes):
10854
      raise errors.OpPrereqError("Nodes in same group as '%s' changed since"
10855
                                 " locks were acquired, current nodes are"
10856
                                 " are '%s', used to be '%s'; retry the"
10857
                                 " operation" %
10858
                                 (self.op.node_name,
10859
                                  utils.CommaJoin(need_nodes),
10860
                                  utils.CommaJoin(owned_nodes)),
10861
                                 errors.ECODE_STATE)
10862

    
10863
    wanted_groups = self.cfg.GetNodeGroupsFromNodes(owned_nodes)
10864
    if owned_groups != wanted_groups:
10865
      raise errors.OpExecError("Node groups changed since locks were acquired,"
10866
                               " current groups are '%s', used to be '%s';"
10867
                               " retry the operation" %
10868
                               (utils.CommaJoin(wanted_groups),
10869
                                utils.CommaJoin(owned_groups)))
10870

    
10871
    # Determine affected instances
10872
    self.instances = self._DetermineInstances()
10873
    self.instance_names = [i.name for i in self.instances]
10874

    
10875
    if set(self.instance_names) != owned_instances:
10876
      raise errors.OpExecError("Instances on node '%s' changed since locks"
10877
                               " were acquired, current instances are '%s',"
10878
                               " used to be '%s'; retry the operation" %
10879
                               (self.op.node_name,
10880
                                utils.CommaJoin(self.instance_names),
10881
                                utils.CommaJoin(owned_instances)))
10882

    
10883
    if self.instance_names:
10884
      self.LogInfo("Evacuating instances from node '%s': %s",
10885
                   self.op.node_name,
10886
                   utils.CommaJoin(utils.NiceSort(self.instance_names)))
10887
    else:
10888
      self.LogInfo("No instances to evacuate from node '%s'",
10889
                   self.op.node_name)
10890

    
10891
    if self.op.remote_node is not None:
10892
      for i in self.instances:
10893
        if i.primary_node == self.op.remote_node:
10894
          raise errors.OpPrereqError("Node %s is the primary node of"
10895
                                     " instance %s, cannot use it as"
10896
                                     " secondary" %
10897
                                     (self.op.remote_node, i.name),
10898
                                     errors.ECODE_INVAL)
10899

    
10900
  def Exec(self, feedback_fn):
10901
    assert (self.op.iallocator is not None) ^ (self.op.remote_node is not None)
10902

    
10903
    if not self.instance_names:
10904
      # No instances to evacuate
10905
      jobs = []
10906

    
10907
    elif self.op.iallocator is not None:
10908
      # TODO: Implement relocation to other group
10909
      ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_NODE_EVAC,
10910
                       evac_mode=self._MODE2IALLOCATOR[self.op.mode],
10911
                       instances=list(self.instance_names))
10912

    
10913
      ial.Run(self.op.iallocator)
10914

    
10915
      if not ial.success:
10916
        raise errors.OpPrereqError("Can't compute node evacuation using"
10917
                                   " iallocator '%s': %s" %
10918
                                   (self.op.iallocator, ial.info),
10919
                                   errors.ECODE_NORES)
10920

    
10921
      jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, True)
10922

    
10923
    elif self.op.remote_node is not None:
10924
      assert self.op.mode == constants.NODE_EVAC_SEC
10925
      jobs = [
10926
        [opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
10927
                                        remote_node=self.op.remote_node,
10928
                                        disks=[],
10929
                                        mode=constants.REPLACE_DISK_CHG,
10930
                                        early_release=self.op.early_release)]
10931
        for instance_name in self.instance_names
10932
        ]
10933

    
10934
    else:
10935
      raise errors.ProgrammerError("No iallocator or remote node")
10936

    
10937
    return ResultWithJobs(jobs)
10938

    
10939

    
10940
def _SetOpEarlyRelease(early_release, op):
10941
  """Sets C{early_release} flag on opcodes if available.
10942

10943
  """
10944
  try:
10945
    op.early_release = early_release
10946
  except AttributeError:
10947
    assert not isinstance(op, opcodes.OpInstanceReplaceDisks)
10948

    
10949
  return op
10950

    
10951

    
10952
def _NodeEvacDest(use_nodes, group, nodes):
10953
  """Returns group or nodes depending on caller's choice.
10954

10955
  """
10956
  if use_nodes:
10957
    return utils.CommaJoin(nodes)
10958
  else:
10959
    return group
10960

    
10961

    
10962
def _LoadNodeEvacResult(lu, alloc_result, early_release, use_nodes):
10963
  """Unpacks the result of change-group and node-evacuate iallocator requests.
10964

10965
  Iallocator modes L{constants.IALLOCATOR_MODE_NODE_EVAC} and
10966
  L{constants.IALLOCATOR_MODE_CHG_GROUP}.
10967

10968
  @type lu: L{LogicalUnit}
10969
  @param lu: Logical unit instance
10970
  @type alloc_result: tuple/list
10971
  @param alloc_result: Result from iallocator
10972
  @type early_release: bool
10973
  @param early_release: Whether to release locks early if possible
10974
  @type use_nodes: bool
10975
  @param use_nodes: Whether to display node names instead of groups
10976

10977
  """
10978
  (moved, failed, jobs) = alloc_result
10979

    
10980
  if failed:
10981
    failreason = utils.CommaJoin("%s (%s)" % (name, reason)
10982
                                 for (name, reason) in failed)
10983
    lu.LogWarning("Unable to evacuate instances %s", failreason)
10984
    raise errors.OpExecError("Unable to evacuate instances %s" % failreason)
10985

    
10986
  if moved:
10987
    lu.LogInfo("Instances to be moved: %s",
10988
               utils.CommaJoin("%s (to %s)" %
10989
                               (name, _NodeEvacDest(use_nodes, group, nodes))
10990
                               for (name, group, nodes) in moved))
10991

    
10992
  return [map(compat.partial(_SetOpEarlyRelease, early_release),
10993
              map(opcodes.OpCode.LoadOpCode, ops))
10994
          for ops in jobs]
10995

    
10996

    
10997
class LUInstanceGrowDisk(LogicalUnit):
10998
  """Grow a disk of an instance.
10999

11000
  """
11001
  HPATH = "disk-grow"
11002
  HTYPE = constants.HTYPE_INSTANCE
11003
  REQ_BGL = False
11004

    
11005
  def ExpandNames(self):
11006
    self._ExpandAndLockInstance()
11007
    self.needed_locks[locking.LEVEL_NODE] = []
11008
    self.needed_locks[locking.LEVEL_NODE_RES] = []
11009
    self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
11010

    
11011
  def DeclareLocks(self, level):
11012
    if level == locking.LEVEL_NODE:
11013
      self._LockInstancesNodes()
11014
    elif level == locking.LEVEL_NODE_RES:
11015
      # Copy node locks
11016
      self.needed_locks[locking.LEVEL_NODE_RES] = \
11017
        self.needed_locks[locking.LEVEL_NODE][:]
11018

    
11019
  def BuildHooksEnv(self):
11020
    """Build hooks env.
11021

11022
    This runs on the master, the primary and all the secondaries.
11023

11024
    """
11025
    env = {
11026
      "DISK": self.op.disk,
11027
      "AMOUNT": self.op.amount,
11028
      }
11029
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
11030
    return env
11031

    
11032
  def BuildHooksNodes(self):
11033
    """Build hooks nodes.
11034

11035
    """
11036
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
11037
    return (nl, nl)
11038

    
11039
  def CheckPrereq(self):
11040
    """Check prerequisites.
11041

11042
    This checks that the instance is in the cluster.
11043

11044
    """
11045
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
11046
    assert instance is not None, \
11047
      "Cannot retrieve locked instance %s" % self.op.instance_name
11048
    nodenames = list(instance.all_nodes)
11049
    for node in nodenames:
11050
      _CheckNodeOnline(self, node)
11051

    
11052
    self.instance = instance
11053

    
11054
    if instance.disk_template not in constants.DTS_GROWABLE:
11055
      raise errors.OpPrereqError("Instance's disk layout does not support"
11056
                                 " growing", errors.ECODE_INVAL)
11057

    
11058
    self.disk = instance.FindDisk(self.op.disk)
11059

    
11060
    if instance.disk_template not in (constants.DT_FILE,
11061
                                      constants.DT_SHARED_FILE):
11062
      # TODO: check the free disk space for file, when that feature will be
11063
      # supported
11064
      _CheckNodesFreeDiskPerVG(self, nodenames,
11065
                               self.disk.ComputeGrowth(self.op.amount))
11066

    
11067
  def Exec(self, feedback_fn):
11068
    """Execute disk grow.
11069

11070
    """
11071
    instance = self.instance
11072
    disk = self.disk
11073

    
11074
    assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
11075
    assert (self.owned_locks(locking.LEVEL_NODE) ==
11076
            self.owned_locks(locking.LEVEL_NODE_RES))
11077

    
11078
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
11079
    if not disks_ok:
11080
      raise errors.OpExecError("Cannot activate block device to grow")
11081

    
11082
    feedback_fn("Growing disk %s of instance '%s' by %s" %
11083
                (self.op.disk, instance.name,
11084
                 utils.FormatUnit(self.op.amount, "h")))
11085

    
11086
    # First run all grow ops in dry-run mode
11087
    for node in instance.all_nodes:
11088
      self.cfg.SetDiskID(disk, node)
11089
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, True)
11090
      result.Raise("Grow request failed to node %s" % node)
11091

    
11092
    # We know that (as far as we can test) operations across different
11093
    # nodes will succeed, time to run it for real
11094
    for node in instance.all_nodes:
11095
      self.cfg.SetDiskID(disk, node)
11096
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, False)
11097
      result.Raise("Grow request failed to node %s" % node)
11098

    
11099
      # TODO: Rewrite code to work properly
11100
      # DRBD goes into sync mode for a short amount of time after executing the
11101
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
11102
      # calling "resize" in sync mode fails. Sleeping for a short amount of
11103
      # time is a work-around.
11104
      time.sleep(5)
11105

    
11106
    disk.RecordGrow(self.op.amount)
11107
    self.cfg.Update(instance, feedback_fn)
11108

    
11109
    # Changes have been recorded, release node lock
11110
    _ReleaseLocks(self, locking.LEVEL_NODE)
11111

    
11112
    # Downgrade lock while waiting for sync
11113
    self.glm.downgrade(locking.LEVEL_INSTANCE)
11114

    
11115
    if self.op.wait_for_sync:
11116
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
11117
      if disk_abort:
11118
        self.proc.LogWarning("Disk sync-ing has not returned a good"
11119
                             " status; please check the instance")
11120
      if instance.admin_state != constants.ADMINST_UP:
11121
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
11122
    elif instance.admin_state != constants.ADMINST_UP:
11123
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
11124
                           " not supposed to be running because no wait for"
11125
                           " sync mode was requested")
11126

    
11127
    assert self.owned_locks(locking.LEVEL_NODE_RES)
11128
    assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
11129

    
11130

    
11131
class LUInstanceQueryData(NoHooksLU):
11132
  """Query runtime instance data.
11133

11134
  """
11135
  REQ_BGL = False
11136

    
11137
  def ExpandNames(self):
11138
    self.needed_locks = {}
11139

    
11140
    # Use locking if requested or when non-static information is wanted
11141
    if not (self.op.static or self.op.use_locking):
11142
      self.LogWarning("Non-static data requested, locks need to be acquired")
11143
      self.op.use_locking = True
11144

    
11145
    if self.op.instances or not self.op.use_locking:
11146
      # Expand instance names right here
11147
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
11148
    else:
11149
      # Will use acquired locks
11150
      self.wanted_names = None
11151

    
11152
    if self.op.use_locking:
11153
      self.share_locks = _ShareAll()
11154

    
11155
      if self.wanted_names is None:
11156
        self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
11157
      else:
11158
        self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
11159

    
11160
      self.needed_locks[locking.LEVEL_NODE] = []
11161
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
11162

    
11163
  def DeclareLocks(self, level):
11164
    if self.op.use_locking and level == locking.LEVEL_NODE:
11165
      self._LockInstancesNodes()
11166

    
11167
  def CheckPrereq(self):
11168
    """Check prerequisites.
11169

11170
    This only checks the optional instance list against the existing names.
11171

11172
    """
11173
    if self.wanted_names is None:
11174
      assert self.op.use_locking, "Locking was not used"
11175
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
11176

    
11177
    self.wanted_instances = \
11178
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
11179

    
11180
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
11181
    """Returns the status of a block device
11182

11183
    """
11184
    if self.op.static or not node:
11185
      return None
11186

    
11187
    self.cfg.SetDiskID(dev, node)
11188

    
11189
    result = self.rpc.call_blockdev_find(node, dev)
11190
    if result.offline:
11191
      return None
11192

    
11193
    result.Raise("Can't compute disk status for %s" % instance_name)
11194

    
11195
    status = result.payload
11196
    if status is None:
11197
      return None
11198

    
11199
    return (status.dev_path, status.major, status.minor,
11200
            status.sync_percent, status.estimated_time,
11201
            status.is_degraded, status.ldisk_status)
11202

    
11203
  def _ComputeDiskStatus(self, instance, snode, dev):
11204
    """Compute block device status.
11205

11206
    """
11207
    if dev.dev_type in constants.LDS_DRBD:
11208
      # we change the snode then (otherwise we use the one passed in)
11209
      if dev.logical_id[0] == instance.primary_node:
11210
        snode = dev.logical_id[1]
11211
      else:
11212
        snode = dev.logical_id[0]
11213

    
11214
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
11215
                                              instance.name, dev)
11216
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
11217

    
11218
    if dev.children:
11219
      dev_children = map(compat.partial(self._ComputeDiskStatus,
11220
                                        instance, snode),
11221
                         dev.children)
11222
    else:
11223
      dev_children = []
11224

    
11225
    return {
11226
      "iv_name": dev.iv_name,
11227
      "dev_type": dev.dev_type,
11228
      "logical_id": dev.logical_id,
11229
      "physical_id": dev.physical_id,
11230
      "pstatus": dev_pstatus,
11231
      "sstatus": dev_sstatus,
11232
      "children": dev_children,
11233
      "mode": dev.mode,
11234
      "size": dev.size,
11235
      }
11236

    
11237
  def Exec(self, feedback_fn):
11238
    """Gather and return data"""
11239
    result = {}
11240

    
11241
    cluster = self.cfg.GetClusterInfo()
11242

    
11243
    pri_nodes = self.cfg.GetMultiNodeInfo(i.primary_node
11244
                                          for i in self.wanted_instances)
11245
    for instance, (_, pnode) in zip(self.wanted_instances, pri_nodes):
11246
      if self.op.static or pnode.offline:
11247
        remote_state = None
11248
        if pnode.offline:
11249
          self.LogWarning("Primary node %s is marked offline, returning static"
11250
                          " information only for instance %s" %
11251
                          (pnode.name, instance.name))
11252
      else:
11253
        remote_info = self.rpc.call_instance_info(instance.primary_node,
11254
                                                  instance.name,
11255
                                                  instance.hypervisor)
11256
        remote_info.Raise("Error checking node %s" % instance.primary_node)
11257
        remote_info = remote_info.payload
11258
        if remote_info and "state" in remote_info:
11259
          remote_state = "up"
11260
        else:
11261
          if instance.admin_state == constants.ADMINST_UP:
11262
            remote_state = "down"
11263
          else:
11264
            remote_state = instance.admin_state
11265

    
11266
      disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
11267
                  instance.disks)
11268

    
11269
      result[instance.name] = {
11270
        "name": instance.name,
11271
        "config_state": instance.admin_state,
11272
        "run_state": remote_state,
11273
        "pnode": instance.primary_node,
11274
        "snodes": instance.secondary_nodes,
11275
        "os": instance.os,
11276
        # this happens to be the same format used for hooks
11277
        "nics": _NICListToTuple(self, instance.nics),
11278
        "disk_template": instance.disk_template,
11279
        "disks": disks,
11280
        "hypervisor": instance.hypervisor,
11281
        "network_port": instance.network_port,
11282
        "hv_instance": instance.hvparams,
11283
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
11284
        "be_instance": instance.beparams,
11285
        "be_actual": cluster.FillBE(instance),
11286
        "os_instance": instance.osparams,
11287
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
11288
        "serial_no": instance.serial_no,
11289
        "mtime": instance.mtime,
11290
        "ctime": instance.ctime,
11291
        "uuid": instance.uuid,
11292
        }
11293

    
11294
    return result
11295

    
11296

    
11297
class LUInstanceSetParams(LogicalUnit):
11298
  """Modifies an instances's parameters.
11299

11300
  """
11301
  HPATH = "instance-modify"
11302
  HTYPE = constants.HTYPE_INSTANCE
11303
  REQ_BGL = False
11304

    
11305
  def CheckArguments(self):
11306
    if not (self.op.nics or self.op.disks or self.op.disk_template or
11307
            self.op.hvparams or self.op.beparams or self.op.os_name or
11308
            self.op.online_inst or self.op.offline_inst):
11309
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
11310

    
11311
    if self.op.hvparams:
11312
      _CheckGlobalHvParams(self.op.hvparams)
11313

    
11314
    # Disk validation
11315
    disk_addremove = 0
11316
    for disk_op, disk_dict in self.op.disks:
11317
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
11318
      if disk_op == constants.DDM_REMOVE:
11319
        disk_addremove += 1
11320
        continue
11321
      elif disk_op == constants.DDM_ADD:
11322
        disk_addremove += 1
11323
      else:
11324
        if not isinstance(disk_op, int):
11325
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
11326
        if not isinstance(disk_dict, dict):
11327
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
11328
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
11329

    
11330
      if disk_op == constants.DDM_ADD:
11331
        mode = disk_dict.setdefault(constants.IDISK_MODE, constants.DISK_RDWR)
11332
        if mode not in constants.DISK_ACCESS_SET:
11333
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
11334
                                     errors.ECODE_INVAL)
11335
        size = disk_dict.get(constants.IDISK_SIZE, None)
11336
        if size is None:
11337
          raise errors.OpPrereqError("Required disk parameter size missing",
11338
                                     errors.ECODE_INVAL)
11339
        try:
11340
          size = int(size)
11341
        except (TypeError, ValueError), err:
11342
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
11343
                                     str(err), errors.ECODE_INVAL)
11344
        disk_dict[constants.IDISK_SIZE] = size
11345
      else:
11346
        # modification of disk
11347
        if constants.IDISK_SIZE in disk_dict:
11348
          raise errors.OpPrereqError("Disk size change not possible, use"
11349
                                     " grow-disk", errors.ECODE_INVAL)
11350

    
11351
    if disk_addremove > 1:
11352
      raise errors.OpPrereqError("Only one disk add or remove operation"
11353
                                 " supported at a time", errors.ECODE_INVAL)
11354

    
11355
    if self.op.disks and self.op.disk_template is not None:
11356
      raise errors.OpPrereqError("Disk template conversion and other disk"
11357
                                 " changes not supported at the same time",
11358
                                 errors.ECODE_INVAL)
11359

    
11360
    if (self.op.disk_template and
11361
        self.op.disk_template in constants.DTS_INT_MIRROR and
11362
        self.op.remote_node is None):
11363
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
11364
                                 " one requires specifying a secondary node",
11365
                                 errors.ECODE_INVAL)
11366

    
11367
    # NIC validation
11368
    nic_addremove = 0
11369
    for nic_op, nic_dict in self.op.nics:
11370
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
11371
      if nic_op == constants.DDM_REMOVE:
11372
        nic_addremove += 1
11373
        continue
11374
      elif nic_op == constants.DDM_ADD:
11375
        nic_addremove += 1
11376
      else:
11377
        if not isinstance(nic_op, int):
11378
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
11379
        if not isinstance(nic_dict, dict):
11380
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
11381
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
11382

    
11383
      # nic_dict should be a dict
11384
      nic_ip = nic_dict.get(constants.INIC_IP, None)
11385
      if nic_ip is not None:
11386
        if nic_ip.lower() == constants.VALUE_NONE:
11387
          nic_dict[constants.INIC_IP] = None
11388
        else:
11389
          if not netutils.IPAddress.IsValid(nic_ip):
11390
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
11391
                                       errors.ECODE_INVAL)
11392

    
11393
      nic_bridge = nic_dict.get("bridge", None)
11394
      nic_link = nic_dict.get(constants.INIC_LINK, None)
11395
      if nic_bridge and nic_link:
11396
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
11397
                                   " at the same time", errors.ECODE_INVAL)
11398
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
11399
        nic_dict["bridge"] = None
11400
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
11401
        nic_dict[constants.INIC_LINK] = None
11402

    
11403
      if nic_op == constants.DDM_ADD:
11404
        nic_mac = nic_dict.get(constants.INIC_MAC, None)
11405
        if nic_mac is None:
11406
          nic_dict[constants.INIC_MAC] = constants.VALUE_AUTO
11407

    
11408
      if constants.INIC_MAC in nic_dict:
11409
        nic_mac = nic_dict[constants.INIC_MAC]
11410
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
11411
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
11412

    
11413
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
11414
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
11415
                                     " modifying an existing nic",
11416
                                     errors.ECODE_INVAL)
11417

    
11418
    if nic_addremove > 1:
11419
      raise errors.OpPrereqError("Only one NIC add or remove operation"
11420
                                 " supported at a time", errors.ECODE_INVAL)
11421

    
11422
  def ExpandNames(self):
11423
    self._ExpandAndLockInstance()
11424
    # Can't even acquire node locks in shared mode as upcoming changes in
11425
    # Ganeti 2.6 will start to modify the node object on disk conversion
11426
    self.needed_locks[locking.LEVEL_NODE] = []
11427
    self.needed_locks[locking.LEVEL_NODE_RES] = []
11428
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
11429

    
11430
  def DeclareLocks(self, level):
11431
    if level == locking.LEVEL_NODE:
11432
      self._LockInstancesNodes()
11433
      if self.op.disk_template and self.op.remote_node:
11434
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
11435
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
11436
    elif level == locking.LEVEL_NODE_RES and self.op.disk_template:
11437
      # Copy node locks
11438
      self.needed_locks[locking.LEVEL_NODE_RES] = \
11439
        self.needed_locks[locking.LEVEL_NODE][:]
11440

    
11441
  def BuildHooksEnv(self):
11442
    """Build hooks env.
11443

11444
    This runs on the master, primary and secondaries.
11445

11446
    """
11447
    args = dict()
11448
    if constants.BE_MINMEM in self.be_new:
11449
      args["minmem"] = self.be_new[constants.BE_MINMEM]
11450
    if constants.BE_MAXMEM in self.be_new:
11451
      args["maxmem"] = self.be_new[constants.BE_MAXMEM]
11452
    if constants.BE_VCPUS in self.be_new:
11453
      args["vcpus"] = self.be_new[constants.BE_VCPUS]
11454
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
11455
    # information at all.
11456
    if self.op.nics:
11457
      args["nics"] = []
11458
      nic_override = dict(self.op.nics)
11459
      for idx, nic in enumerate(self.instance.nics):
11460
        if idx in nic_override:
11461
          this_nic_override = nic_override[idx]
11462
        else:
11463
          this_nic_override = {}
11464
        if constants.INIC_IP in this_nic_override:
11465
          ip = this_nic_override[constants.INIC_IP]
11466
        else:
11467
          ip = nic.ip
11468
        if constants.INIC_MAC in this_nic_override:
11469
          mac = this_nic_override[constants.INIC_MAC]
11470
        else:
11471
          mac = nic.mac
11472
        if idx in self.nic_pnew:
11473
          nicparams = self.nic_pnew[idx]
11474
        else:
11475
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
11476
        mode = nicparams[constants.NIC_MODE]
11477
        link = nicparams[constants.NIC_LINK]
11478
        args["nics"].append((ip, mac, mode, link))
11479
      if constants.DDM_ADD in nic_override:
11480
        ip = nic_override[constants.DDM_ADD].get(constants.INIC_IP, None)
11481
        mac = nic_override[constants.DDM_ADD][constants.INIC_MAC]
11482
        nicparams = self.nic_pnew[constants.DDM_ADD]
11483
        mode = nicparams[constants.NIC_MODE]
11484
        link = nicparams[constants.NIC_LINK]
11485
        args["nics"].append((ip, mac, mode, link))
11486
      elif constants.DDM_REMOVE in nic_override:
11487
        del args["nics"][-1]
11488

    
11489
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
11490
    if self.op.disk_template:
11491
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
11492

    
11493
    return env
11494

    
11495
  def BuildHooksNodes(self):
11496
    """Build hooks nodes.
11497

11498
    """
11499
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
11500
    return (nl, nl)
11501

    
11502
  def CheckPrereq(self):
11503
    """Check prerequisites.
11504

11505
    This only checks the instance list against the existing names.
11506

11507
    """
11508
    # checking the new params on the primary/secondary nodes
11509

    
11510
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
11511
    cluster = self.cluster = self.cfg.GetClusterInfo()
11512
    assert self.instance is not None, \
11513
      "Cannot retrieve locked instance %s" % self.op.instance_name
11514
    pnode = instance.primary_node
11515
    nodelist = list(instance.all_nodes)
11516
    pnode_info = self.cfg.GetNodeInfo(pnode)
11517
    self.diskparams = self.cfg.GetNodeGroup(pnode_info.group).diskparams
11518

    
11519
    # OS change
11520
    if self.op.os_name and not self.op.force:
11521
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
11522
                      self.op.force_variant)
11523
      instance_os = self.op.os_name
11524
    else:
11525
      instance_os = instance.os
11526

    
11527
    if self.op.disk_template:
11528
      if instance.disk_template == self.op.disk_template:
11529
        raise errors.OpPrereqError("Instance already has disk template %s" %
11530
                                   instance.disk_template, errors.ECODE_INVAL)
11531

    
11532
      if (instance.disk_template,
11533
          self.op.disk_template) not in self._DISK_CONVERSIONS:
11534
        raise errors.OpPrereqError("Unsupported disk template conversion from"
11535
                                   " %s to %s" % (instance.disk_template,
11536
                                                  self.op.disk_template),
11537
                                   errors.ECODE_INVAL)
11538
      _CheckInstanceState(self, instance, INSTANCE_DOWN,
11539
                          msg="cannot change disk template")
11540
      if self.op.disk_template in constants.DTS_INT_MIRROR:
11541
        if self.op.remote_node == pnode:
11542
          raise errors.OpPrereqError("Given new secondary node %s is the same"
11543
                                     " as the primary node of the instance" %
11544
                                     self.op.remote_node, errors.ECODE_STATE)
11545
        _CheckNodeOnline(self, self.op.remote_node)
11546
        _CheckNodeNotDrained(self, self.op.remote_node)
11547
        # FIXME: here we assume that the old instance type is DT_PLAIN
11548
        assert instance.disk_template == constants.DT_PLAIN
11549
        disks = [{constants.IDISK_SIZE: d.size,
11550
                  constants.IDISK_VG: d.logical_id[0]}
11551
                 for d in instance.disks]
11552
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
11553
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
11554

    
11555
        snode_info = self.cfg.GetNodeInfo(self.op.remote_node)
11556
        if pnode_info.group != snode_info.group:
11557
          self.LogWarning("The primary and secondary nodes are in two"
11558
                          " different node groups; the disk parameters"
11559
                          " from the first disk's node group will be"
11560
                          " used")
11561

    
11562
    # hvparams processing
11563
    if self.op.hvparams:
11564
      hv_type = instance.hypervisor
11565
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
11566
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
11567
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
11568

    
11569
      # local check
11570
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
11571
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
11572
      self.hv_proposed = self.hv_new = hv_new # the new actual values
11573
      self.hv_inst = i_hvdict # the new dict (without defaults)
11574
    else:
11575
      self.hv_proposed = cluster.SimpleFillHV(instance.hypervisor, instance.os,
11576
                                              instance.hvparams)
11577
      self.hv_new = self.hv_inst = {}
11578

    
11579
    # beparams processing
11580
    if self.op.beparams:
11581
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
11582
                                   use_none=True)
11583
      objects.UpgradeBeParams(i_bedict)
11584
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
11585
      be_new = cluster.SimpleFillBE(i_bedict)
11586
      self.be_proposed = self.be_new = be_new # the new actual values
11587
      self.be_inst = i_bedict # the new dict (without defaults)
11588
    else:
11589
      self.be_new = self.be_inst = {}
11590
      self.be_proposed = cluster.SimpleFillBE(instance.beparams)
11591
    be_old = cluster.FillBE(instance)
11592

    
11593
    # CPU param validation -- checking every time a paramtere is
11594
    # changed to cover all cases where either CPU mask or vcpus have
11595
    # changed
11596
    if (constants.BE_VCPUS in self.be_proposed and
11597
        constants.HV_CPU_MASK in self.hv_proposed):
11598
      cpu_list = \
11599
        utils.ParseMultiCpuMask(self.hv_proposed[constants.HV_CPU_MASK])
11600
      # Verify mask is consistent with number of vCPUs. Can skip this
11601
      # test if only 1 entry in the CPU mask, which means same mask
11602
      # is applied to all vCPUs.
11603
      if (len(cpu_list) > 1 and
11604
          len(cpu_list) != self.be_proposed[constants.BE_VCPUS]):
11605
        raise errors.OpPrereqError("Number of vCPUs [%d] does not match the"
11606
                                   " CPU mask [%s]" %
11607
                                   (self.be_proposed[constants.BE_VCPUS],
11608
                                    self.hv_proposed[constants.HV_CPU_MASK]),
11609
                                   errors.ECODE_INVAL)
11610

    
11611
      # Only perform this test if a new CPU mask is given
11612
      if constants.HV_CPU_MASK in self.hv_new:
11613
        # Calculate the largest CPU number requested
11614
        max_requested_cpu = max(map(max, cpu_list))
11615
        # Check that all of the instance's nodes have enough physical CPUs to
11616
        # satisfy the requested CPU mask
11617
        _CheckNodesPhysicalCPUs(self, instance.all_nodes,
11618
                                max_requested_cpu + 1, instance.hypervisor)
11619

    
11620
    # osparams processing
11621
    if self.op.osparams:
11622
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
11623
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
11624
      self.os_inst = i_osdict # the new dict (without defaults)
11625
    else:
11626
      self.os_inst = {}
11627

    
11628
    self.warn = []
11629

    
11630
    #TODO(dynmem): do the appropriate check involving MINMEM
11631
    if (constants.BE_MAXMEM in self.op.beparams and not self.op.force and
11632
        be_new[constants.BE_MAXMEM] > be_old[constants.BE_MAXMEM]):
11633
      mem_check_list = [pnode]
11634
      if be_new[constants.BE_AUTO_BALANCE]:
11635
        # either we changed auto_balance to yes or it was from before
11636
        mem_check_list.extend(instance.secondary_nodes)
11637
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
11638
                                                  instance.hypervisor)
11639
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
11640
                                         [instance.hypervisor])
11641
      pninfo = nodeinfo[pnode]
11642
      msg = pninfo.fail_msg
11643
      if msg:
11644
        # Assume the primary node is unreachable and go ahead
11645
        self.warn.append("Can't get info from primary node %s: %s" %
11646
                         (pnode, msg))
11647
      else:
11648
        (_, _, (pnhvinfo, )) = pninfo.payload
11649
        if not isinstance(pnhvinfo.get("memory_free", None), int):
11650
          self.warn.append("Node data from primary node %s doesn't contain"
11651
                           " free memory information" % pnode)
11652
        elif instance_info.fail_msg:
11653
          self.warn.append("Can't get instance runtime information: %s" %
11654
                          instance_info.fail_msg)
11655
        else:
11656
          if instance_info.payload:
11657
            current_mem = int(instance_info.payload["memory"])
11658
          else:
11659
            # Assume instance not running
11660
            # (there is a slight race condition here, but it's not very
11661
            # probable, and we have no other way to check)
11662
            # TODO: Describe race condition
11663
            current_mem = 0
11664
          #TODO(dynmem): do the appropriate check involving MINMEM
11665
          miss_mem = (be_new[constants.BE_MAXMEM] - current_mem -
11666
                      pnhvinfo["memory_free"])
11667
          if miss_mem > 0:
11668
            raise errors.OpPrereqError("This change will prevent the instance"
11669
                                       " from starting, due to %d MB of memory"
11670
                                       " missing on its primary node" %
11671
                                       miss_mem,
11672
                                       errors.ECODE_NORES)
11673

    
11674
      if be_new[constants.BE_AUTO_BALANCE]:
11675
        for node, nres in nodeinfo.items():
11676
          if node not in instance.secondary_nodes:
11677
            continue
11678
          nres.Raise("Can't get info from secondary node %s" % node,
11679
                     prereq=True, ecode=errors.ECODE_STATE)
11680
          (_, _, (nhvinfo, )) = nres.payload
11681
          if not isinstance(nhvinfo.get("memory_free", None), int):
11682
            raise errors.OpPrereqError("Secondary node %s didn't return free"
11683
                                       " memory information" % node,
11684
                                       errors.ECODE_STATE)
11685
          #TODO(dynmem): do the appropriate check involving MINMEM
11686
          elif be_new[constants.BE_MAXMEM] > nhvinfo["memory_free"]:
11687
            raise errors.OpPrereqError("This change will prevent the instance"
11688
                                       " from failover to its secondary node"
11689
                                       " %s, due to not enough memory" % node,
11690
                                       errors.ECODE_STATE)
11691

    
11692
    # NIC processing
11693
    self.nic_pnew = {}
11694
    self.nic_pinst = {}
11695
    for nic_op, nic_dict in self.op.nics:
11696
      if nic_op == constants.DDM_REMOVE:
11697
        if not instance.nics:
11698
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
11699
                                     errors.ECODE_INVAL)
11700
        continue
11701
      if nic_op != constants.DDM_ADD:
11702
        # an existing nic
11703
        if not instance.nics:
11704
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
11705
                                     " no NICs" % nic_op,
11706
                                     errors.ECODE_INVAL)
11707
        if nic_op < 0 or nic_op >= len(instance.nics):
11708
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
11709
                                     " are 0 to %d" %
11710
                                     (nic_op, len(instance.nics) - 1),
11711
                                     errors.ECODE_INVAL)
11712
        old_nic_params = instance.nics[nic_op].nicparams
11713
        old_nic_ip = instance.nics[nic_op].ip
11714
      else:
11715
        old_nic_params = {}
11716
        old_nic_ip = None
11717

    
11718
      update_params_dict = dict([(key, nic_dict[key])
11719
                                 for key in constants.NICS_PARAMETERS
11720
                                 if key in nic_dict])
11721

    
11722
      if "bridge" in nic_dict:
11723
        update_params_dict[constants.NIC_LINK] = nic_dict["bridge"]
11724

    
11725
      new_nic_params = _GetUpdatedParams(old_nic_params,
11726
                                         update_params_dict)
11727
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
11728
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
11729
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
11730
      self.nic_pinst[nic_op] = new_nic_params
11731
      self.nic_pnew[nic_op] = new_filled_nic_params
11732
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
11733

    
11734
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
11735
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
11736
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
11737
        if msg:
11738
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
11739
          if self.op.force:
11740
            self.warn.append(msg)
11741
          else:
11742
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
11743
      if new_nic_mode == constants.NIC_MODE_ROUTED:
11744
        if constants.INIC_IP in nic_dict:
11745
          nic_ip = nic_dict[constants.INIC_IP]
11746
        else:
11747
          nic_ip = old_nic_ip
11748
        if nic_ip is None:
11749
          raise errors.OpPrereqError("Cannot set the nic ip to None"
11750
                                     " on a routed nic", errors.ECODE_INVAL)
11751
      if constants.INIC_MAC in nic_dict:
11752
        nic_mac = nic_dict[constants.INIC_MAC]
11753
        if nic_mac is None:
11754
          raise errors.OpPrereqError("Cannot set the nic mac to None",
11755
                                     errors.ECODE_INVAL)
11756
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
11757
          # otherwise generate the mac
11758
          nic_dict[constants.INIC_MAC] = \
11759
            self.cfg.GenerateMAC(self.proc.GetECId())
11760
        else:
11761
          # or validate/reserve the current one
11762
          try:
11763
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
11764
          except errors.ReservationError:
11765
            raise errors.OpPrereqError("MAC address %s already in use"
11766
                                       " in cluster" % nic_mac,
11767
                                       errors.ECODE_NOTUNIQUE)
11768

    
11769
    # DISK processing
11770
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
11771
      raise errors.OpPrereqError("Disk operations not supported for"
11772
                                 " diskless instances",
11773
                                 errors.ECODE_INVAL)
11774
    for disk_op, _ in self.op.disks:
11775
      if disk_op == constants.DDM_REMOVE:
11776
        if len(instance.disks) == 1:
11777
          raise errors.OpPrereqError("Cannot remove the last disk of"
11778
                                     " an instance", errors.ECODE_INVAL)
11779
        _CheckInstanceState(self, instance, INSTANCE_DOWN,
11780
                            msg="cannot remove disks")
11781

    
11782
      if (disk_op == constants.DDM_ADD and
11783
          len(instance.disks) >= constants.MAX_DISKS):
11784
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
11785
                                   " add more" % constants.MAX_DISKS,
11786
                                   errors.ECODE_STATE)
11787
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
11788
        # an existing disk
11789
        if disk_op < 0 or disk_op >= len(instance.disks):
11790
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
11791
                                     " are 0 to %d" %
11792
                                     (disk_op, len(instance.disks)),
11793
                                     errors.ECODE_INVAL)
11794

    
11795
    # disabling the instance
11796
    if self.op.offline_inst:
11797
      _CheckInstanceState(self, instance, INSTANCE_DOWN,
11798
                          msg="cannot change instance state to offline")
11799

    
11800
    # enabling the instance
11801
    if self.op.online_inst:
11802
      _CheckInstanceState(self, instance, INSTANCE_OFFLINE,
11803
                          msg="cannot make instance go online")
11804

    
11805
  def _ConvertPlainToDrbd(self, feedback_fn):
11806
    """Converts an instance from plain to drbd.
11807

11808
    """
11809
    feedback_fn("Converting template to drbd")
11810
    instance = self.instance
11811
    pnode = instance.primary_node
11812
    snode = self.op.remote_node
11813

    
11814
    assert instance.disk_template == constants.DT_PLAIN
11815

    
11816
    # create a fake disk info for _GenerateDiskTemplate
11817
    disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
11818
                  constants.IDISK_VG: d.logical_id[0]}
11819
                 for d in instance.disks]
11820
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
11821
                                      instance.name, pnode, [snode],
11822
                                      disk_info, None, None, 0, feedback_fn,
11823
                                      self.diskparams)
11824
    info = _GetInstanceInfoText(instance)
11825
    feedback_fn("Creating aditional volumes...")
11826
    # first, create the missing data and meta devices
11827
    for disk in new_disks:
11828
      # unfortunately this is... not too nice
11829
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
11830
                            info, True)
11831
      for child in disk.children:
11832
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
11833
    # at this stage, all new LVs have been created, we can rename the
11834
    # old ones
11835
    feedback_fn("Renaming original volumes...")
11836
    rename_list = [(o, n.children[0].logical_id)
11837
                   for (o, n) in zip(instance.disks, new_disks)]
11838
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
11839
    result.Raise("Failed to rename original LVs")
11840

    
11841
    feedback_fn("Initializing DRBD devices...")
11842
    # all child devices are in place, we can now create the DRBD devices
11843
    for disk in new_disks:
11844
      for node in [pnode, snode]:
11845
        f_create = node == pnode
11846
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
11847

    
11848
    # at this point, the instance has been modified
11849
    instance.disk_template = constants.DT_DRBD8
11850
    instance.disks = new_disks
11851
    self.cfg.Update(instance, feedback_fn)
11852

    
11853
    # Release node locks while waiting for sync
11854
    _ReleaseLocks(self, locking.LEVEL_NODE)
11855

    
11856
    # disks are created, waiting for sync
11857
    disk_abort = not _WaitForSync(self, instance,
11858
                                  oneshot=not self.op.wait_for_sync)
11859
    if disk_abort:
11860
      raise errors.OpExecError("There are some degraded disks for"
11861
                               " this instance, please cleanup manually")
11862

    
11863
    # Node resource locks will be released by caller
11864

    
11865
  def _ConvertDrbdToPlain(self, feedback_fn):
11866
    """Converts an instance from drbd to plain.
11867

11868
    """
11869
    instance = self.instance
11870

    
11871
    assert len(instance.secondary_nodes) == 1
11872
    assert instance.disk_template == constants.DT_DRBD8
11873

    
11874
    pnode = instance.primary_node
11875
    snode = instance.secondary_nodes[0]
11876
    feedback_fn("Converting template to plain")
11877

    
11878
    old_disks = instance.disks
11879
    new_disks = [d.children[0] for d in old_disks]
11880

    
11881
    # copy over size and mode
11882
    for parent, child in zip(old_disks, new_disks):
11883
      child.size = parent.size
11884
      child.mode = parent.mode
11885

    
11886
    # update instance structure
11887
    instance.disks = new_disks
11888
    instance.disk_template = constants.DT_PLAIN
11889
    self.cfg.Update(instance, feedback_fn)
11890

    
11891
    # Release locks in case removing disks takes a while
11892
    _ReleaseLocks(self, locking.LEVEL_NODE)
11893

    
11894
    feedback_fn("Removing volumes on the secondary node...")
11895
    for disk in old_disks:
11896
      self.cfg.SetDiskID(disk, snode)
11897
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
11898
      if msg:
11899
        self.LogWarning("Could not remove block device %s on node %s,"
11900
                        " continuing anyway: %s", disk.iv_name, snode, msg)
11901

    
11902
    feedback_fn("Removing unneeded volumes on the primary node...")
11903
    for idx, disk in enumerate(old_disks):
11904
      meta = disk.children[1]
11905
      self.cfg.SetDiskID(meta, pnode)
11906
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
11907
      if msg:
11908
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
11909
                        " continuing anyway: %s", idx, pnode, msg)
11910

    
11911
    # this is a DRBD disk, return its port to the pool
11912
    for disk in old_disks:
11913
      tcp_port = disk.logical_id[2]
11914
      self.cfg.AddTcpUdpPort(tcp_port)
11915

    
11916
    # Node resource locks will be released by caller
11917

    
11918
  def Exec(self, feedback_fn):
11919
    """Modifies an instance.
11920

11921
    All parameters take effect only at the next restart of the instance.
11922

11923
    """
11924
    # Process here the warnings from CheckPrereq, as we don't have a
11925
    # feedback_fn there.
11926
    for warn in self.warn:
11927
      feedback_fn("WARNING: %s" % warn)
11928

    
11929
    assert ((self.op.disk_template is None) ^
11930
            bool(self.owned_locks(locking.LEVEL_NODE_RES))), \
11931
      "Not owning any node resource locks"
11932

    
11933
    result = []
11934
    instance = self.instance
11935
    # disk changes
11936
    for disk_op, disk_dict in self.op.disks:
11937
      if disk_op == constants.DDM_REMOVE:
11938
        # remove the last disk
11939
        device = instance.disks.pop()
11940
        device_idx = len(instance.disks)
11941
        for node, disk in device.ComputeNodeTree(instance.primary_node):
11942
          self.cfg.SetDiskID(disk, node)
11943
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
11944
          if msg:
11945
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
11946
                            " continuing anyway", device_idx, node, msg)
11947
        result.append(("disk/%d" % device_idx, "remove"))
11948

    
11949
        # if this is a DRBD disk, return its port to the pool
11950
        if device.dev_type in constants.LDS_DRBD:
11951
          tcp_port = device.logical_id[2]
11952
          self.cfg.AddTcpUdpPort(tcp_port)
11953
      elif disk_op == constants.DDM_ADD:
11954
        # add a new disk
11955
        if instance.disk_template in (constants.DT_FILE,
11956
                                        constants.DT_SHARED_FILE):
11957
          file_driver, file_path = instance.disks[0].logical_id
11958
          file_path = os.path.dirname(file_path)
11959
        else:
11960
          file_driver = file_path = None
11961
        disk_idx_base = len(instance.disks)
11962
        new_disk = _GenerateDiskTemplate(self,
11963
                                         instance.disk_template,
11964
                                         instance.name, instance.primary_node,
11965
                                         instance.secondary_nodes,
11966
                                         [disk_dict],
11967
                                         file_path,
11968
                                         file_driver,
11969
                                         disk_idx_base,
11970
                                         feedback_fn,
11971
                                         self.diskparams)[0]
11972
        instance.disks.append(new_disk)
11973
        info = _GetInstanceInfoText(instance)
11974

    
11975
        logging.info("Creating volume %s for instance %s",
11976
                     new_disk.iv_name, instance.name)
11977
        # Note: this needs to be kept in sync with _CreateDisks
11978
        #HARDCODE
11979
        for node in instance.all_nodes:
11980
          f_create = node == instance.primary_node
11981
          try:
11982
            _CreateBlockDev(self, node, instance, new_disk,
11983
                            f_create, info, f_create)
11984
          except errors.OpExecError, err:
11985
            self.LogWarning("Failed to create volume %s (%s) on"
11986
                            " node %s: %s",
11987
                            new_disk.iv_name, new_disk, node, err)
11988
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
11989
                       (new_disk.size, new_disk.mode)))
11990
      else:
11991
        # change a given disk
11992
        instance.disks[disk_op].mode = disk_dict[constants.IDISK_MODE]
11993
        result.append(("disk.mode/%d" % disk_op,
11994
                       disk_dict[constants.IDISK_MODE]))
11995

    
11996
    if self.op.disk_template:
11997
      if __debug__:
11998
        check_nodes = set(instance.all_nodes)
11999
        if self.op.remote_node:
12000
          check_nodes.add(self.op.remote_node)
12001
        for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]:
12002
          owned = self.owned_locks(level)
12003
          assert not (check_nodes - owned), \
12004
            ("Not owning the correct locks, owning %r, expected at least %r" %
12005
             (owned, check_nodes))
12006

    
12007
      r_shut = _ShutdownInstanceDisks(self, instance)
12008
      if not r_shut:
12009
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
12010
                                 " proceed with disk template conversion")
12011
      mode = (instance.disk_template, self.op.disk_template)
12012
      try:
12013
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
12014
      except:
12015
        self.cfg.ReleaseDRBDMinors(instance.name)
12016
        raise
12017
      result.append(("disk_template", self.op.disk_template))
12018

    
12019
      assert instance.disk_template == self.op.disk_template, \
12020
        ("Expected disk template '%s', found '%s'" %
12021
         (self.op.disk_template, instance.disk_template))
12022

    
12023
    # Release node and resource locks if there are any (they might already have
12024
    # been released during disk conversion)
12025
    _ReleaseLocks(self, locking.LEVEL_NODE)
12026
    _ReleaseLocks(self, locking.LEVEL_NODE_RES)
12027

    
12028
    # NIC changes
12029
    for nic_op, nic_dict in self.op.nics:
12030
      if nic_op == constants.DDM_REMOVE:
12031
        # remove the last nic
12032
        del instance.nics[-1]
12033
        result.append(("nic.%d" % len(instance.nics), "remove"))
12034
      elif nic_op == constants.DDM_ADD:
12035
        # mac and bridge should be set, by now
12036
        mac = nic_dict[constants.INIC_MAC]
12037
        ip = nic_dict.get(constants.INIC_IP, None)
12038
        nicparams = self.nic_pinst[constants.DDM_ADD]
12039
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
12040
        instance.nics.append(new_nic)
12041
        result.append(("nic.%d" % (len(instance.nics) - 1),
12042
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
12043
                       (new_nic.mac, new_nic.ip,
12044
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
12045
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
12046
                       )))
12047
      else:
12048
        for key in (constants.INIC_MAC, constants.INIC_IP):
12049
          if key in nic_dict:
12050
            setattr(instance.nics[nic_op], key, nic_dict[key])
12051
        if nic_op in self.nic_pinst:
12052
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
12053
        for key, val in nic_dict.iteritems():
12054
          result.append(("nic.%s/%d" % (key, nic_op), val))
12055

    
12056
    # hvparams changes
12057
    if self.op.hvparams:
12058
      instance.hvparams = self.hv_inst
12059
      for key, val in self.op.hvparams.iteritems():
12060
        result.append(("hv/%s" % key, val))
12061

    
12062
    # beparams changes
12063
    if self.op.beparams:
12064
      instance.beparams = self.be_inst
12065
      for key, val in self.op.beparams.iteritems():
12066
        result.append(("be/%s" % key, val))
12067

    
12068
    # OS change
12069
    if self.op.os_name:
12070
      instance.os = self.op.os_name
12071

    
12072
    # osparams changes
12073
    if self.op.osparams:
12074
      instance.osparams = self.os_inst
12075
      for key, val in self.op.osparams.iteritems():
12076
        result.append(("os/%s" % key, val))
12077

    
12078
    # online/offline instance
12079
    if self.op.online_inst:
12080
      self.cfg.MarkInstanceDown(instance.name)
12081
      result.append(("admin_state", constants.ADMINST_DOWN))
12082
    if self.op.offline_inst:
12083
      self.cfg.MarkInstanceOffline(instance.name)
12084
      result.append(("admin_state", constants.ADMINST_OFFLINE))
12085

    
12086
    self.cfg.Update(instance, feedback_fn)
12087

    
12088
    assert not (self.owned_locks(locking.LEVEL_NODE_RES) or
12089
                self.owned_locks(locking.LEVEL_NODE)), \
12090
      "All node locks should have been released by now"
12091

    
12092
    return result
12093

    
12094
  _DISK_CONVERSIONS = {
12095
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
12096
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
12097
    }
12098

    
12099

    
12100
class LUInstanceChangeGroup(LogicalUnit):
12101
  HPATH = "instance-change-group"
12102
  HTYPE = constants.HTYPE_INSTANCE
12103
  REQ_BGL = False
12104

    
12105
  def ExpandNames(self):
12106
    self.share_locks = _ShareAll()
12107
    self.needed_locks = {
12108
      locking.LEVEL_NODEGROUP: [],
12109
      locking.LEVEL_NODE: [],
12110
      }
12111

    
12112
    self._ExpandAndLockInstance()
12113

    
12114
    if self.op.target_groups:
12115
      self.req_target_uuids = map(self.cfg.LookupNodeGroup,
12116
                                  self.op.target_groups)
12117
    else:
12118
      self.req_target_uuids = None
12119

    
12120
    self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
12121

    
12122
  def DeclareLocks(self, level):
12123
    if level == locking.LEVEL_NODEGROUP:
12124
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
12125

    
12126
      if self.req_target_uuids:
12127
        lock_groups = set(self.req_target_uuids)
12128

    
12129
        # Lock all groups used by instance optimistically; this requires going
12130
        # via the node before it's locked, requiring verification later on
12131
        instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
12132
        lock_groups.update(instance_groups)
12133
      else:
12134
        # No target groups, need to lock all of them
12135
        lock_groups = locking.ALL_SET
12136

    
12137
      self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
12138

    
12139
    elif level == locking.LEVEL_NODE:
12140
      if self.req_target_uuids:
12141
        # Lock all nodes used by instances
12142
        self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
12143
        self._LockInstancesNodes()
12144

    
12145
        # Lock all nodes in all potential target groups
12146
        lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
12147
                       self.cfg.GetInstanceNodeGroups(self.op.instance_name))
12148
        member_nodes = [node_name
12149
                        for group in lock_groups
12150
                        for node_name in self.cfg.GetNodeGroup(group).members]
12151
        self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
12152
      else:
12153
        # Lock all nodes as all groups are potential targets
12154
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
12155

    
12156
  def CheckPrereq(self):
12157
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
12158
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
12159
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
12160

    
12161
    assert (self.req_target_uuids is None or
12162
            owned_groups.issuperset(self.req_target_uuids))
12163
    assert owned_instances == set([self.op.instance_name])
12164

    
12165
    # Get instance information
12166
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
12167

    
12168
    # Check if node groups for locked instance are still correct
12169
    assert owned_nodes.issuperset(self.instance.all_nodes), \
12170
      ("Instance %s's nodes changed while we kept the lock" %
12171
       self.op.instance_name)
12172

    
12173
    inst_groups = _CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
12174
                                           owned_groups)
12175

    
12176
    if self.req_target_uuids:
12177
      # User requested specific target groups
12178
      self.target_uuids = self.req_target_uuids
12179
    else:
12180
      # All groups except those used by the instance are potential targets
12181
      self.target_uuids = owned_groups - inst_groups
12182

    
12183
    conflicting_groups = self.target_uuids & inst_groups
12184
    if conflicting_groups:
12185
      raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
12186
                                 " used by the instance '%s'" %
12187
                                 (utils.CommaJoin(conflicting_groups),
12188
                                  self.op.instance_name),
12189
                                 errors.ECODE_INVAL)
12190

    
12191
    if not self.target_uuids:
12192
      raise errors.OpPrereqError("There are no possible target groups",
12193
                                 errors.ECODE_INVAL)
12194

    
12195
  def BuildHooksEnv(self):
12196
    """Build hooks env.
12197

12198
    """
12199
    assert self.target_uuids
12200

    
12201
    env = {
12202
      "TARGET_GROUPS": " ".join(self.target_uuids),
12203
      }
12204

    
12205
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
12206

    
12207
    return env
12208

    
12209
  def BuildHooksNodes(self):
12210
    """Build hooks nodes.
12211

12212
    """
12213
    mn = self.cfg.GetMasterNode()
12214
    return ([mn], [mn])
12215

    
12216
  def Exec(self, feedback_fn):
12217
    instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
12218

    
12219
    assert instances == [self.op.instance_name], "Instance not locked"
12220

    
12221
    ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
12222
                     instances=instances, target_groups=list(self.target_uuids))
12223

    
12224
    ial.Run(self.op.iallocator)
12225

    
12226
    if not ial.success:
12227
      raise errors.OpPrereqError("Can't compute solution for changing group of"
12228
                                 " instance '%s' using iallocator '%s': %s" %
12229
                                 (self.op.instance_name, self.op.iallocator,
12230
                                  ial.info),
12231
                                 errors.ECODE_NORES)
12232

    
12233
    jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
12234

    
12235
    self.LogInfo("Iallocator returned %s job(s) for changing group of"
12236
                 " instance '%s'", len(jobs), self.op.instance_name)
12237

    
12238
    return ResultWithJobs(jobs)
12239

    
12240

    
12241
class LUBackupQuery(NoHooksLU):
12242
  """Query the exports list
12243

12244
  """
12245
  REQ_BGL = False
12246

    
12247
  def ExpandNames(self):
12248
    self.needed_locks = {}
12249
    self.share_locks[locking.LEVEL_NODE] = 1
12250
    if not self.op.nodes:
12251
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
12252
    else:
12253
      self.needed_locks[locking.LEVEL_NODE] = \
12254
        _GetWantedNodes(self, self.op.nodes)
12255

    
12256
  def Exec(self, feedback_fn):
12257
    """Compute the list of all the exported system images.
12258

12259
    @rtype: dict
12260
    @return: a dictionary with the structure node->(export-list)
12261
        where export-list is a list of the instances exported on
12262
        that node.
12263

12264
    """
12265
    self.nodes = self.owned_locks(locking.LEVEL_NODE)
12266
    rpcresult = self.rpc.call_export_list(self.nodes)
12267
    result = {}
12268
    for node in rpcresult:
12269
      if rpcresult[node].fail_msg:
12270
        result[node] = False
12271
      else:
12272
        result[node] = rpcresult[node].payload
12273

    
12274
    return result
12275

    
12276

    
12277
class LUBackupPrepare(NoHooksLU):
12278
  """Prepares an instance for an export and returns useful information.
12279

12280
  """
12281
  REQ_BGL = False
12282

    
12283
  def ExpandNames(self):
12284
    self._ExpandAndLockInstance()
12285

    
12286
  def CheckPrereq(self):
12287
    """Check prerequisites.
12288

12289
    """
12290
    instance_name = self.op.instance_name
12291

    
12292
    self.instance = self.cfg.GetInstanceInfo(instance_name)
12293
    assert self.instance is not None, \
12294
          "Cannot retrieve locked instance %s" % self.op.instance_name
12295
    _CheckNodeOnline(self, self.instance.primary_node)
12296

    
12297
    self._cds = _GetClusterDomainSecret()
12298

    
12299
  def Exec(self, feedback_fn):
12300
    """Prepares an instance for an export.
12301

12302
    """
12303
    instance = self.instance
12304

    
12305
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
12306
      salt = utils.GenerateSecret(8)
12307

    
12308
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
12309
      result = self.rpc.call_x509_cert_create(instance.primary_node,
12310
                                              constants.RIE_CERT_VALIDITY)
12311
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
12312

    
12313
      (name, cert_pem) = result.payload
12314

    
12315
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
12316
                                             cert_pem)
12317

    
12318
      return {
12319
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
12320
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
12321
                          salt),
12322
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
12323
        }
12324

    
12325
    return None
12326

    
12327

    
12328
class LUBackupExport(LogicalUnit):
12329
  """Export an instance to an image in the cluster.
12330

12331
  """
12332
  HPATH = "instance-export"
12333
  HTYPE = constants.HTYPE_INSTANCE
12334
  REQ_BGL = False
12335

    
12336
  def CheckArguments(self):
12337
    """Check the arguments.
12338

12339
    """
12340
    self.x509_key_name = self.op.x509_key_name
12341
    self.dest_x509_ca_pem = self.op.destination_x509_ca
12342

    
12343
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
12344
      if not self.x509_key_name:
12345
        raise errors.OpPrereqError("Missing X509 key name for encryption",
12346
                                   errors.ECODE_INVAL)
12347

    
12348
      if not self.dest_x509_ca_pem:
12349
        raise errors.OpPrereqError("Missing destination X509 CA",
12350
                                   errors.ECODE_INVAL)
12351

    
12352
  def ExpandNames(self):
12353
    self._ExpandAndLockInstance()
12354

    
12355
    # Lock all nodes for local exports
12356
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
12357
      # FIXME: lock only instance primary and destination node
12358
      #
12359
      # Sad but true, for now we have do lock all nodes, as we don't know where
12360
      # the previous export might be, and in this LU we search for it and
12361
      # remove it from its current node. In the future we could fix this by:
12362
      #  - making a tasklet to search (share-lock all), then create the
12363
      #    new one, then one to remove, after
12364
      #  - removing the removal operation altogether
12365
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
12366

    
12367
  def DeclareLocks(self, level):
12368
    """Last minute lock declaration."""
12369
    # All nodes are locked anyway, so nothing to do here.
12370

    
12371
  def BuildHooksEnv(self):
12372
    """Build hooks env.
12373

12374
    This will run on the master, primary node and target node.
12375

12376
    """
12377
    env = {
12378
      "EXPORT_MODE": self.op.mode,
12379
      "EXPORT_NODE": self.op.target_node,
12380
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
12381
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
12382
      # TODO: Generic function for boolean env variables
12383
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
12384
      }
12385

    
12386
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
12387

    
12388
    return env
12389

    
12390
  def BuildHooksNodes(self):
12391
    """Build hooks nodes.
12392

12393
    """
12394
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
12395

    
12396
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
12397
      nl.append(self.op.target_node)
12398

    
12399
    return (nl, nl)
12400

    
12401
  def CheckPrereq(self):
12402
    """Check prerequisites.
12403

12404
    This checks that the instance and node names are valid.
12405

12406
    """
12407
    instance_name = self.op.instance_name
12408

    
12409
    self.instance = self.cfg.GetInstanceInfo(instance_name)
12410
    assert self.instance is not None, \
12411
          "Cannot retrieve locked instance %s" % self.op.instance_name
12412
    _CheckNodeOnline(self, self.instance.primary_node)
12413

    
12414
    if (self.op.remove_instance and
12415
        self.instance.admin_state == constants.ADMINST_UP and
12416
        not self.op.shutdown):
12417
      raise errors.OpPrereqError("Can not remove instance without shutting it"
12418
                                 " down before")
12419

    
12420
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
12421
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
12422
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
12423
      assert self.dst_node is not None
12424

    
12425
      _CheckNodeOnline(self, self.dst_node.name)
12426
      _CheckNodeNotDrained(self, self.dst_node.name)
12427

    
12428
      self._cds = None
12429
      self.dest_disk_info = None
12430
      self.dest_x509_ca = None
12431

    
12432
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
12433
      self.dst_node = None
12434

    
12435
      if len(self.op.target_node) != len(self.instance.disks):
12436
        raise errors.OpPrereqError(("Received destination information for %s"
12437
                                    " disks, but instance %s has %s disks") %
12438
                                   (len(self.op.target_node), instance_name,
12439
                                    len(self.instance.disks)),
12440
                                   errors.ECODE_INVAL)
12441

    
12442
      cds = _GetClusterDomainSecret()
12443

    
12444
      # Check X509 key name
12445
      try:
12446
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
12447
      except (TypeError, ValueError), err:
12448
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
12449

    
12450
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
12451
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
12452
                                   errors.ECODE_INVAL)
12453

    
12454
      # Load and verify CA
12455
      try:
12456
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
12457
      except OpenSSL.crypto.Error, err:
12458
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
12459
                                   (err, ), errors.ECODE_INVAL)
12460

    
12461
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
12462
      if errcode is not None:
12463
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
12464
                                   (msg, ), errors.ECODE_INVAL)
12465

    
12466
      self.dest_x509_ca = cert
12467

    
12468
      # Verify target information
12469
      disk_info = []
12470
      for idx, disk_data in enumerate(self.op.target_node):
12471
        try:
12472
          (host, port, magic) = \
12473
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
12474
        except errors.GenericError, err:
12475
          raise errors.OpPrereqError("Target info for disk %s: %s" %
12476
                                     (idx, err), errors.ECODE_INVAL)
12477

    
12478
        disk_info.append((host, port, magic))
12479

    
12480
      assert len(disk_info) == len(self.op.target_node)
12481
      self.dest_disk_info = disk_info
12482

    
12483
    else:
12484
      raise errors.ProgrammerError("Unhandled export mode %r" %
12485
                                   self.op.mode)
12486

    
12487
    # instance disk type verification
12488
    # TODO: Implement export support for file-based disks
12489
    for disk in self.instance.disks:
12490
      if disk.dev_type == constants.LD_FILE:
12491
        raise errors.OpPrereqError("Export not supported for instances with"
12492
                                   " file-based disks", errors.ECODE_INVAL)
12493

    
12494
  def _CleanupExports(self, feedback_fn):
12495
    """Removes exports of current instance from all other nodes.
12496

12497
    If an instance in a cluster with nodes A..D was exported to node C, its
12498
    exports will be removed from the nodes A, B and D.
12499

12500
    """
12501
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
12502

    
12503
    nodelist = self.cfg.GetNodeList()
12504
    nodelist.remove(self.dst_node.name)
12505

    
12506
    # on one-node clusters nodelist will be empty after the removal
12507
    # if we proceed the backup would be removed because OpBackupQuery
12508
    # substitutes an empty list with the full cluster node list.
12509
    iname = self.instance.name
12510
    if nodelist:
12511
      feedback_fn("Removing old exports for instance %s" % iname)
12512
      exportlist = self.rpc.call_export_list(nodelist)
12513
      for node in exportlist:
12514
        if exportlist[node].fail_msg:
12515
          continue
12516
        if iname in exportlist[node].payload:
12517
          msg = self.rpc.call_export_remove(node, iname).fail_msg
12518
          if msg:
12519
            self.LogWarning("Could not remove older export for instance %s"
12520
                            " on node %s: %s", iname, node, msg)
12521

    
12522
  def Exec(self, feedback_fn):
12523
    """Export an instance to an image in the cluster.
12524

12525
    """
12526
    assert self.op.mode in constants.EXPORT_MODES
12527

    
12528
    instance = self.instance
12529
    src_node = instance.primary_node
12530

    
12531
    if self.op.shutdown:
12532
      # shutdown the instance, but not the disks
12533
      feedback_fn("Shutting down instance %s" % instance.name)
12534
      result = self.rpc.call_instance_shutdown(src_node, instance,
12535
                                               self.op.shutdown_timeout)
12536
      # TODO: Maybe ignore failures if ignore_remove_failures is set
12537
      result.Raise("Could not shutdown instance %s on"
12538
                   " node %s" % (instance.name, src_node))
12539

    
12540
    # set the disks ID correctly since call_instance_start needs the
12541
    # correct drbd minor to create the symlinks
12542
    for disk in instance.disks:
12543
      self.cfg.SetDiskID(disk, src_node)
12544

    
12545
    activate_disks = (instance.admin_state != constants.ADMINST_UP)
12546

    
12547
    if activate_disks:
12548
      # Activate the instance disks if we'exporting a stopped instance
12549
      feedback_fn("Activating disks for %s" % instance.name)
12550
      _StartInstanceDisks(self, instance, None)
12551

    
12552
    try:
12553
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
12554
                                                     instance)
12555

    
12556
      helper.CreateSnapshots()
12557
      try:
12558
        if (self.op.shutdown and
12559
            instance.admin_state == constants.ADMINST_UP and
12560
            not self.op.remove_instance):
12561
          assert not activate_disks
12562
          feedback_fn("Starting instance %s" % instance.name)
12563
          result = self.rpc.call_instance_start(src_node,
12564
                                                (instance, None, None), False)
12565
          msg = result.fail_msg
12566
          if msg:
12567
            feedback_fn("Failed to start instance: %s" % msg)
12568
            _ShutdownInstanceDisks(self, instance)
12569
            raise errors.OpExecError("Could not start instance: %s" % msg)
12570

    
12571
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
12572
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
12573
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
12574
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
12575
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
12576

    
12577
          (key_name, _, _) = self.x509_key_name
12578

    
12579
          dest_ca_pem = \
12580
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
12581
                                            self.dest_x509_ca)
12582

    
12583
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
12584
                                                     key_name, dest_ca_pem,
12585
                                                     timeouts)
12586
      finally:
12587
        helper.Cleanup()
12588

    
12589
      # Check for backwards compatibility
12590
      assert len(dresults) == len(instance.disks)
12591
      assert compat.all(isinstance(i, bool) for i in dresults), \
12592
             "Not all results are boolean: %r" % dresults
12593

    
12594
    finally:
12595
      if activate_disks:
12596
        feedback_fn("Deactivating disks for %s" % instance.name)
12597
        _ShutdownInstanceDisks(self, instance)
12598

    
12599
    if not (compat.all(dresults) and fin_resu):
12600
      failures = []
12601
      if not fin_resu:
12602
        failures.append("export finalization")
12603
      if not compat.all(dresults):
12604
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
12605
                               if not dsk)
12606
        failures.append("disk export: disk(s) %s" % fdsk)
12607

    
12608
      raise errors.OpExecError("Export failed, errors in %s" %
12609
                               utils.CommaJoin(failures))
12610

    
12611
    # At this point, the export was successful, we can cleanup/finish
12612

    
12613
    # Remove instance if requested
12614
    if self.op.remove_instance:
12615
      feedback_fn("Removing instance %s" % instance.name)
12616
      _RemoveInstance(self, feedback_fn, instance,
12617
                      self.op.ignore_remove_failures)
12618

    
12619
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
12620
      self._CleanupExports(feedback_fn)
12621

    
12622
    return fin_resu, dresults
12623

    
12624

    
12625
class LUBackupRemove(NoHooksLU):
12626
  """Remove exports related to the named instance.
12627

12628
  """
12629
  REQ_BGL = False
12630

    
12631
  def ExpandNames(self):
12632
    self.needed_locks = {}
12633
    # We need all nodes to be locked in order for RemoveExport to work, but we
12634
    # don't need to lock the instance itself, as nothing will happen to it (and
12635
    # we can remove exports also for a removed instance)
12636
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
12637

    
12638
  def Exec(self, feedback_fn):
12639
    """Remove any export.
12640

12641
    """
12642
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
12643
    # If the instance was not found we'll try with the name that was passed in.
12644
    # This will only work if it was an FQDN, though.
12645
    fqdn_warn = False
12646
    if not instance_name:
12647
      fqdn_warn = True
12648
      instance_name = self.op.instance_name
12649

    
12650
    locked_nodes = self.owned_locks(locking.LEVEL_NODE)
12651
    exportlist = self.rpc.call_export_list(locked_nodes)
12652
    found = False
12653
    for node in exportlist:
12654
      msg = exportlist[node].fail_msg
12655
      if msg:
12656
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
12657
        continue
12658
      if instance_name in exportlist[node].payload:
12659
        found = True
12660
        result = self.rpc.call_export_remove(node, instance_name)
12661
        msg = result.fail_msg
12662
        if msg:
12663
          logging.error("Could not remove export for instance %s"
12664
                        " on node %s: %s", instance_name, node, msg)
12665

    
12666
    if fqdn_warn and not found:
12667
      feedback_fn("Export not found. If trying to remove an export belonging"
12668
                  " to a deleted instance please use its Fully Qualified"
12669
                  " Domain Name.")
12670

    
12671

    
12672
class LUGroupAdd(LogicalUnit):
12673
  """Logical unit for creating node groups.
12674

12675
  """
12676
  HPATH = "group-add"
12677
  HTYPE = constants.HTYPE_GROUP
12678
  REQ_BGL = False
12679

    
12680
  def ExpandNames(self):
12681
    # We need the new group's UUID here so that we can create and acquire the
12682
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
12683
    # that it should not check whether the UUID exists in the configuration.
12684
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
12685
    self.needed_locks = {}
12686
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
12687

    
12688
  def CheckPrereq(self):
12689
    """Check prerequisites.
12690

12691
    This checks that the given group name is not an existing node group
12692
    already.
12693

12694
    """
12695
    try:
12696
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12697
    except errors.OpPrereqError:
12698
      pass
12699
    else:
12700
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
12701
                                 " node group (UUID: %s)" %
12702
                                 (self.op.group_name, existing_uuid),
12703
                                 errors.ECODE_EXISTS)
12704

    
12705
    if self.op.ndparams:
12706
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
12707

    
12708
    if self.op.diskparams:
12709
      for templ in constants.DISK_TEMPLATES:
12710
        if templ not in self.op.diskparams:
12711
          self.op.diskparams[templ] = {}
12712
        utils.ForceDictType(self.op.diskparams[templ], constants.DISK_DT_TYPES)
12713
    else:
12714
      self.op.diskparams = self.cfg.GetClusterInfo().diskparams
12715

    
12716
  def BuildHooksEnv(self):
12717
    """Build hooks env.
12718

12719
    """
12720
    return {
12721
      "GROUP_NAME": self.op.group_name,
12722
      }
12723

    
12724
  def BuildHooksNodes(self):
12725
    """Build hooks nodes.
12726

12727
    """
12728
    mn = self.cfg.GetMasterNode()
12729
    return ([mn], [mn])
12730

    
12731
  def Exec(self, feedback_fn):
12732
    """Add the node group to the cluster.
12733

12734
    """
12735
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
12736
                                  uuid=self.group_uuid,
12737
                                  alloc_policy=self.op.alloc_policy,
12738
                                  ndparams=self.op.ndparams,
12739
                                  diskparams=self.op.diskparams)
12740

    
12741
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
12742
    del self.remove_locks[locking.LEVEL_NODEGROUP]
12743

    
12744

    
12745
class LUGroupAssignNodes(NoHooksLU):
12746
  """Logical unit for assigning nodes to groups.
12747

12748
  """
12749
  REQ_BGL = False
12750

    
12751
  def ExpandNames(self):
12752
    # These raise errors.OpPrereqError on their own:
12753
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12754
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
12755

    
12756
    # We want to lock all the affected nodes and groups. We have readily
12757
    # available the list of nodes, and the *destination* group. To gather the
12758
    # list of "source" groups, we need to fetch node information later on.
12759
    self.needed_locks = {
12760
      locking.LEVEL_NODEGROUP: set([self.group_uuid]),
12761
      locking.LEVEL_NODE: self.op.nodes,
12762
      }
12763

    
12764
  def DeclareLocks(self, level):
12765
    if level == locking.LEVEL_NODEGROUP:
12766
      assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
12767

    
12768
      # Try to get all affected nodes' groups without having the group or node
12769
      # lock yet. Needs verification later in the code flow.
12770
      groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
12771

    
12772
      self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
12773

    
12774
  def CheckPrereq(self):
12775
    """Check prerequisites.
12776

12777
    """
12778
    assert self.needed_locks[locking.LEVEL_NODEGROUP]
12779
    assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
12780
            frozenset(self.op.nodes))
12781

    
12782
    expected_locks = (set([self.group_uuid]) |
12783
                      self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
12784
    actual_locks = self.owned_locks(locking.LEVEL_NODEGROUP)
12785
    if actual_locks != expected_locks:
12786
      raise errors.OpExecError("Nodes changed groups since locks were acquired,"
12787
                               " current groups are '%s', used to be '%s'" %
12788
                               (utils.CommaJoin(expected_locks),
12789
                                utils.CommaJoin(actual_locks)))
12790

    
12791
    self.node_data = self.cfg.GetAllNodesInfo()
12792
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
12793
    instance_data = self.cfg.GetAllInstancesInfo()
12794

    
12795
    if self.group is None:
12796
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
12797
                               (self.op.group_name, self.group_uuid))
12798

    
12799
    (new_splits, previous_splits) = \
12800
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
12801
                                             for node in self.op.nodes],
12802
                                            self.node_data, instance_data)
12803

    
12804
    if new_splits:
12805
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
12806

    
12807
      if not self.op.force:
12808
        raise errors.OpExecError("The following instances get split by this"
12809
                                 " change and --force was not given: %s" %
12810
                                 fmt_new_splits)
12811
      else:
12812
        self.LogWarning("This operation will split the following instances: %s",
12813
                        fmt_new_splits)
12814

    
12815
        if previous_splits:
12816
          self.LogWarning("In addition, these already-split instances continue"
12817
                          " to be split across groups: %s",
12818
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
12819

    
12820
  def Exec(self, feedback_fn):
12821
    """Assign nodes to a new group.
12822

12823
    """
12824
    mods = [(node_name, self.group_uuid) for node_name in self.op.nodes]
12825

    
12826
    self.cfg.AssignGroupNodes(mods)
12827

    
12828
  @staticmethod
12829
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
12830
    """Check for split instances after a node assignment.
12831

12832
    This method considers a series of node assignments as an atomic operation,
12833
    and returns information about split instances after applying the set of
12834
    changes.
12835

12836
    In particular, it returns information about newly split instances, and
12837
    instances that were already split, and remain so after the change.
12838

12839
    Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
12840
    considered.
12841

12842
    @type changes: list of (node_name, new_group_uuid) pairs.
12843
    @param changes: list of node assignments to consider.
12844
    @param node_data: a dict with data for all nodes
12845
    @param instance_data: a dict with all instances to consider
12846
    @rtype: a two-tuple
12847
    @return: a list of instances that were previously okay and result split as a
12848
      consequence of this change, and a list of instances that were previously
12849
      split and this change does not fix.
12850

12851
    """
12852
    changed_nodes = dict((node, group) for node, group in changes
12853
                         if node_data[node].group != group)
12854

    
12855
    all_split_instances = set()
12856
    previously_split_instances = set()
12857

    
12858
    def InstanceNodes(instance):
12859
      return [instance.primary_node] + list(instance.secondary_nodes)
12860

    
12861
    for inst in instance_data.values():
12862
      if inst.disk_template not in constants.DTS_INT_MIRROR:
12863
        continue
12864

    
12865
      instance_nodes = InstanceNodes(inst)
12866

    
12867
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
12868
        previously_split_instances.add(inst.name)
12869

    
12870
      if len(set(changed_nodes.get(node, node_data[node].group)
12871
                 for node in instance_nodes)) > 1:
12872
        all_split_instances.add(inst.name)
12873

    
12874
    return (list(all_split_instances - previously_split_instances),
12875
            list(previously_split_instances & all_split_instances))
12876

    
12877

    
12878
class _GroupQuery(_QueryBase):
12879
  FIELDS = query.GROUP_FIELDS
12880

    
12881
  def ExpandNames(self, lu):
12882
    lu.needed_locks = {}
12883

    
12884
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
12885
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
12886

    
12887
    if not self.names:
12888
      self.wanted = [name_to_uuid[name]
12889
                     for name in utils.NiceSort(name_to_uuid.keys())]
12890
    else:
12891
      # Accept names to be either names or UUIDs.
12892
      missing = []
12893
      self.wanted = []
12894
      all_uuid = frozenset(self._all_groups.keys())
12895

    
12896
      for name in self.names:
12897
        if name in all_uuid:
12898
          self.wanted.append(name)
12899
        elif name in name_to_uuid:
12900
          self.wanted.append(name_to_uuid[name])
12901
        else:
12902
          missing.append(name)
12903

    
12904
      if missing:
12905
        raise errors.OpPrereqError("Some groups do not exist: %s" %
12906
                                   utils.CommaJoin(missing),
12907
                                   errors.ECODE_NOENT)
12908

    
12909
  def DeclareLocks(self, lu, level):
12910
    pass
12911

    
12912
  def _GetQueryData(self, lu):
12913
    """Computes the list of node groups and their attributes.
12914

12915
    """
12916
    do_nodes = query.GQ_NODE in self.requested_data
12917
    do_instances = query.GQ_INST in self.requested_data
12918

    
12919
    group_to_nodes = None
12920
    group_to_instances = None
12921

    
12922
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
12923
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
12924
    # latter GetAllInstancesInfo() is not enough, for we have to go through
12925
    # instance->node. Hence, we will need to process nodes even if we only need
12926
    # instance information.
12927
    if do_nodes or do_instances:
12928
      all_nodes = lu.cfg.GetAllNodesInfo()
12929
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
12930
      node_to_group = {}
12931

    
12932
      for node in all_nodes.values():
12933
        if node.group in group_to_nodes:
12934
          group_to_nodes[node.group].append(node.name)
12935
          node_to_group[node.name] = node.group
12936

    
12937
      if do_instances:
12938
        all_instances = lu.cfg.GetAllInstancesInfo()
12939
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
12940

    
12941
        for instance in all_instances.values():
12942
          node = instance.primary_node
12943
          if node in node_to_group:
12944
            group_to_instances[node_to_group[node]].append(instance.name)
12945

    
12946
        if not do_nodes:
12947
          # Do not pass on node information if it was not requested.
12948
          group_to_nodes = None
12949

    
12950
    return query.GroupQueryData([self._all_groups[uuid]
12951
                                 for uuid in self.wanted],
12952
                                group_to_nodes, group_to_instances)
12953

    
12954

    
12955
class LUGroupQuery(NoHooksLU):
12956
  """Logical unit for querying node groups.
12957

12958
  """
12959
  REQ_BGL = False
12960

    
12961
  def CheckArguments(self):
12962
    self.gq = _GroupQuery(qlang.MakeSimpleFilter("name", self.op.names),
12963
                          self.op.output_fields, False)
12964

    
12965
  def ExpandNames(self):
12966
    self.gq.ExpandNames(self)
12967

    
12968
  def DeclareLocks(self, level):
12969
    self.gq.DeclareLocks(self, level)
12970

    
12971
  def Exec(self, feedback_fn):
12972
    return self.gq.OldStyleQuery(self)
12973

    
12974

    
12975
class LUGroupSetParams(LogicalUnit):
12976
  """Modifies the parameters of a node group.
12977

12978
  """
12979
  HPATH = "group-modify"
12980
  HTYPE = constants.HTYPE_GROUP
12981
  REQ_BGL = False
12982

    
12983
  def CheckArguments(self):
12984
    all_changes = [
12985
      self.op.ndparams,
12986
      self.op.diskparams,
12987
      self.op.alloc_policy,
12988
      self.op.hv_state,
12989
      self.op.disk_state
12990
      ]
12991

    
12992
    if all_changes.count(None) == len(all_changes):
12993
      raise errors.OpPrereqError("Please pass at least one modification",
12994
                                 errors.ECODE_INVAL)
12995

    
12996
  def ExpandNames(self):
12997
    # This raises errors.OpPrereqError on its own:
12998
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12999

    
13000
    self.needed_locks = {
13001
      locking.LEVEL_NODEGROUP: [self.group_uuid],
13002
      }
13003

    
13004
  def CheckPrereq(self):
13005
    """Check prerequisites.
13006

13007
    """
13008
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
13009

    
13010
    if self.group is None:
13011
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
13012
                               (self.op.group_name, self.group_uuid))
13013

    
13014
    if self.op.ndparams:
13015
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
13016
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
13017
      self.new_ndparams = new_ndparams
13018

    
13019
    if self.op.diskparams:
13020
      self.new_diskparams = dict()
13021
      for templ in constants.DISK_TEMPLATES:
13022
        if templ not in self.op.diskparams:
13023
          self.op.diskparams[templ] = {}
13024
        new_templ_params = _GetUpdatedParams(self.group.diskparams[templ],
13025
                                             self.op.diskparams[templ])
13026
        utils.ForceDictType(new_templ_params, constants.DISK_DT_TYPES)
13027
        self.new_diskparams[templ] = new_templ_params
13028

    
13029
    if self.op.hv_state:
13030
      self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
13031
                                                 self.group.hv_state_static)
13032

    
13033
    if self.op.disk_state:
13034
      self.new_disk_state = \
13035
        _MergeAndVerifyDiskState(self.op.disk_state,
13036
                                 self.group.disk_state_static)
13037

    
13038
  def BuildHooksEnv(self):
13039
    """Build hooks env.
13040

13041
    """
13042
    return {
13043
      "GROUP_NAME": self.op.group_name,
13044
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
13045
      }
13046

    
13047
  def BuildHooksNodes(self):
13048
    """Build hooks nodes.
13049

13050
    """
13051
    mn = self.cfg.GetMasterNode()
13052
    return ([mn], [mn])
13053

    
13054
  def Exec(self, feedback_fn):
13055
    """Modifies the node group.
13056

13057
    """
13058
    result = []
13059

    
13060
    if self.op.ndparams:
13061
      self.group.ndparams = self.new_ndparams
13062
      result.append(("ndparams", str(self.group.ndparams)))
13063

    
13064
    if self.op.diskparams:
13065
      self.group.diskparams = self.new_diskparams
13066
      result.append(("diskparams", str(self.group.diskparams)))
13067

    
13068
    if self.op.alloc_policy:
13069
      self.group.alloc_policy = self.op.alloc_policy
13070

    
13071
    if self.op.hv_state:
13072
      self.group.hv_state_static = self.new_hv_state
13073

    
13074
    if self.op.disk_state:
13075
      self.group.disk_state_static = self.new_disk_state
13076

    
13077
    self.cfg.Update(self.group, feedback_fn)
13078
    return result
13079

    
13080

    
13081
class LUGroupRemove(LogicalUnit):
13082
  HPATH = "group-remove"
13083
  HTYPE = constants.HTYPE_GROUP
13084
  REQ_BGL = False
13085

    
13086
  def ExpandNames(self):
13087
    # This will raises errors.OpPrereqError on its own:
13088
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13089
    self.needed_locks = {
13090
      locking.LEVEL_NODEGROUP: [self.group_uuid],
13091
      }
13092

    
13093
  def CheckPrereq(self):
13094
    """Check prerequisites.
13095

13096
    This checks that the given group name exists as a node group, that is
13097
    empty (i.e., contains no nodes), and that is not the last group of the
13098
    cluster.
13099

13100
    """
13101
    # Verify that the group is empty.
13102
    group_nodes = [node.name
13103
                   for node in self.cfg.GetAllNodesInfo().values()
13104
                   if node.group == self.group_uuid]
13105

    
13106
    if group_nodes:
13107
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
13108
                                 " nodes: %s" %
13109
                                 (self.op.group_name,
13110
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
13111
                                 errors.ECODE_STATE)
13112

    
13113
    # Verify the cluster would not be left group-less.
13114
    if len(self.cfg.GetNodeGroupList()) == 1:
13115
      raise errors.OpPrereqError("Group '%s' is the only group,"
13116
                                 " cannot be removed" %
13117
                                 self.op.group_name,
13118
                                 errors.ECODE_STATE)
13119

    
13120
  def BuildHooksEnv(self):
13121
    """Build hooks env.
13122

13123
    """
13124
    return {
13125
      "GROUP_NAME": self.op.group_name,
13126
      }
13127

    
13128
  def BuildHooksNodes(self):
13129
    """Build hooks nodes.
13130

13131
    """
13132
    mn = self.cfg.GetMasterNode()
13133
    return ([mn], [mn])
13134

    
13135
  def Exec(self, feedback_fn):
13136
    """Remove the node group.
13137

13138
    """
13139
    try:
13140
      self.cfg.RemoveNodeGroup(self.group_uuid)
13141
    except errors.ConfigurationError:
13142
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
13143
                               (self.op.group_name, self.group_uuid))
13144

    
13145
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
13146

    
13147

    
13148
class LUGroupRename(LogicalUnit):
13149
  HPATH = "group-rename"
13150
  HTYPE = constants.HTYPE_GROUP
13151
  REQ_BGL = False
13152

    
13153
  def ExpandNames(self):
13154
    # This raises errors.OpPrereqError on its own:
13155
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13156

    
13157
    self.needed_locks = {
13158
      locking.LEVEL_NODEGROUP: [self.group_uuid],
13159
      }
13160

    
13161
  def CheckPrereq(self):
13162
    """Check prerequisites.
13163

13164
    Ensures requested new name is not yet used.
13165

13166
    """
13167
    try:
13168
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
13169
    except errors.OpPrereqError:
13170
      pass
13171
    else:
13172
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
13173
                                 " node group (UUID: %s)" %
13174
                                 (self.op.new_name, new_name_uuid),
13175
                                 errors.ECODE_EXISTS)
13176

    
13177
  def BuildHooksEnv(self):
13178
    """Build hooks env.
13179

13180
    """
13181
    return {
13182
      "OLD_NAME": self.op.group_name,
13183
      "NEW_NAME": self.op.new_name,
13184
      }
13185

    
13186
  def BuildHooksNodes(self):
13187
    """Build hooks nodes.
13188

13189
    """
13190
    mn = self.cfg.GetMasterNode()
13191

    
13192
    all_nodes = self.cfg.GetAllNodesInfo()
13193
    all_nodes.pop(mn, None)
13194

    
13195
    run_nodes = [mn]
13196
    run_nodes.extend(node.name for node in all_nodes.values()
13197
                     if node.group == self.group_uuid)
13198

    
13199
    return (run_nodes, run_nodes)
13200

    
13201
  def Exec(self, feedback_fn):
13202
    """Rename the node group.
13203

13204
    """
13205
    group = self.cfg.GetNodeGroup(self.group_uuid)
13206

    
13207
    if group is None:
13208
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
13209
                               (self.op.group_name, self.group_uuid))
13210

    
13211
    group.name = self.op.new_name
13212
    self.cfg.Update(group, feedback_fn)
13213

    
13214
    return self.op.new_name
13215

    
13216

    
13217
class LUGroupEvacuate(LogicalUnit):
13218
  HPATH = "group-evacuate"
13219
  HTYPE = constants.HTYPE_GROUP
13220
  REQ_BGL = False
13221

    
13222
  def ExpandNames(self):
13223
    # This raises errors.OpPrereqError on its own:
13224
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
13225

    
13226
    if self.op.target_groups:
13227
      self.req_target_uuids = map(self.cfg.LookupNodeGroup,
13228
                                  self.op.target_groups)
13229
    else:
13230
      self.req_target_uuids = []
13231

    
13232
    if self.group_uuid in self.req_target_uuids:
13233
      raise errors.OpPrereqError("Group to be evacuated (%s) can not be used"
13234
                                 " as a target group (targets are %s)" %
13235
                                 (self.group_uuid,
13236
                                  utils.CommaJoin(self.req_target_uuids)),
13237
                                 errors.ECODE_INVAL)
13238

    
13239
    self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
13240

    
13241
    self.share_locks = _ShareAll()
13242
    self.needed_locks = {
13243
      locking.LEVEL_INSTANCE: [],
13244
      locking.LEVEL_NODEGROUP: [],
13245
      locking.LEVEL_NODE: [],
13246
      }
13247

    
13248
  def DeclareLocks(self, level):
13249
    if level == locking.LEVEL_INSTANCE:
13250
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
13251

    
13252
      # Lock instances optimistically, needs verification once node and group
13253
      # locks have been acquired
13254
      self.needed_locks[locking.LEVEL_INSTANCE] = \
13255
        self.cfg.GetNodeGroupInstances(self.group_uuid)
13256

    
13257
    elif level == locking.LEVEL_NODEGROUP:
13258
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
13259

    
13260
      if self.req_target_uuids:
13261
        lock_groups = set([self.group_uuid] + self.req_target_uuids)
13262

    
13263
        # Lock all groups used by instances optimistically; this requires going
13264
        # via the node before it's locked, requiring verification later on
13265
        lock_groups.update(group_uuid
13266
                           for instance_name in
13267
                             self.owned_locks(locking.LEVEL_INSTANCE)
13268
                           for group_uuid in
13269
                             self.cfg.GetInstanceNodeGroups(instance_name))
13270
      else:
13271
        # No target groups, need to lock all of them
13272
        lock_groups = locking.ALL_SET
13273

    
13274
      self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
13275

    
13276
    elif level == locking.LEVEL_NODE:
13277
      # This will only lock the nodes in the group to be evacuated which
13278
      # contain actual instances
13279
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
13280
      self._LockInstancesNodes()
13281

    
13282
      # Lock all nodes in group to be evacuated and target groups
13283
      owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
13284
      assert self.group_uuid in owned_groups
13285
      member_nodes = [node_name
13286
                      for group in owned_groups
13287
                      for node_name in self.cfg.GetNodeGroup(group).members]
13288
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
13289

    
13290
  def CheckPrereq(self):
13291
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
13292
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
13293
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
13294

    
13295
    assert owned_groups.issuperset(self.req_target_uuids)
13296
    assert self.group_uuid in owned_groups
13297

    
13298
    # Check if locked instances are still correct
13299
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
13300

    
13301
    # Get instance information
13302
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
13303

    
13304
    # Check if node groups for locked instances are still correct
13305
    for instance_name in owned_instances:
13306
      inst = self.instances[instance_name]
13307
      assert owned_nodes.issuperset(inst.all_nodes), \
13308
        "Instance %s's nodes changed while we kept the lock" % instance_name
13309

    
13310
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
13311
                                             owned_groups)
13312

    
13313
      assert self.group_uuid in inst_groups, \
13314
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
13315

    
13316
    if self.req_target_uuids:
13317
      # User requested specific target groups
13318
      self.target_uuids = self.req_target_uuids
13319
    else:
13320
      # All groups except the one to be evacuated are potential targets
13321
      self.target_uuids = [group_uuid for group_uuid in owned_groups
13322
                           if group_uuid != self.group_uuid]
13323

    
13324
      if not self.target_uuids:
13325
        raise errors.OpPrereqError("There are no possible target groups",
13326
                                   errors.ECODE_INVAL)
13327

    
13328
  def BuildHooksEnv(self):
13329
    """Build hooks env.
13330

13331
    """
13332
    return {
13333
      "GROUP_NAME": self.op.group_name,
13334
      "TARGET_GROUPS": " ".join(self.target_uuids),
13335
      }
13336

    
13337
  def BuildHooksNodes(self):
13338
    """Build hooks nodes.
13339

13340
    """
13341
    mn = self.cfg.GetMasterNode()
13342

    
13343
    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
13344

    
13345
    run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members
13346

    
13347
    return (run_nodes, run_nodes)
13348

    
13349
  def Exec(self, feedback_fn):
13350
    instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
13351

    
13352
    assert self.group_uuid not in self.target_uuids
13353

    
13354
    ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
13355
                     instances=instances, target_groups=self.target_uuids)
13356

    
13357
    ial.Run(self.op.iallocator)
13358

    
13359
    if not ial.success:
13360
      raise errors.OpPrereqError("Can't compute group evacuation using"
13361
                                 " iallocator '%s': %s" %
13362
                                 (self.op.iallocator, ial.info),
13363
                                 errors.ECODE_NORES)
13364

    
13365
    jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
13366

    
13367
    self.LogInfo("Iallocator returned %s job(s) for evacuating node group %s",
13368
                 len(jobs), self.op.group_name)
13369

    
13370
    return ResultWithJobs(jobs)
13371

    
13372

    
13373
class TagsLU(NoHooksLU): # pylint: disable=W0223
13374
  """Generic tags LU.
13375

13376
  This is an abstract class which is the parent of all the other tags LUs.
13377

13378
  """
13379
  def ExpandNames(self):
13380
    self.group_uuid = None
13381
    self.needed_locks = {}
13382
    if self.op.kind == constants.TAG_NODE:
13383
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
13384
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
13385
    elif self.op.kind == constants.TAG_INSTANCE:
13386
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
13387
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
13388
    elif self.op.kind == constants.TAG_NODEGROUP:
13389
      self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)
13390

    
13391
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
13392
    # not possible to acquire the BGL based on opcode parameters)
13393

    
13394
  def CheckPrereq(self):
13395
    """Check prerequisites.
13396

13397
    """
13398
    if self.op.kind == constants.TAG_CLUSTER:
13399
      self.target = self.cfg.GetClusterInfo()
13400
    elif self.op.kind == constants.TAG_NODE:
13401
      self.target = self.cfg.GetNodeInfo(self.op.name)
13402
    elif self.op.kind == constants.TAG_INSTANCE:
13403
      self.target = self.cfg.GetInstanceInfo(self.op.name)
13404
    elif self.op.kind == constants.TAG_NODEGROUP:
13405
      self.target = self.cfg.GetNodeGroup(self.group_uuid)
13406
    else:
13407
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
13408
                                 str(self.op.kind), errors.ECODE_INVAL)
13409

    
13410

    
13411
class LUTagsGet(TagsLU):
13412
  """Returns the tags of a given object.
13413

13414
  """
13415
  REQ_BGL = False
13416

    
13417
  def ExpandNames(self):
13418
    TagsLU.ExpandNames(self)
13419

    
13420
    # Share locks as this is only a read operation
13421
    self.share_locks = _ShareAll()
13422

    
13423
  def Exec(self, feedback_fn):
13424
    """Returns the tag list.
13425

13426
    """
13427
    return list(self.target.GetTags())
13428

    
13429

    
13430
class LUTagsSearch(NoHooksLU):
13431
  """Searches the tags for a given pattern.
13432

13433
  """
13434
  REQ_BGL = False
13435

    
13436
  def ExpandNames(self):
13437
    self.needed_locks = {}
13438

    
13439
  def CheckPrereq(self):
13440
    """Check prerequisites.
13441

13442
    This checks the pattern passed for validity by compiling it.
13443

13444
    """
13445
    try:
13446
      self.re = re.compile(self.op.pattern)
13447
    except re.error, err:
13448
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
13449
                                 (self.op.pattern, err), errors.ECODE_INVAL)
13450

    
13451
  def Exec(self, feedback_fn):
13452
    """Returns the tag list.
13453

13454
    """
13455
    cfg = self.cfg
13456
    tgts = [("/cluster", cfg.GetClusterInfo())]
13457
    ilist = cfg.GetAllInstancesInfo().values()
13458
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
13459
    nlist = cfg.GetAllNodesInfo().values()
13460
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
13461
    tgts.extend(("/nodegroup/%s" % n.name, n)
13462
                for n in cfg.GetAllNodeGroupsInfo().values())
13463
    results = []
13464
    for path, target in tgts:
13465
      for tag in target.GetTags():
13466
        if self.re.search(tag):
13467
          results.append((path, tag))
13468
    return results
13469

    
13470

    
13471
class LUTagsSet(TagsLU):
13472
  """Sets a tag on a given object.
13473

13474
  """
13475
  REQ_BGL = False
13476

    
13477
  def CheckPrereq(self):
13478
    """Check prerequisites.
13479

13480
    This checks the type and length of the tag name and value.
13481

13482
    """
13483
    TagsLU.CheckPrereq(self)
13484
    for tag in self.op.tags:
13485
      objects.TaggableObject.ValidateTag(tag)
13486

    
13487
  def Exec(self, feedback_fn):
13488
    """Sets the tag.
13489

13490
    """
13491
    try:
13492
      for tag in self.op.tags:
13493
        self.target.AddTag(tag)
13494
    except errors.TagError, err:
13495
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
13496
    self.cfg.Update(self.target, feedback_fn)
13497

    
13498

    
13499
class LUTagsDel(TagsLU):
13500
  """Delete a list of tags from a given object.
13501

13502
  """
13503
  REQ_BGL = False
13504

    
13505
  def CheckPrereq(self):
13506
    """Check prerequisites.
13507

13508
    This checks that we have the given tag.
13509

13510
    """
13511
    TagsLU.CheckPrereq(self)
13512
    for tag in self.op.tags:
13513
      objects.TaggableObject.ValidateTag(tag)
13514
    del_tags = frozenset(self.op.tags)
13515
    cur_tags = self.target.GetTags()
13516

    
13517
    diff_tags = del_tags - cur_tags
13518
    if diff_tags:
13519
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
13520
      raise errors.OpPrereqError("Tag(s) %s not found" %
13521
                                 (utils.CommaJoin(diff_names), ),
13522
                                 errors.ECODE_NOENT)
13523

    
13524
  def Exec(self, feedback_fn):
13525
    """Remove the tag from the object.
13526

13527
    """
13528
    for tag in self.op.tags:
13529
      self.target.RemoveTag(tag)
13530
    self.cfg.Update(self.target, feedback_fn)
13531

    
13532

    
13533
class LUTestDelay(NoHooksLU):
13534
  """Sleep for a specified amount of time.
13535

13536
  This LU sleeps on the master and/or nodes for a specified amount of
13537
  time.
13538

13539
  """
13540
  REQ_BGL = False
13541

    
13542
  def ExpandNames(self):
13543
    """Expand names and set required locks.
13544

13545
    This expands the node list, if any.
13546

13547
    """
13548
    self.needed_locks = {}
13549
    if self.op.on_nodes:
13550
      # _GetWantedNodes can be used here, but is not always appropriate to use
13551
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
13552
      # more information.
13553
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
13554
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
13555

    
13556
  def _TestDelay(self):
13557
    """Do the actual sleep.
13558

13559
    """
13560
    if self.op.on_master:
13561
      if not utils.TestDelay(self.op.duration):
13562
        raise errors.OpExecError("Error during master delay test")
13563
    if self.op.on_nodes:
13564
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
13565
      for node, node_result in result.items():
13566
        node_result.Raise("Failure during rpc call to node %s" % node)
13567

    
13568
  def Exec(self, feedback_fn):
13569
    """Execute the test delay opcode, with the wanted repetitions.
13570

13571
    """
13572
    if self.op.repeat == 0:
13573
      self._TestDelay()
13574
    else:
13575
      top_value = self.op.repeat - 1
13576
      for i in range(self.op.repeat):
13577
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
13578
        self._TestDelay()
13579

    
13580

    
13581
class LUTestJqueue(NoHooksLU):
13582
  """Utility LU to test some aspects of the job queue.
13583

13584
  """
13585
  REQ_BGL = False
13586

    
13587
  # Must be lower than default timeout for WaitForJobChange to see whether it
13588
  # notices changed jobs
13589
  _CLIENT_CONNECT_TIMEOUT = 20.0
13590
  _CLIENT_CONFIRM_TIMEOUT = 60.0
13591

    
13592
  @classmethod
13593
  def _NotifyUsingSocket(cls, cb, errcls):
13594
    """Opens a Unix socket and waits for another program to connect.
13595

13596
    @type cb: callable
13597
    @param cb: Callback to send socket name to client
13598
    @type errcls: class
13599
    @param errcls: Exception class to use for errors
13600

13601
    """
13602
    # Using a temporary directory as there's no easy way to create temporary
13603
    # sockets without writing a custom loop around tempfile.mktemp and
13604
    # socket.bind
13605
    tmpdir = tempfile.mkdtemp()
13606
    try:
13607
      tmpsock = utils.PathJoin(tmpdir, "sock")
13608

    
13609
      logging.debug("Creating temporary socket at %s", tmpsock)
13610
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
13611
      try:
13612
        sock.bind(tmpsock)
13613
        sock.listen(1)
13614

    
13615
        # Send details to client
13616
        cb(tmpsock)
13617

    
13618
        # Wait for client to connect before continuing
13619
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
13620
        try:
13621
          (conn, _) = sock.accept()
13622
        except socket.error, err:
13623
          raise errcls("Client didn't connect in time (%s)" % err)
13624
      finally:
13625
        sock.close()
13626
    finally:
13627
      # Remove as soon as client is connected
13628
      shutil.rmtree(tmpdir)
13629

    
13630
    # Wait for client to close
13631
    try:
13632
      try:
13633
        # pylint: disable=E1101
13634
        # Instance of '_socketobject' has no ... member
13635
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
13636
        conn.recv(1)
13637
      except socket.error, err:
13638
        raise errcls("Client failed to confirm notification (%s)" % err)
13639
    finally:
13640
      conn.close()
13641

    
13642
  def _SendNotification(self, test, arg, sockname):
13643
    """Sends a notification to the client.
13644

13645
    @type test: string
13646
    @param test: Test name
13647
    @param arg: Test argument (depends on test)
13648
    @type sockname: string
13649
    @param sockname: Socket path
13650

13651
    """
13652
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
13653

    
13654
  def _Notify(self, prereq, test, arg):
13655
    """Notifies the client of a test.
13656

13657
    @type prereq: bool
13658
    @param prereq: Whether this is a prereq-phase test
13659
    @type test: string
13660
    @param test: Test name
13661
    @param arg: Test argument (depends on test)
13662

13663
    """
13664
    if prereq:
13665
      errcls = errors.OpPrereqError
13666
    else:
13667
      errcls = errors.OpExecError
13668

    
13669
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
13670
                                                  test, arg),
13671
                                   errcls)
13672

    
13673
  def CheckArguments(self):
13674
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
13675
    self.expandnames_calls = 0
13676

    
13677
  def ExpandNames(self):
13678
    checkargs_calls = getattr(self, "checkargs_calls", 0)
13679
    if checkargs_calls < 1:
13680
      raise errors.ProgrammerError("CheckArguments was not called")
13681

    
13682
    self.expandnames_calls += 1
13683

    
13684
    if self.op.notify_waitlock:
13685
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
13686

    
13687
    self.LogInfo("Expanding names")
13688

    
13689
    # Get lock on master node (just to get a lock, not for a particular reason)
13690
    self.needed_locks = {
13691
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
13692
      }
13693

    
13694
  def Exec(self, feedback_fn):
13695
    if self.expandnames_calls < 1:
13696
      raise errors.ProgrammerError("ExpandNames was not called")
13697

    
13698
    if self.op.notify_exec:
13699
      self._Notify(False, constants.JQT_EXEC, None)
13700

    
13701
    self.LogInfo("Executing")
13702

    
13703
    if self.op.log_messages:
13704
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
13705
      for idx, msg in enumerate(self.op.log_messages):
13706
        self.LogInfo("Sending log message %s", idx + 1)
13707
        feedback_fn(constants.JQT_MSGPREFIX + msg)
13708
        # Report how many test messages have been sent
13709
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
13710

    
13711
    if self.op.fail:
13712
      raise errors.OpExecError("Opcode failure was requested")
13713

    
13714
    return True
13715

    
13716

    
13717
class IAllocator(object):
13718
  """IAllocator framework.
13719

13720
  An IAllocator instance has three sets of attributes:
13721
    - cfg that is needed to query the cluster
13722
    - input data (all members of the _KEYS class attribute are required)
13723
    - four buffer attributes (in|out_data|text), that represent the
13724
      input (to the external script) in text and data structure format,
13725
      and the output from it, again in two formats
13726
    - the result variables from the script (success, info, nodes) for
13727
      easy usage
13728

13729
  """
13730
  # pylint: disable=R0902
13731
  # lots of instance attributes
13732

    
13733
  def __init__(self, cfg, rpc_runner, mode, **kwargs):
13734
    self.cfg = cfg
13735
    self.rpc = rpc_runner
13736
    # init buffer variables
13737
    self.in_text = self.out_text = self.in_data = self.out_data = None
13738
    # init all input fields so that pylint is happy
13739
    self.mode = mode
13740
    self.memory = self.disks = self.disk_template = None
13741
    self.os = self.tags = self.nics = self.vcpus = None
13742
    self.hypervisor = None
13743
    self.relocate_from = None
13744
    self.name = None
13745
    self.instances = None
13746
    self.evac_mode = None
13747
    self.target_groups = []
13748
    # computed fields
13749
    self.required_nodes = None
13750
    # init result fields
13751
    self.success = self.info = self.result = None
13752

    
13753
    try:
13754
      (fn, keydata, self._result_check) = self._MODE_DATA[self.mode]
13755
    except KeyError:
13756
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
13757
                                   " IAllocator" % self.mode)
13758

    
13759
    keyset = [n for (n, _) in keydata]
13760

    
13761
    for key in kwargs:
13762
      if key not in keyset:
13763
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
13764
                                     " IAllocator" % key)
13765
      setattr(self, key, kwargs[key])
13766

    
13767
    for key in keyset:
13768
      if key not in kwargs:
13769
        raise errors.ProgrammerError("Missing input parameter '%s' to"
13770
                                     " IAllocator" % key)
13771
    self._BuildInputData(compat.partial(fn, self), keydata)
13772

    
13773
  def _ComputeClusterData(self):
13774
    """Compute the generic allocator input data.
13775

13776
    This is the data that is independent of the actual operation.
13777

13778
    """
13779
    cfg = self.cfg
13780
    cluster_info = cfg.GetClusterInfo()
13781
    # cluster data
13782
    data = {
13783
      "version": constants.IALLOCATOR_VERSION,
13784
      "cluster_name": cfg.GetClusterName(),
13785
      "cluster_tags": list(cluster_info.GetTags()),
13786
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
13787
      # we don't have job IDs
13788
      }
13789
    ninfo = cfg.GetAllNodesInfo()
13790
    iinfo = cfg.GetAllInstancesInfo().values()
13791
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
13792

    
13793
    # node data
13794
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
13795

    
13796
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
13797
      hypervisor_name = self.hypervisor
13798
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
13799
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
13800
    else:
13801
      hypervisor_name = cluster_info.primary_hypervisor
13802

    
13803
    node_data = self.rpc.call_node_info(node_list, [cfg.GetVGName()],
13804
                                        [hypervisor_name])
13805
    node_iinfo = \
13806
      self.rpc.call_all_instances_info(node_list,
13807
                                       cluster_info.enabled_hypervisors)
13808

    
13809
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
13810

    
13811
    config_ndata = self._ComputeBasicNodeData(ninfo)
13812
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
13813
                                                 i_list, config_ndata)
13814
    assert len(data["nodes"]) == len(ninfo), \
13815
        "Incomplete node data computed"
13816

    
13817
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
13818

    
13819
    self.in_data = data
13820

    
13821
  @staticmethod
13822
  def _ComputeNodeGroupData(cfg):
13823
    """Compute node groups data.
13824

13825
    """
13826
    ng = dict((guuid, {
13827
      "name": gdata.name,
13828
      "alloc_policy": gdata.alloc_policy,
13829
      })
13830
      for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
13831

    
13832
    return ng
13833

    
13834
  @staticmethod
13835
  def _ComputeBasicNodeData(node_cfg):
13836
    """Compute global node data.
13837

13838
    @rtype: dict
13839
    @returns: a dict of name: (node dict, node config)
13840

13841
    """
13842
    # fill in static (config-based) values
13843
    node_results = dict((ninfo.name, {
13844
      "tags": list(ninfo.GetTags()),
13845
      "primary_ip": ninfo.primary_ip,
13846
      "secondary_ip": ninfo.secondary_ip,
13847
      "offline": ninfo.offline,
13848
      "drained": ninfo.drained,
13849
      "master_candidate": ninfo.master_candidate,
13850
      "group": ninfo.group,
13851
      "master_capable": ninfo.master_capable,
13852
      "vm_capable": ninfo.vm_capable,
13853
      })
13854
      for ninfo in node_cfg.values())
13855

    
13856
    return node_results
13857

    
13858
  @staticmethod
13859
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
13860
                              node_results):
13861
    """Compute global node data.
13862

13863
    @param node_results: the basic node structures as filled from the config
13864

13865
    """
13866
    #TODO(dynmem): compute the right data on MAX and MIN memory
13867
    # make a copy of the current dict
13868
    node_results = dict(node_results)
13869
    for nname, nresult in node_data.items():
13870
      assert nname in node_results, "Missing basic data for node %s" % nname
13871
      ninfo = node_cfg[nname]
13872

    
13873
      if not (ninfo.offline or ninfo.drained):
13874
        nresult.Raise("Can't get data for node %s" % nname)
13875
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
13876
                                nname)
13877
        remote_info = _MakeLegacyNodeInfo(nresult.payload)
13878

    
13879
        for attr in ["memory_total", "memory_free", "memory_dom0",
13880
                     "vg_size", "vg_free", "cpu_total"]:
13881
          if attr not in remote_info:
13882
            raise errors.OpExecError("Node '%s' didn't return attribute"
13883
                                     " '%s'" % (nname, attr))
13884
          if not isinstance(remote_info[attr], int):
13885
            raise errors.OpExecError("Node '%s' returned invalid value"
13886
                                     " for '%s': %s" %
13887
                                     (nname, attr, remote_info[attr]))
13888
        # compute memory used by primary instances
13889
        i_p_mem = i_p_up_mem = 0
13890
        for iinfo, beinfo in i_list:
13891
          if iinfo.primary_node == nname:
13892
            i_p_mem += beinfo[constants.BE_MAXMEM]
13893
            if iinfo.name not in node_iinfo[nname].payload:
13894
              i_used_mem = 0
13895
            else:
13896
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
13897
            i_mem_diff = beinfo[constants.BE_MAXMEM] - i_used_mem
13898
            remote_info["memory_free"] -= max(0, i_mem_diff)
13899

    
13900
            if iinfo.admin_state == constants.ADMINST_UP:
13901
              i_p_up_mem += beinfo[constants.BE_MAXMEM]
13902

    
13903
        # compute memory used by instances
13904
        pnr_dyn = {
13905
          "total_memory": remote_info["memory_total"],
13906
          "reserved_memory": remote_info["memory_dom0"],
13907
          "free_memory": remote_info["memory_free"],
13908
          "total_disk": remote_info["vg_size"],
13909
          "free_disk": remote_info["vg_free"],
13910
          "total_cpus": remote_info["cpu_total"],
13911
          "i_pri_memory": i_p_mem,
13912
          "i_pri_up_memory": i_p_up_mem,
13913
          }
13914
        pnr_dyn.update(node_results[nname])
13915
        node_results[nname] = pnr_dyn
13916

    
13917
    return node_results
13918

    
13919
  @staticmethod
13920
  def _ComputeInstanceData(cluster_info, i_list):
13921
    """Compute global instance data.
13922

13923
    """
13924
    instance_data = {}
13925
    for iinfo, beinfo in i_list:
13926
      nic_data = []
13927
      for nic in iinfo.nics:
13928
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
13929
        nic_dict = {
13930
          "mac": nic.mac,
13931
          "ip": nic.ip,
13932
          "mode": filled_params[constants.NIC_MODE],
13933
          "link": filled_params[constants.NIC_LINK],
13934
          }
13935
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
13936
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
13937
        nic_data.append(nic_dict)
13938
      pir = {
13939
        "tags": list(iinfo.GetTags()),
13940
        "admin_state": iinfo.admin_state,
13941
        "vcpus": beinfo[constants.BE_VCPUS],
13942
        "memory": beinfo[constants.BE_MAXMEM],
13943
        "os": iinfo.os,
13944
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
13945
        "nics": nic_data,
13946
        "disks": [{constants.IDISK_SIZE: dsk.size,
13947
                   constants.IDISK_MODE: dsk.mode}
13948
                  for dsk in iinfo.disks],
13949
        "disk_template": iinfo.disk_template,
13950
        "hypervisor": iinfo.hypervisor,
13951
        }
13952
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
13953
                                                 pir["disks"])
13954
      instance_data[iinfo.name] = pir
13955

    
13956
    return instance_data
13957

    
13958
  def _AddNewInstance(self):
13959
    """Add new instance data to allocator structure.
13960

13961
    This in combination with _AllocatorGetClusterData will create the
13962
    correct structure needed as input for the allocator.
13963

13964
    The checks for the completeness of the opcode must have already been
13965
    done.
13966

13967
    """
13968
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
13969

    
13970
    if self.disk_template in constants.DTS_INT_MIRROR:
13971
      self.required_nodes = 2
13972
    else:
13973
      self.required_nodes = 1
13974

    
13975
    request = {
13976
      "name": self.name,
13977
      "disk_template": self.disk_template,
13978
      "tags": self.tags,
13979
      "os": self.os,
13980
      "vcpus": self.vcpus,
13981
      "memory": self.memory,
13982
      "disks": self.disks,
13983
      "disk_space_total": disk_space,
13984
      "nics": self.nics,
13985
      "required_nodes": self.required_nodes,
13986
      "hypervisor": self.hypervisor,
13987
      }
13988

    
13989
    return request
13990

    
13991
  def _AddRelocateInstance(self):
13992
    """Add relocate instance data to allocator structure.
13993

13994
    This in combination with _IAllocatorGetClusterData will create the
13995
    correct structure needed as input for the allocator.
13996

13997
    The checks for the completeness of the opcode must have already been
13998
    done.
13999

14000
    """
14001
    instance = self.cfg.GetInstanceInfo(self.name)
14002
    if instance is None:
14003
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
14004
                                   " IAllocator" % self.name)
14005

    
14006
    if instance.disk_template not in constants.DTS_MIRRORED:
14007
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
14008
                                 errors.ECODE_INVAL)
14009

    
14010
    if instance.disk_template in constants.DTS_INT_MIRROR and \
14011
        len(instance.secondary_nodes) != 1:
14012
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
14013
                                 errors.ECODE_STATE)
14014

    
14015
    self.required_nodes = 1
14016
    disk_sizes = [{constants.IDISK_SIZE: disk.size} for disk in instance.disks]
14017
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
14018

    
14019
    request = {
14020
      "name": self.name,
14021
      "disk_space_total": disk_space,
14022
      "required_nodes": self.required_nodes,
14023
      "relocate_from": self.relocate_from,
14024
      }
14025
    return request
14026

    
14027
  def _AddNodeEvacuate(self):
14028
    """Get data for node-evacuate requests.
14029

14030
    """
14031
    return {
14032
      "instances": self.instances,
14033
      "evac_mode": self.evac_mode,
14034
      }
14035

    
14036
  def _AddChangeGroup(self):
14037
    """Get data for node-evacuate requests.
14038

14039
    """
14040
    return {
14041
      "instances": self.instances,
14042
      "target_groups": self.target_groups,
14043
      }
14044

    
14045
  def _BuildInputData(self, fn, keydata):
14046
    """Build input data structures.
14047

14048
    """
14049
    self._ComputeClusterData()
14050

    
14051
    request = fn()
14052
    request["type"] = self.mode
14053
    for keyname, keytype in keydata:
14054
      if keyname not in request:
14055
        raise errors.ProgrammerError("Request parameter %s is missing" %
14056
                                     keyname)
14057
      val = request[keyname]
14058
      if not keytype(val):
14059
        raise errors.ProgrammerError("Request parameter %s doesn't pass"
14060
                                     " validation, value %s, expected"
14061
                                     " type %s" % (keyname, val, keytype))
14062
    self.in_data["request"] = request
14063

    
14064
    self.in_text = serializer.Dump(self.in_data)
14065

    
14066
  _STRING_LIST = ht.TListOf(ht.TString)
14067
  _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
14068
     # pylint: disable=E1101
14069
     # Class '...' has no 'OP_ID' member
14070
     "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
14071
                          opcodes.OpInstanceMigrate.OP_ID,
14072
                          opcodes.OpInstanceReplaceDisks.OP_ID])
14073
     })))
14074

    
14075
  _NEVAC_MOVED = \
14076
    ht.TListOf(ht.TAnd(ht.TIsLength(3),
14077
                       ht.TItems([ht.TNonEmptyString,
14078
                                  ht.TNonEmptyString,
14079
                                  ht.TListOf(ht.TNonEmptyString),
14080
                                 ])))
14081
  _NEVAC_FAILED = \
14082
    ht.TListOf(ht.TAnd(ht.TIsLength(2),
14083
                       ht.TItems([ht.TNonEmptyString,
14084
                                  ht.TMaybeString,
14085
                                 ])))
14086
  _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
14087
                          ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
14088

    
14089
  _MODE_DATA = {
14090
    constants.IALLOCATOR_MODE_ALLOC:
14091
      (_AddNewInstance,
14092
       [
14093
        ("name", ht.TString),
14094
        ("memory", ht.TInt),
14095
        ("disks", ht.TListOf(ht.TDict)),
14096
        ("disk_template", ht.TString),
14097
        ("os", ht.TString),
14098
        ("tags", _STRING_LIST),
14099
        ("nics", ht.TListOf(ht.TDict)),
14100
        ("vcpus", ht.TInt),
14101
        ("hypervisor", ht.TString),
14102
        ], ht.TList),
14103
    constants.IALLOCATOR_MODE_RELOC:
14104
      (_AddRelocateInstance,
14105
       [("name", ht.TString), ("relocate_from", _STRING_LIST)],
14106
       ht.TList),
14107
     constants.IALLOCATOR_MODE_NODE_EVAC:
14108
      (_AddNodeEvacuate, [
14109
        ("instances", _STRING_LIST),
14110
        ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
14111
        ], _NEVAC_RESULT),
14112
     constants.IALLOCATOR_MODE_CHG_GROUP:
14113
      (_AddChangeGroup, [
14114
        ("instances", _STRING_LIST),
14115
        ("target_groups", _STRING_LIST),
14116
        ], _NEVAC_RESULT),
14117
    }
14118

    
14119
  def Run(self, name, validate=True, call_fn=None):
14120
    """Run an instance allocator and return the results.
14121

14122
    """
14123
    if call_fn is None:
14124
      call_fn = self.rpc.call_iallocator_runner
14125

    
14126
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
14127
    result.Raise("Failure while running the iallocator script")
14128

    
14129
    self.out_text = result.payload
14130
    if validate:
14131
      self._ValidateResult()
14132

    
14133
  def _ValidateResult(self):
14134
    """Process the allocator results.
14135

14136
    This will process and if successful save the result in
14137
    self.out_data and the other parameters.
14138

14139
    """
14140
    try:
14141
      rdict = serializer.Load(self.out_text)
14142
    except Exception, err:
14143
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
14144

    
14145
    if not isinstance(rdict, dict):
14146
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
14147

    
14148
    # TODO: remove backwards compatiblity in later versions
14149
    if "nodes" in rdict and "result" not in rdict:
14150
      rdict["result"] = rdict["nodes"]
14151
      del rdict["nodes"]
14152

    
14153
    for key in "success", "info", "result":
14154
      if key not in rdict:
14155
        raise errors.OpExecError("Can't parse iallocator results:"
14156
                                 " missing key '%s'" % key)
14157
      setattr(self, key, rdict[key])
14158

    
14159
    if not self._result_check(self.result):
14160
      raise errors.OpExecError("Iallocator returned invalid result,"
14161
                               " expected %s, got %s" %
14162
                               (self._result_check, self.result),
14163
                               errors.ECODE_INVAL)
14164

    
14165
    if self.mode == constants.IALLOCATOR_MODE_RELOC:
14166
      assert self.relocate_from is not None
14167
      assert self.required_nodes == 1
14168

    
14169
      node2group = dict((name, ndata["group"])
14170
                        for (name, ndata) in self.in_data["nodes"].items())
14171

    
14172
      fn = compat.partial(self._NodesToGroups, node2group,
14173
                          self.in_data["nodegroups"])
14174

    
14175
      instance = self.cfg.GetInstanceInfo(self.name)
14176
      request_groups = fn(self.relocate_from + [instance.primary_node])
14177
      result_groups = fn(rdict["result"] + [instance.primary_node])
14178

    
14179
      if self.success and not set(result_groups).issubset(request_groups):
14180
        raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
14181
                                 " differ from original groups (%s)" %
14182
                                 (utils.CommaJoin(result_groups),
14183
                                  utils.CommaJoin(request_groups)))
14184

    
14185
    elif self.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
14186
      assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
14187

    
14188
    self.out_data = rdict
14189

    
14190
  @staticmethod
14191
  def _NodesToGroups(node2group, groups, nodes):
14192
    """Returns a list of unique group names for a list of nodes.
14193

14194
    @type node2group: dict
14195
    @param node2group: Map from node name to group UUID
14196
    @type groups: dict
14197
    @param groups: Group information
14198
    @type nodes: list
14199
    @param nodes: Node names
14200

14201
    """
14202
    result = set()
14203

    
14204
    for node in nodes:
14205
      try:
14206
        group_uuid = node2group[node]
14207
      except KeyError:
14208
        # Ignore unknown node
14209
        pass
14210
      else:
14211
        try:
14212
          group = groups[group_uuid]
14213
        except KeyError:
14214
          # Can't find group, let's use UUID
14215
          group_name = group_uuid
14216
        else:
14217
          group_name = group["name"]
14218

    
14219
        result.add(group_name)
14220

    
14221
    return sorted(result)
14222

    
14223

    
14224
class LUTestAllocator(NoHooksLU):
14225
  """Run allocator tests.
14226

14227
  This LU runs the allocator tests
14228

14229
  """
14230
  def CheckPrereq(self):
14231
    """Check prerequisites.
14232

14233
    This checks the opcode parameters depending on the director and mode test.
14234

14235
    """
14236
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
14237
      for attr in ["memory", "disks", "disk_template",
14238
                   "os", "tags", "nics", "vcpus"]:
14239
        if not hasattr(self.op, attr):
14240
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
14241
                                     attr, errors.ECODE_INVAL)
14242
      iname = self.cfg.ExpandInstanceName(self.op.name)
14243
      if iname is not None:
14244
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
14245
                                   iname, errors.ECODE_EXISTS)
14246
      if not isinstance(self.op.nics, list):
14247
        raise errors.OpPrereqError("Invalid parameter 'nics'",
14248
                                   errors.ECODE_INVAL)
14249
      if not isinstance(self.op.disks, list):
14250
        raise errors.OpPrereqError("Invalid parameter 'disks'",
14251
                                   errors.ECODE_INVAL)
14252
      for row in self.op.disks:
14253
        if (not isinstance(row, dict) or
14254
            constants.IDISK_SIZE not in row or
14255
            not isinstance(row[constants.IDISK_SIZE], int) or
14256
            constants.IDISK_MODE not in row or
14257
            row[constants.IDISK_MODE] not in constants.DISK_ACCESS_SET):
14258
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
14259
                                     " parameter", errors.ECODE_INVAL)
14260
      if self.op.hypervisor is None:
14261
        self.op.hypervisor = self.cfg.GetHypervisorType()
14262
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
14263
      fname = _ExpandInstanceName(self.cfg, self.op.name)
14264
      self.op.name = fname
14265
      self.relocate_from = \
14266
          list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
14267
    elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
14268
                          constants.IALLOCATOR_MODE_NODE_EVAC):
14269
      if not self.op.instances:
14270
        raise errors.OpPrereqError("Missing instances", errors.ECODE_INVAL)
14271
      self.op.instances = _GetWantedInstances(self, self.op.instances)
14272
    else:
14273
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
14274
                                 self.op.mode, errors.ECODE_INVAL)
14275

    
14276
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
14277
      if self.op.allocator is None:
14278
        raise errors.OpPrereqError("Missing allocator name",
14279
                                   errors.ECODE_INVAL)
14280
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
14281
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
14282
                                 self.op.direction, errors.ECODE_INVAL)
14283

    
14284
  def Exec(self, feedback_fn):
14285
    """Run the allocator test.
14286

14287
    """
14288
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
14289
      ial = IAllocator(self.cfg, self.rpc,
14290
                       mode=self.op.mode,
14291
                       name=self.op.name,
14292
                       memory=self.op.memory,
14293
                       disks=self.op.disks,
14294
                       disk_template=self.op.disk_template,
14295
                       os=self.op.os,
14296
                       tags=self.op.tags,
14297
                       nics=self.op.nics,
14298
                       vcpus=self.op.vcpus,
14299
                       hypervisor=self.op.hypervisor,
14300
                       )
14301
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
14302
      ial = IAllocator(self.cfg, self.rpc,
14303
                       mode=self.op.mode,
14304
                       name=self.op.name,
14305
                       relocate_from=list(self.relocate_from),
14306
                       )
14307
    elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
14308
      ial = IAllocator(self.cfg, self.rpc,
14309
                       mode=self.op.mode,
14310
                       instances=self.op.instances,
14311
                       target_groups=self.op.target_groups)
14312
    elif self.op.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
14313
      ial = IAllocator(self.cfg, self.rpc,
14314
                       mode=self.op.mode,
14315
                       instances=self.op.instances,
14316
                       evac_mode=self.op.evac_mode)
14317
    else:
14318
      raise errors.ProgrammerError("Uncatched mode %s in"
14319
                                   " LUTestAllocator.Exec", self.op.mode)
14320

    
14321
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
14322
      result = ial.in_text
14323
    else:
14324
      ial.Run(self.op.allocator, validate=False)
14325
      result = ial.out_text
14326
    return result
14327

    
14328

    
14329
#: Query type implementations
14330
_QUERY_IMPL = {
14331
  constants.QR_INSTANCE: _InstanceQuery,
14332
  constants.QR_NODE: _NodeQuery,
14333
  constants.QR_GROUP: _GroupQuery,
14334
  constants.QR_OS: _OsQuery,
14335
  }
14336

    
14337
assert set(_QUERY_IMPL.keys()) == constants.QR_VIA_OP
14338

    
14339

    
14340
def _GetQueryImplementation(name):
14341
  """Returns the implemtnation for a query type.
14342

14343
  @param name: Query type, must be one of L{constants.QR_VIA_OP}
14344

14345
  """
14346
  try:
14347
    return _QUERY_IMPL[name]
14348
  except KeyError:
14349
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
14350
                               errors.ECODE_INVAL)