Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 6a1434d7

History | View | Annotate | Download (476.6 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

    
63
import ganeti.masterd.instance # pylint: disable=W0611
64

    
65

    
66
class ResultWithJobs:
67
  """Data container for LU results with jobs.
68

69
  Instances of this class returned from L{LogicalUnit.Exec} will be recognized
70
  by L{mcpu.Processor._ProcessResult}. The latter will then submit the jobs
71
  contained in the C{jobs} attribute and include the job IDs in the opcode
72
  result.
73

74
  """
75
  def __init__(self, jobs, **kwargs):
76
    """Initializes this class.
77

78
    Additional return values can be specified as keyword arguments.
79

80
    @type jobs: list of lists of L{opcode.OpCode}
81
    @param jobs: A list of lists of opcode objects
82

83
    """
84
    self.jobs = jobs
85
    self.other = kwargs
86

    
87

    
88
class LogicalUnit(object):
89
  """Logical Unit base class.
90

91
  Subclasses must follow these rules:
92
    - implement ExpandNames
93
    - implement CheckPrereq (except when tasklets are used)
94
    - implement Exec (except when tasklets are used)
95
    - implement BuildHooksEnv
96
    - implement BuildHooksNodes
97
    - redefine HPATH and HTYPE
98
    - optionally redefine their run requirements:
99
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
100

101
  Note that all commands require root permissions.
102

103
  @ivar dry_run_result: the value (if any) that will be returned to the caller
104
      in dry-run mode (signalled by opcode dry_run parameter)
105

106
  """
107
  HPATH = None
108
  HTYPE = None
109
  REQ_BGL = True
110

    
111
  def __init__(self, processor, op, context, rpc):
112
    """Constructor for LogicalUnit.
113

114
    This needs to be overridden in derived classes in order to check op
115
    validity.
116

117
    """
118
    self.proc = processor
119
    self.op = op
120
    self.cfg = context.cfg
121
    self.glm = context.glm
122
    # readability alias
123
    self.owned_locks = context.glm.list_owned
124
    self.context = context
125
    self.rpc = rpc
126
    # Dicts used to declare locking needs to mcpu
127
    self.needed_locks = None
128
    self.share_locks = dict.fromkeys(locking.LEVELS, 0)
129
    self.add_locks = {}
130
    self.remove_locks = {}
131
    # Used to force good behavior when calling helper functions
132
    self.recalculate_locks = {}
133
    # logging
134
    self.Log = processor.Log # pylint: disable=C0103
135
    self.LogWarning = processor.LogWarning # pylint: disable=C0103
136
    self.LogInfo = processor.LogInfo # pylint: disable=C0103
137
    self.LogStep = processor.LogStep # pylint: disable=C0103
138
    # support for dry-run
139
    self.dry_run_result = None
140
    # support for generic debug attribute
141
    if (not hasattr(self.op, "debug_level") or
142
        not isinstance(self.op.debug_level, int)):
143
      self.op.debug_level = 0
144

    
145
    # Tasklets
146
    self.tasklets = None
147

    
148
    # Validate opcode parameters and set defaults
149
    self.op.Validate(True)
150

    
151
    self.CheckArguments()
152

    
153
  def CheckArguments(self):
154
    """Check syntactic validity for the opcode arguments.
155

156
    This method is for doing a simple syntactic check and ensure
157
    validity of opcode parameters, without any cluster-related
158
    checks. While the same can be accomplished in ExpandNames and/or
159
    CheckPrereq, doing these separate is better because:
160

161
      - ExpandNames is left as as purely a lock-related function
162
      - CheckPrereq is run after we have acquired locks (and possible
163
        waited for them)
164

165
    The function is allowed to change the self.op attribute so that
166
    later methods can no longer worry about missing parameters.
167

168
    """
169
    pass
170

    
171
  def ExpandNames(self):
172
    """Expand names for this LU.
173

174
    This method is called before starting to execute the opcode, and it should
175
    update all the parameters of the opcode to their canonical form (e.g. a
176
    short node name must be fully expanded after this method has successfully
177
    completed). This way locking, hooks, logging, etc. can work correctly.
178

179
    LUs which implement this method must also populate the self.needed_locks
180
    member, as a dict with lock levels as keys, and a list of needed lock names
181
    as values. Rules:
182

183
      - use an empty dict if you don't need any lock
184
      - if you don't need any lock at a particular level omit that level
185
      - don't put anything for the BGL level
186
      - if you want all locks at a level use locking.ALL_SET as a value
187

188
    If you need to share locks (rather than acquire them exclusively) at one
189
    level you can modify self.share_locks, setting a true value (usually 1) for
190
    that level. By default locks are not shared.
191

192
    This function can also define a list of tasklets, which then will be
193
    executed in order instead of the usual LU-level CheckPrereq and Exec
194
    functions, if those are not defined by the LU.
195

196
    Examples::
197

198
      # Acquire all nodes and one instance
199
      self.needed_locks = {
200
        locking.LEVEL_NODE: locking.ALL_SET,
201
        locking.LEVEL_INSTANCE: ['instance1.example.com'],
202
      }
203
      # Acquire just two nodes
204
      self.needed_locks = {
205
        locking.LEVEL_NODE: ['node1.example.com', 'node2.example.com'],
206
      }
207
      # Acquire no locks
208
      self.needed_locks = {} # No, you can't leave it to the default value None
209

210
    """
211
    # The implementation of this method is mandatory only if the new LU is
212
    # concurrent, so that old LUs don't need to be changed all at the same
213
    # time.
214
    if self.REQ_BGL:
215
      self.needed_locks = {} # Exclusive LUs don't need locks.
216
    else:
217
      raise NotImplementedError
218

    
219
  def DeclareLocks(self, level):
220
    """Declare LU locking needs for a level
221

222
    While most LUs can just declare their locking needs at ExpandNames time,
223
    sometimes there's the need to calculate some locks after having acquired
224
    the ones before. This function is called just before acquiring locks at a
225
    particular level, but after acquiring the ones at lower levels, and permits
226
    such calculations. It can be used to modify self.needed_locks, and by
227
    default it does nothing.
228

229
    This function is only called if you have something already set in
230
    self.needed_locks for the level.
231

232
    @param level: Locking level which is going to be locked
233
    @type level: member of ganeti.locking.LEVELS
234

235
    """
236

    
237
  def CheckPrereq(self):
238
    """Check prerequisites for this LU.
239

240
    This method should check that the prerequisites for the execution
241
    of this LU are fulfilled. It can do internode communication, but
242
    it should be idempotent - no cluster or system changes are
243
    allowed.
244

245
    The method should raise errors.OpPrereqError in case something is
246
    not fulfilled. Its return value is ignored.
247

248
    This method should also update all the parameters of the opcode to
249
    their canonical form if it hasn't been done by ExpandNames before.
250

251
    """
252
    if self.tasklets is not None:
253
      for (idx, tl) in enumerate(self.tasklets):
254
        logging.debug("Checking prerequisites for tasklet %s/%s",
255
                      idx + 1, len(self.tasklets))
256
        tl.CheckPrereq()
257
    else:
258
      pass
259

    
260
  def Exec(self, feedback_fn):
261
    """Execute the LU.
262

263
    This method should implement the actual work. It should raise
264
    errors.OpExecError for failures that are somewhat dealt with in
265
    code, or expected.
266

267
    """
268
    if self.tasklets is not None:
269
      for (idx, tl) in enumerate(self.tasklets):
270
        logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
271
        tl.Exec(feedback_fn)
272
    else:
273
      raise NotImplementedError
274

    
275
  def BuildHooksEnv(self):
276
    """Build hooks environment for this LU.
277

278
    @rtype: dict
279
    @return: Dictionary containing the environment that will be used for
280
      running the hooks for this LU. The keys of the dict must not be prefixed
281
      with "GANETI_"--that'll be added by the hooks runner. The hooks runner
282
      will extend the environment with additional variables. If no environment
283
      should be defined, an empty dictionary should be returned (not C{None}).
284
    @note: If the C{HPATH} attribute of the LU class is C{None}, this function
285
      will not be called.
286

287
    """
288
    raise NotImplementedError
289

    
290
  def BuildHooksNodes(self):
291
    """Build list of nodes to run LU's hooks.
292

293
    @rtype: tuple; (list, list)
294
    @return: Tuple containing a list of node names on which the hook
295
      should run before the execution and a list of node names on which the
296
      hook should run after the execution. No nodes should be returned as an
297
      empty list (and not None).
298
    @note: If the C{HPATH} attribute of the LU class is C{None}, this function
299
      will not be called.
300

301
    """
302
    raise NotImplementedError
303

    
304
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
305
    """Notify the LU about the results of its hooks.
306

307
    This method is called every time a hooks phase is executed, and notifies
308
    the Logical Unit about the hooks' result. The LU can then use it to alter
309
    its result based on the hooks.  By default the method does nothing and the
310
    previous result is passed back unchanged but any LU can define it if it
311
    wants to use the local cluster hook-scripts somehow.
312

313
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
314
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
315
    @param hook_results: the results of the multi-node hooks rpc call
316
    @param feedback_fn: function used send feedback back to the caller
317
    @param lu_result: the previous Exec result this LU had, or None
318
        in the PRE phase
319
    @return: the new Exec result, based on the previous result
320
        and hook results
321

322
    """
323
    # API must be kept, thus we ignore the unused argument and could
324
    # be a function warnings
325
    # pylint: disable=W0613,R0201
326
    return lu_result
327

    
328
  def _ExpandAndLockInstance(self):
329
    """Helper function to expand and lock an instance.
330

331
    Many LUs that work on an instance take its name in self.op.instance_name
332
    and need to expand it and then declare the expanded name for locking. This
333
    function does it, and then updates self.op.instance_name to the expanded
334
    name. It also initializes needed_locks as a dict, if this hasn't been done
335
    before.
336

337
    """
338
    if self.needed_locks is None:
339
      self.needed_locks = {}
340
    else:
341
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
342
        "_ExpandAndLockInstance called with instance-level locks set"
343
    self.op.instance_name = _ExpandInstanceName(self.cfg,
344
                                                self.op.instance_name)
345
    self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
346

    
347
  def _LockInstancesNodes(self, primary_only=False):
348
    """Helper function to declare instances' nodes for locking.
349

350
    This function should be called after locking one or more instances to lock
351
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
352
    with all primary or secondary nodes for instances already locked and
353
    present in self.needed_locks[locking.LEVEL_INSTANCE].
354

355
    It should be called from DeclareLocks, and for safety only works if
356
    self.recalculate_locks[locking.LEVEL_NODE] is set.
357

358
    In the future it may grow parameters to just lock some instance's nodes, or
359
    to just lock primaries or secondary nodes, if needed.
360

361
    If should be called in DeclareLocks in a way similar to::
362

363
      if level == locking.LEVEL_NODE:
364
        self._LockInstancesNodes()
365

366
    @type primary_only: boolean
367
    @param primary_only: only lock primary nodes of locked instances
368

369
    """
370
    assert locking.LEVEL_NODE in self.recalculate_locks, \
371
      "_LockInstancesNodes helper function called with no nodes to recalculate"
372

    
373
    # TODO: check if we're really been called with the instance locks held
374

    
375
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
376
    # future we might want to have different behaviors depending on the value
377
    # of self.recalculate_locks[locking.LEVEL_NODE]
378
    wanted_nodes = []
379
    locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
380
    for _, instance in self.cfg.GetMultiInstanceInfo(locked_i):
381
      wanted_nodes.append(instance.primary_node)
382
      if not primary_only:
383
        wanted_nodes.extend(instance.secondary_nodes)
384

    
385
    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
386
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
387
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
388
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
389

    
390
    del self.recalculate_locks[locking.LEVEL_NODE]
391

    
392

    
393
class NoHooksLU(LogicalUnit): # pylint: disable=W0223
394
  """Simple LU which runs no hooks.
395

396
  This LU is intended as a parent for other LogicalUnits which will
397
  run no hooks, in order to reduce duplicate code.
398

399
  """
400
  HPATH = None
401
  HTYPE = None
402

    
403
  def BuildHooksEnv(self):
404
    """Empty BuildHooksEnv for NoHooksLu.
405

406
    This just raises an error.
407

408
    """
409
    raise AssertionError("BuildHooksEnv called for NoHooksLUs")
410

    
411
  def BuildHooksNodes(self):
412
    """Empty BuildHooksNodes for NoHooksLU.
413

414
    """
415
    raise AssertionError("BuildHooksNodes called for NoHooksLU")
416

    
417

    
418
class Tasklet:
419
  """Tasklet base class.
420

421
  Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
422
  they can mix legacy code with tasklets. Locking needs to be done in the LU,
423
  tasklets know nothing about locks.
424

425
  Subclasses must follow these rules:
426
    - Implement CheckPrereq
427
    - Implement Exec
428

429
  """
430
  def __init__(self, lu):
431
    self.lu = lu
432

    
433
    # Shortcuts
434
    self.cfg = lu.cfg
435
    self.rpc = lu.rpc
436

    
437
  def CheckPrereq(self):
438
    """Check prerequisites for this tasklets.
439

440
    This method should check whether the prerequisites for the execution of
441
    this tasklet are fulfilled. It can do internode communication, but it
442
    should be idempotent - no cluster or system changes are allowed.
443

444
    The method should raise errors.OpPrereqError in case something is not
445
    fulfilled. Its return value is ignored.
446

447
    This method should also update all parameters to their canonical form if it
448
    hasn't been done before.
449

450
    """
451
    pass
452

    
453
  def Exec(self, feedback_fn):
454
    """Execute the tasklet.
455

456
    This method should implement the actual work. It should raise
457
    errors.OpExecError for failures that are somewhat dealt with in code, or
458
    expected.
459

460
    """
461
    raise NotImplementedError
462

    
463

    
464
class _QueryBase:
465
  """Base for query utility classes.
466

467
  """
468
  #: Attribute holding field definitions
469
  FIELDS = None
470

    
471
  def __init__(self, filter_, fields, use_locking):
472
    """Initializes this class.
473

474
    """
475
    self.use_locking = use_locking
476

    
477
    self.query = query.Query(self.FIELDS, fields, filter_=filter_,
478
                             namefield="name")
479
    self.requested_data = self.query.RequestedData()
480
    self.names = self.query.RequestedNames()
481

    
482
    # Sort only if no names were requested
483
    self.sort_by_name = not self.names
484

    
485
    self.do_locking = None
486
    self.wanted = None
487

    
488
  def _GetNames(self, lu, all_names, lock_level):
489
    """Helper function to determine names asked for in the query.
490

491
    """
492
    if self.do_locking:
493
      names = lu.owned_locks(lock_level)
494
    else:
495
      names = all_names
496

    
497
    if self.wanted == locking.ALL_SET:
498
      assert not self.names
499
      # caller didn't specify names, so ordering is not important
500
      return utils.NiceSort(names)
501

    
502
    # caller specified names and we must keep the same order
503
    assert self.names
504
    assert not self.do_locking or lu.glm.is_owned(lock_level)
505

    
506
    missing = set(self.wanted).difference(names)
507
    if missing:
508
      raise errors.OpExecError("Some items were removed before retrieving"
509
                               " their data: %s" % missing)
510

    
511
    # Return expanded names
512
    return self.wanted
513

    
514
  def ExpandNames(self, lu):
515
    """Expand names for this query.
516

517
    See L{LogicalUnit.ExpandNames}.
518

519
    """
520
    raise NotImplementedError()
521

    
522
  def DeclareLocks(self, lu, level):
523
    """Declare locks for this query.
524

525
    See L{LogicalUnit.DeclareLocks}.
526

527
    """
528
    raise NotImplementedError()
529

    
530
  def _GetQueryData(self, lu):
531
    """Collects all data for this query.
532

533
    @return: Query data object
534

535
    """
536
    raise NotImplementedError()
537

    
538
  def NewStyleQuery(self, lu):
539
    """Collect data and execute query.
540

541
    """
542
    return query.GetQueryResponse(self.query, self._GetQueryData(lu),
543
                                  sort_by_name=self.sort_by_name)
544

    
545
  def OldStyleQuery(self, lu):
546
    """Collect data and execute query.
547

548
    """
549
    return self.query.OldStyleQuery(self._GetQueryData(lu),
550
                                    sort_by_name=self.sort_by_name)
551

    
552

    
553
def _ShareAll():
554
  """Returns a dict declaring all lock levels shared.
555

556
  """
557
  return dict.fromkeys(locking.LEVELS, 1)
558

    
559

    
560
def _CheckInstanceNodeGroups(cfg, instance_name, owned_groups):
561
  """Checks if the owned node groups are still correct for an instance.
562

563
  @type cfg: L{config.ConfigWriter}
564
  @param cfg: The cluster configuration
565
  @type instance_name: string
566
  @param instance_name: Instance name
567
  @type owned_groups: set or frozenset
568
  @param owned_groups: List of currently owned node groups
569

570
  """
571
  inst_groups = cfg.GetInstanceNodeGroups(instance_name)
572

    
573
  if not owned_groups.issuperset(inst_groups):
574
    raise errors.OpPrereqError("Instance %s's node groups changed since"
575
                               " locks were acquired, current groups are"
576
                               " are '%s', owning groups '%s'; retry the"
577
                               " operation" %
578
                               (instance_name,
579
                                utils.CommaJoin(inst_groups),
580
                                utils.CommaJoin(owned_groups)),
581
                               errors.ECODE_STATE)
582

    
583
  return inst_groups
584

    
585

    
586
def _CheckNodeGroupInstances(cfg, group_uuid, owned_instances):
587
  """Checks if the instances in a node group are still correct.
588

589
  @type cfg: L{config.ConfigWriter}
590
  @param cfg: The cluster configuration
591
  @type group_uuid: string
592
  @param group_uuid: Node group UUID
593
  @type owned_instances: set or frozenset
594
  @param owned_instances: List of currently owned instances
595

596
  """
597
  wanted_instances = cfg.GetNodeGroupInstances(group_uuid)
598
  if owned_instances != wanted_instances:
599
    raise errors.OpPrereqError("Instances in node group '%s' changed since"
600
                               " locks were acquired, wanted '%s', have '%s';"
601
                               " retry the operation" %
602
                               (group_uuid,
603
                                utils.CommaJoin(wanted_instances),
604
                                utils.CommaJoin(owned_instances)),
605
                               errors.ECODE_STATE)
606

    
607
  return wanted_instances
608

    
609

    
610
def _SupportsOob(cfg, node):
611
  """Tells if node supports OOB.
612

613
  @type cfg: L{config.ConfigWriter}
614
  @param cfg: The cluster configuration
615
  @type node: L{objects.Node}
616
  @param node: The node
617
  @return: The OOB script if supported or an empty string otherwise
618

619
  """
620
  return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
621

    
622

    
623
def _GetWantedNodes(lu, nodes):
624
  """Returns list of checked and expanded node names.
625

626
  @type lu: L{LogicalUnit}
627
  @param lu: the logical unit on whose behalf we execute
628
  @type nodes: list
629
  @param nodes: list of node names or None for all nodes
630
  @rtype: list
631
  @return: the list of nodes, sorted
632
  @raise errors.ProgrammerError: if the nodes parameter is wrong type
633

634
  """
635
  if nodes:
636
    return [_ExpandNodeName(lu.cfg, name) for name in nodes]
637

    
638
  return utils.NiceSort(lu.cfg.GetNodeList())
639

    
640

    
641
def _GetWantedInstances(lu, instances):
642
  """Returns list of checked and expanded instance names.
643

644
  @type lu: L{LogicalUnit}
645
  @param lu: the logical unit on whose behalf we execute
646
  @type instances: list
647
  @param instances: list of instance names or None for all instances
648
  @rtype: list
649
  @return: the list of instances, sorted
650
  @raise errors.OpPrereqError: if the instances parameter is wrong type
651
  @raise errors.OpPrereqError: if any of the passed instances is not found
652

653
  """
654
  if instances:
655
    wanted = [_ExpandInstanceName(lu.cfg, name) for name in instances]
656
  else:
657
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
658
  return wanted
659

    
660

    
661
def _GetUpdatedParams(old_params, update_dict,
662
                      use_default=True, use_none=False):
663
  """Return the new version of a parameter dictionary.
664

665
  @type old_params: dict
666
  @param old_params: old parameters
667
  @type update_dict: dict
668
  @param update_dict: dict containing new parameter values, or
669
      constants.VALUE_DEFAULT to reset the parameter to its default
670
      value
671
  @param use_default: boolean
672
  @type use_default: whether to recognise L{constants.VALUE_DEFAULT}
673
      values as 'to be deleted' values
674
  @param use_none: boolean
675
  @type use_none: whether to recognise C{None} values as 'to be
676
      deleted' values
677
  @rtype: dict
678
  @return: the new parameter dictionary
679

680
  """
681
  params_copy = copy.deepcopy(old_params)
682
  for key, val in update_dict.iteritems():
683
    if ((use_default and val == constants.VALUE_DEFAULT) or
684
        (use_none and val is None)):
685
      try:
686
        del params_copy[key]
687
      except KeyError:
688
        pass
689
    else:
690
      params_copy[key] = val
691
  return params_copy
692

    
693

    
694
def _ReleaseLocks(lu, level, names=None, keep=None):
695
  """Releases locks owned by an LU.
696

697
  @type lu: L{LogicalUnit}
698
  @param level: Lock level
699
  @type names: list or None
700
  @param names: Names of locks to release
701
  @type keep: list or None
702
  @param keep: Names of locks to retain
703

704
  """
705
  assert not (keep is not None and names is not None), \
706
         "Only one of the 'names' and the 'keep' parameters can be given"
707

    
708
  if names is not None:
709
    should_release = names.__contains__
710
  elif keep:
711
    should_release = lambda name: name not in keep
712
  else:
713
    should_release = None
714

    
715
  if should_release:
716
    retain = []
717
    release = []
718

    
719
    # Determine which locks to release
720
    for name in lu.owned_locks(level):
721
      if should_release(name):
722
        release.append(name)
723
      else:
724
        retain.append(name)
725

    
726
    assert len(lu.owned_locks(level)) == (len(retain) + len(release))
727

    
728
    # Release just some locks
729
    lu.glm.release(level, names=release)
730

    
731
    assert frozenset(lu.owned_locks(level)) == frozenset(retain)
732
  else:
733
    # Release everything
734
    lu.glm.release(level)
735

    
736
    assert not lu.glm.is_owned(level), "No locks should be owned"
737

    
738

    
739
def _MapInstanceDisksToNodes(instances):
740
  """Creates a map from (node, volume) to instance name.
741

742
  @type instances: list of L{objects.Instance}
743
  @rtype: dict; tuple of (node name, volume name) as key, instance name as value
744

745
  """
746
  return dict(((node, vol), inst.name)
747
              for inst in instances
748
              for (node, vols) in inst.MapLVsByNode().items()
749
              for vol in vols)
750

    
751

    
752
def _RunPostHook(lu, node_name):
753
  """Runs the post-hook for an opcode on a single node.
754

755
  """
756
  hm = lu.proc.hmclass(lu.rpc.call_hooks_runner, lu)
757
  try:
758
    hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
759
  except:
760
    # pylint: disable=W0702
761
    lu.LogWarning("Errors occurred running hooks on %s" % node_name)
762

    
763

    
764
def _CheckOutputFields(static, dynamic, selected):
765
  """Checks whether all selected fields are valid.
766

767
  @type static: L{utils.FieldSet}
768
  @param static: static fields set
769
  @type dynamic: L{utils.FieldSet}
770
  @param dynamic: dynamic fields set
771

772
  """
773
  f = utils.FieldSet()
774
  f.Extend(static)
775
  f.Extend(dynamic)
776

    
777
  delta = f.NonMatching(selected)
778
  if delta:
779
    raise errors.OpPrereqError("Unknown output fields selected: %s"
780
                               % ",".join(delta), errors.ECODE_INVAL)
781

    
782

    
783
def _CheckGlobalHvParams(params):
784
  """Validates that given hypervisor params are not global ones.
785

786
  This will ensure that instances don't get customised versions of
787
  global params.
788

789
  """
790
  used_globals = constants.HVC_GLOBALS.intersection(params)
791
  if used_globals:
792
    msg = ("The following hypervisor parameters are global and cannot"
793
           " be customized at instance level, please modify them at"
794
           " cluster level: %s" % utils.CommaJoin(used_globals))
795
    raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
796

    
797

    
798
def _CheckNodeOnline(lu, node, msg=None):
799
  """Ensure that a given node is online.
800

801
  @param lu: the LU on behalf of which we make the check
802
  @param node: the node to check
803
  @param msg: if passed, should be a message to replace the default one
804
  @raise errors.OpPrereqError: if the node is offline
805

806
  """
807
  if msg is None:
808
    msg = "Can't use offline node"
809
  if lu.cfg.GetNodeInfo(node).offline:
810
    raise errors.OpPrereqError("%s: %s" % (msg, node), errors.ECODE_STATE)
811

    
812

    
813
def _CheckNodeNotDrained(lu, node):
814
  """Ensure that a given node is not drained.
815

816
  @param lu: the LU on behalf of which we make the check
817
  @param node: the node to check
818
  @raise errors.OpPrereqError: if the node is drained
819

820
  """
821
  if lu.cfg.GetNodeInfo(node).drained:
822
    raise errors.OpPrereqError("Can't use drained node %s" % node,
823
                               errors.ECODE_STATE)
824

    
825

    
826
def _CheckNodeVmCapable(lu, node):
827
  """Ensure that a given node is vm capable.
828

829
  @param lu: the LU on behalf of which we make the check
830
  @param node: the node to check
831
  @raise errors.OpPrereqError: if the node is not vm capable
832

833
  """
834
  if not lu.cfg.GetNodeInfo(node).vm_capable:
835
    raise errors.OpPrereqError("Can't use non-vm_capable node %s" % node,
836
                               errors.ECODE_STATE)
837

    
838

    
839
def _CheckNodeHasOS(lu, node, os_name, force_variant):
840
  """Ensure that a node supports a given OS.
841

842
  @param lu: the LU on behalf of which we make the check
843
  @param node: the node to check
844
  @param os_name: the OS to query about
845
  @param force_variant: whether to ignore variant errors
846
  @raise errors.OpPrereqError: if the node is not supporting the OS
847

848
  """
849
  result = lu.rpc.call_os_get(node, os_name)
850
  result.Raise("OS '%s' not in supported OS list for node %s" %
851
               (os_name, node),
852
               prereq=True, ecode=errors.ECODE_INVAL)
853
  if not force_variant:
854
    _CheckOSVariant(result.payload, os_name)
855

    
856

    
857
def _CheckNodeHasSecondaryIP(lu, node, secondary_ip, prereq):
858
  """Ensure that a node has the given secondary ip.
859

860
  @type lu: L{LogicalUnit}
861
  @param lu: the LU on behalf of which we make the check
862
  @type node: string
863
  @param node: the node to check
864
  @type secondary_ip: string
865
  @param secondary_ip: the ip to check
866
  @type prereq: boolean
867
  @param prereq: whether to throw a prerequisite or an execute error
868
  @raise errors.OpPrereqError: if the node doesn't have the ip, and prereq=True
869
  @raise errors.OpExecError: if the node doesn't have the ip, and prereq=False
870

871
  """
872
  result = lu.rpc.call_node_has_ip_address(node, secondary_ip)
873
  result.Raise("Failure checking secondary ip on node %s" % node,
874
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
875
  if not result.payload:
876
    msg = ("Node claims it doesn't have the secondary ip you gave (%s),"
877
           " please fix and re-run this command" % secondary_ip)
878
    if prereq:
879
      raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
880
    else:
881
      raise errors.OpExecError(msg)
882

    
883

    
884
def _GetClusterDomainSecret():
885
  """Reads the cluster domain secret.
886

887
  """
888
  return utils.ReadOneLineFile(constants.CLUSTER_DOMAIN_SECRET_FILE,
889
                               strict=True)
890

    
891

    
892
def _CheckInstanceDown(lu, instance, reason):
893
  """Ensure that an instance is not running."""
894
  if instance.admin_up:
895
    raise errors.OpPrereqError("Instance %s is marked to be up, %s" %
896
                               (instance.name, reason), errors.ECODE_STATE)
897

    
898
  pnode = instance.primary_node
899
  ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
900
  ins_l.Raise("Can't contact node %s for instance information" % pnode,
901
              prereq=True, ecode=errors.ECODE_ENVIRON)
902

    
903
  if instance.name in ins_l.payload:
904
    raise errors.OpPrereqError("Instance %s is running, %s" %
905
                               (instance.name, reason), errors.ECODE_STATE)
906

    
907

    
908
def _ExpandItemName(fn, name, kind):
909
  """Expand an item name.
910

911
  @param fn: the function to use for expansion
912
  @param name: requested item name
913
  @param kind: text description ('Node' or 'Instance')
914
  @return: the resolved (full) name
915
  @raise errors.OpPrereqError: if the item is not found
916

917
  """
918
  full_name = fn(name)
919
  if full_name is None:
920
    raise errors.OpPrereqError("%s '%s' not known" % (kind, name),
921
                               errors.ECODE_NOENT)
922
  return full_name
923

    
924

    
925
def _ExpandNodeName(cfg, name):
926
  """Wrapper over L{_ExpandItemName} for nodes."""
927
  return _ExpandItemName(cfg.ExpandNodeName, name, "Node")
928

    
929

    
930
def _ExpandInstanceName(cfg, name):
931
  """Wrapper over L{_ExpandItemName} for instance."""
932
  return _ExpandItemName(cfg.ExpandInstanceName, name, "Instance")
933

    
934

    
935
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
936
                          memory, vcpus, nics, disk_template, disks,
937
                          bep, hvp, hypervisor_name, tags):
938
  """Builds instance related env variables for hooks
939

940
  This builds the hook environment from individual variables.
941

942
  @type name: string
943
  @param name: the name of the instance
944
  @type primary_node: string
945
  @param primary_node: the name of the instance's primary node
946
  @type secondary_nodes: list
947
  @param secondary_nodes: list of secondary nodes as strings
948
  @type os_type: string
949
  @param os_type: the name of the instance's OS
950
  @type status: boolean
951
  @param status: the should_run status of the instance
952
  @type memory: string
953
  @param memory: the memory size of the instance
954
  @type vcpus: string
955
  @param vcpus: the count of VCPUs the instance has
956
  @type nics: list
957
  @param nics: list of tuples (ip, mac, mode, link) representing
958
      the NICs the instance has
959
  @type disk_template: string
960
  @param disk_template: the disk template of the instance
961
  @type disks: list
962
  @param disks: the list of (size, mode) pairs
963
  @type bep: dict
964
  @param bep: the backend parameters for the instance
965
  @type hvp: dict
966
  @param hvp: the hypervisor parameters for the instance
967
  @type hypervisor_name: string
968
  @param hypervisor_name: the hypervisor for the instance
969
  @type tags: list
970
  @param tags: list of instance tags as strings
971
  @rtype: dict
972
  @return: the hook environment for this instance
973

974
  """
975
  if status:
976
    str_status = "up"
977
  else:
978
    str_status = "down"
979
  env = {
980
    "OP_TARGET": name,
981
    "INSTANCE_NAME": name,
982
    "INSTANCE_PRIMARY": primary_node,
983
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
984
    "INSTANCE_OS_TYPE": os_type,
985
    "INSTANCE_STATUS": str_status,
986
    "INSTANCE_MEMORY": memory,
987
    "INSTANCE_VCPUS": vcpus,
988
    "INSTANCE_DISK_TEMPLATE": disk_template,
989
    "INSTANCE_HYPERVISOR": hypervisor_name,
990
  }
991

    
992
  if nics:
993
    nic_count = len(nics)
994
    for idx, (ip, mac, mode, link) in enumerate(nics):
995
      if ip is None:
996
        ip = ""
997
      env["INSTANCE_NIC%d_IP" % idx] = ip
998
      env["INSTANCE_NIC%d_MAC" % idx] = mac
999
      env["INSTANCE_NIC%d_MODE" % idx] = mode
1000
      env["INSTANCE_NIC%d_LINK" % idx] = link
1001
      if mode == constants.NIC_MODE_BRIDGED:
1002
        env["INSTANCE_NIC%d_BRIDGE" % idx] = link
1003
  else:
1004
    nic_count = 0
1005

    
1006
  env["INSTANCE_NIC_COUNT"] = nic_count
1007

    
1008
  if disks:
1009
    disk_count = len(disks)
1010
    for idx, (size, mode) in enumerate(disks):
1011
      env["INSTANCE_DISK%d_SIZE" % idx] = size
1012
      env["INSTANCE_DISK%d_MODE" % idx] = mode
1013
  else:
1014
    disk_count = 0
1015

    
1016
  env["INSTANCE_DISK_COUNT"] = disk_count
1017

    
1018
  if not tags:
1019
    tags = []
1020

    
1021
  env["INSTANCE_TAGS"] = " ".join(tags)
1022

    
1023
  for source, kind in [(bep, "BE"), (hvp, "HV")]:
1024
    for key, value in source.items():
1025
      env["INSTANCE_%s_%s" % (kind, key)] = value
1026

    
1027
  return env
1028

    
1029

    
1030
def _NICListToTuple(lu, nics):
1031
  """Build a list of nic information tuples.
1032

1033
  This list is suitable to be passed to _BuildInstanceHookEnv or as a return
1034
  value in LUInstanceQueryData.
1035

1036
  @type lu:  L{LogicalUnit}
1037
  @param lu: the logical unit on whose behalf we execute
1038
  @type nics: list of L{objects.NIC}
1039
  @param nics: list of nics to convert to hooks tuples
1040

1041
  """
1042
  hooks_nics = []
1043
  cluster = lu.cfg.GetClusterInfo()
1044
  for nic in nics:
1045
    ip = nic.ip
1046
    mac = nic.mac
1047
    filled_params = cluster.SimpleFillNIC(nic.nicparams)
1048
    mode = filled_params[constants.NIC_MODE]
1049
    link = filled_params[constants.NIC_LINK]
1050
    hooks_nics.append((ip, mac, mode, link))
1051
  return hooks_nics
1052

    
1053

    
1054
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
1055
  """Builds instance related env variables for hooks from an object.
1056

1057
  @type lu: L{LogicalUnit}
1058
  @param lu: the logical unit on whose behalf we execute
1059
  @type instance: L{objects.Instance}
1060
  @param instance: the instance for which we should build the
1061
      environment
1062
  @type override: dict
1063
  @param override: dictionary with key/values that will override
1064
      our values
1065
  @rtype: dict
1066
  @return: the hook environment dictionary
1067

1068
  """
1069
  cluster = lu.cfg.GetClusterInfo()
1070
  bep = cluster.FillBE(instance)
1071
  hvp = cluster.FillHV(instance)
1072
  args = {
1073
    "name": instance.name,
1074
    "primary_node": instance.primary_node,
1075
    "secondary_nodes": instance.secondary_nodes,
1076
    "os_type": instance.os,
1077
    "status": instance.admin_up,
1078
    "memory": bep[constants.BE_MEMORY],
1079
    "vcpus": bep[constants.BE_VCPUS],
1080
    "nics": _NICListToTuple(lu, instance.nics),
1081
    "disk_template": instance.disk_template,
1082
    "disks": [(disk.size, disk.mode) for disk in instance.disks],
1083
    "bep": bep,
1084
    "hvp": hvp,
1085
    "hypervisor_name": instance.hypervisor,
1086
    "tags": instance.tags,
1087
  }
1088
  if override:
1089
    args.update(override)
1090
  return _BuildInstanceHookEnv(**args) # pylint: disable=W0142
1091

    
1092

    
1093
def _AdjustCandidatePool(lu, exceptions):
1094
  """Adjust the candidate pool after node operations.
1095

1096
  """
1097
  mod_list = lu.cfg.MaintainCandidatePool(exceptions)
1098
  if mod_list:
1099
    lu.LogInfo("Promoted nodes to master candidate role: %s",
1100
               utils.CommaJoin(node.name for node in mod_list))
1101
    for name in mod_list:
1102
      lu.context.ReaddNode(name)
1103
  mc_now, mc_max, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1104
  if mc_now > mc_max:
1105
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
1106
               (mc_now, mc_max))
1107

    
1108

    
1109
def _DecideSelfPromotion(lu, exceptions=None):
1110
  """Decide whether I should promote myself as a master candidate.
1111

1112
  """
1113
  cp_size = lu.cfg.GetClusterInfo().candidate_pool_size
1114
  mc_now, mc_should, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1115
  # the new node will increase mc_max with one, so:
1116
  mc_should = min(mc_should + 1, cp_size)
1117
  return mc_now < mc_should
1118

    
1119

    
1120
def _CheckNicsBridgesExist(lu, target_nics, target_node):
1121
  """Check that the brigdes needed by a list of nics exist.
1122

1123
  """
1124
  cluster = lu.cfg.GetClusterInfo()
1125
  paramslist = [cluster.SimpleFillNIC(nic.nicparams) for nic in target_nics]
1126
  brlist = [params[constants.NIC_LINK] for params in paramslist
1127
            if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
1128
  if brlist:
1129
    result = lu.rpc.call_bridges_exist(target_node, brlist)
1130
    result.Raise("Error checking bridges on destination node '%s'" %
1131
                 target_node, prereq=True, ecode=errors.ECODE_ENVIRON)
1132

    
1133

    
1134
def _CheckInstanceBridgesExist(lu, instance, node=None):
1135
  """Check that the brigdes needed by an instance exist.
1136

1137
  """
1138
  if node is None:
1139
    node = instance.primary_node
1140
  _CheckNicsBridgesExist(lu, instance.nics, node)
1141

    
1142

    
1143
def _CheckOSVariant(os_obj, name):
1144
  """Check whether an OS name conforms to the os variants specification.
1145

1146
  @type os_obj: L{objects.OS}
1147
  @param os_obj: OS object to check
1148
  @type name: string
1149
  @param name: OS name passed by the user, to check for validity
1150

1151
  """
1152
  variant = objects.OS.GetVariant(name)
1153
  if not os_obj.supported_variants:
1154
    if variant:
1155
      raise errors.OpPrereqError("OS '%s' doesn't support variants ('%s'"
1156
                                 " passed)" % (os_obj.name, variant),
1157
                                 errors.ECODE_INVAL)
1158
    return
1159
  if not variant:
1160
    raise errors.OpPrereqError("OS name must include a variant",
1161
                               errors.ECODE_INVAL)
1162

    
1163
  if variant not in os_obj.supported_variants:
1164
    raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)
1165

    
1166

    
1167
def _GetNodeInstancesInner(cfg, fn):
1168
  return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]
1169

    
1170

    
1171
def _GetNodeInstances(cfg, node_name):
1172
  """Returns a list of all primary and secondary instances on a node.
1173

1174
  """
1175

    
1176
  return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)
1177

    
1178

    
1179
def _GetNodePrimaryInstances(cfg, node_name):
1180
  """Returns primary instances on a node.
1181

1182
  """
1183
  return _GetNodeInstancesInner(cfg,
1184
                                lambda inst: node_name == inst.primary_node)
1185

    
1186

    
1187
def _GetNodeSecondaryInstances(cfg, node_name):
1188
  """Returns secondary instances on a node.
1189

1190
  """
1191
  return _GetNodeInstancesInner(cfg,
1192
                                lambda inst: node_name in inst.secondary_nodes)
1193

    
1194

    
1195
def _GetStorageTypeArgs(cfg, storage_type):
1196
  """Returns the arguments for a storage type.
1197

1198
  """
1199
  # Special case for file storage
1200
  if storage_type == constants.ST_FILE:
1201
    # storage.FileStorage wants a list of storage directories
1202
    return [[cfg.GetFileStorageDir(), cfg.GetSharedFileStorageDir()]]
1203

    
1204
  return []
1205

    
1206

    
1207
def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
1208
  faulty = []
1209

    
1210
  for dev in instance.disks:
1211
    cfg.SetDiskID(dev, node_name)
1212

    
1213
  result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
1214
  result.Raise("Failed to get disk status from node %s" % node_name,
1215
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
1216

    
1217
  for idx, bdev_status in enumerate(result.payload):
1218
    if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
1219
      faulty.append(idx)
1220

    
1221
  return faulty
1222

    
1223

    
1224
def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
1225
  """Check the sanity of iallocator and node arguments and use the
1226
  cluster-wide iallocator if appropriate.
1227

1228
  Check that at most one of (iallocator, node) is specified. If none is
1229
  specified, then the LU's opcode's iallocator slot is filled with the
1230
  cluster-wide default iallocator.
1231

1232
  @type iallocator_slot: string
1233
  @param iallocator_slot: the name of the opcode iallocator slot
1234
  @type node_slot: string
1235
  @param node_slot: the name of the opcode target node slot
1236

1237
  """
1238
  node = getattr(lu.op, node_slot, None)
1239
  iallocator = getattr(lu.op, iallocator_slot, None)
1240

    
1241
  if node is not None and iallocator is not None:
1242
    raise errors.OpPrereqError("Do not specify both, iallocator and node",
1243
                               errors.ECODE_INVAL)
1244
  elif node is None and iallocator is None:
1245
    default_iallocator = lu.cfg.GetDefaultIAllocator()
1246
    if default_iallocator:
1247
      setattr(lu.op, iallocator_slot, default_iallocator)
1248
    else:
1249
      raise errors.OpPrereqError("No iallocator or node given and no"
1250
                                 " cluster-wide default iallocator found;"
1251
                                 " please specify either an iallocator or a"
1252
                                 " node, or set a cluster-wide default"
1253
                                 " iallocator")
1254

    
1255

    
1256
def _GetDefaultIAllocator(cfg, iallocator):
1257
  """Decides on which iallocator to use.
1258

1259
  @type cfg: L{config.ConfigWriter}
1260
  @param cfg: Cluster configuration object
1261
  @type iallocator: string or None
1262
  @param iallocator: Iallocator specified in opcode
1263
  @rtype: string
1264
  @return: Iallocator name
1265

1266
  """
1267
  if not iallocator:
1268
    # Use default iallocator
1269
    iallocator = cfg.GetDefaultIAllocator()
1270

    
1271
  if not iallocator:
1272
    raise errors.OpPrereqError("No iallocator was specified, neither in the"
1273
                               " opcode nor as a cluster-wide default",
1274
                               errors.ECODE_INVAL)
1275

    
1276
  return iallocator
1277

    
1278

    
1279
class LUClusterPostInit(LogicalUnit):
1280
  """Logical unit for running hooks after cluster initialization.
1281

1282
  """
1283
  HPATH = "cluster-init"
1284
  HTYPE = constants.HTYPE_CLUSTER
1285

    
1286
  def BuildHooksEnv(self):
1287
    """Build hooks env.
1288

1289
    """
1290
    return {
1291
      "OP_TARGET": self.cfg.GetClusterName(),
1292
      }
1293

    
1294
  def BuildHooksNodes(self):
1295
    """Build hooks nodes.
1296

1297
    """
1298
    return ([], [self.cfg.GetMasterNode()])
1299

    
1300
  def Exec(self, feedback_fn):
1301
    """Nothing to do.
1302

1303
    """
1304
    return True
1305

    
1306

    
1307
class LUClusterDestroy(LogicalUnit):
1308
  """Logical unit for destroying the cluster.
1309

1310
  """
1311
  HPATH = "cluster-destroy"
1312
  HTYPE = constants.HTYPE_CLUSTER
1313

    
1314
  def BuildHooksEnv(self):
1315
    """Build hooks env.
1316

1317
    """
1318
    return {
1319
      "OP_TARGET": self.cfg.GetClusterName(),
1320
      }
1321

    
1322
  def BuildHooksNodes(self):
1323
    """Build hooks nodes.
1324

1325
    """
1326
    return ([], [])
1327

    
1328
  def CheckPrereq(self):
1329
    """Check prerequisites.
1330

1331
    This checks whether the cluster is empty.
1332

1333
    Any errors are signaled by raising errors.OpPrereqError.
1334

1335
    """
1336
    master = self.cfg.GetMasterNode()
1337

    
1338
    nodelist = self.cfg.GetNodeList()
1339
    if len(nodelist) != 1 or nodelist[0] != master:
1340
      raise errors.OpPrereqError("There are still %d node(s) in"
1341
                                 " this cluster." % (len(nodelist) - 1),
1342
                                 errors.ECODE_INVAL)
1343
    instancelist = self.cfg.GetInstanceList()
1344
    if instancelist:
1345
      raise errors.OpPrereqError("There are still %d instance(s) in"
1346
                                 " this cluster." % len(instancelist),
1347
                                 errors.ECODE_INVAL)
1348

    
1349
  def Exec(self, feedback_fn):
1350
    """Destroys the cluster.
1351

1352
    """
1353
    master = self.cfg.GetMasterNode()
1354

    
1355
    # Run post hooks on master node before it's removed
1356
    _RunPostHook(self, master)
1357

    
1358
    result = self.rpc.call_node_stop_master(master, False)
1359
    result.Raise("Could not disable the master role")
1360

    
1361
    return master
1362

    
1363

    
1364
def _VerifyCertificate(filename):
1365
  """Verifies a certificate for L{LUClusterVerifyConfig}.
1366

1367
  @type filename: string
1368
  @param filename: Path to PEM file
1369

1370
  """
1371
  try:
1372
    cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1373
                                           utils.ReadFile(filename))
1374
  except Exception, err: # pylint: disable=W0703
1375
    return (LUClusterVerifyConfig.ETYPE_ERROR,
1376
            "Failed to load X509 certificate %s: %s" % (filename, err))
1377

    
1378
  (errcode, msg) = \
1379
    utils.VerifyX509Certificate(cert, constants.SSL_CERT_EXPIRATION_WARN,
1380
                                constants.SSL_CERT_EXPIRATION_ERROR)
1381

    
1382
  if msg:
1383
    fnamemsg = "While verifying %s: %s" % (filename, msg)
1384
  else:
1385
    fnamemsg = None
1386

    
1387
  if errcode is None:
1388
    return (None, fnamemsg)
1389
  elif errcode == utils.CERT_WARNING:
1390
    return (LUClusterVerifyConfig.ETYPE_WARNING, fnamemsg)
1391
  elif errcode == utils.CERT_ERROR:
1392
    return (LUClusterVerifyConfig.ETYPE_ERROR, fnamemsg)
1393

    
1394
  raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
1395

    
1396

    
1397
def _GetAllHypervisorParameters(cluster, instances):
1398
  """Compute the set of all hypervisor parameters.
1399

1400
  @type cluster: L{objects.Cluster}
1401
  @param cluster: the cluster object
1402
  @param instances: list of L{objects.Instance}
1403
  @param instances: additional instances from which to obtain parameters
1404
  @rtype: list of (origin, hypervisor, parameters)
1405
  @return: a list with all parameters found, indicating the hypervisor they
1406
       apply to, and the origin (can be "cluster", "os X", or "instance Y")
1407

1408
  """
1409
  hvp_data = []
1410

    
1411
  for hv_name in cluster.enabled_hypervisors:
1412
    hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
1413

    
1414
  for os_name, os_hvp in cluster.os_hvp.items():
1415
    for hv_name, hv_params in os_hvp.items():
1416
      if hv_params:
1417
        full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
1418
        hvp_data.append(("os %s" % os_name, hv_name, full_params))
1419

    
1420
  # TODO: collapse identical parameter values in a single one
1421
  for instance in instances:
1422
    if instance.hvparams:
1423
      hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
1424
                       cluster.FillHV(instance)))
1425

    
1426
  return hvp_data
1427

    
1428

    
1429
class _VerifyErrors(object):
1430
  """Mix-in for cluster/group verify LUs.
1431

1432
  It provides _Error and _ErrorIf, and updates the self.bad boolean. (Expects
1433
  self.op and self._feedback_fn to be available.)
1434

1435
  """
1436
  TCLUSTER = "cluster"
1437
  TNODE = "node"
1438
  TINSTANCE = "instance"
1439

    
1440
  ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
1441
  ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
1442
  ECLUSTERFILECHECK = (TCLUSTER, "ECLUSTERFILECHECK")
1443
  ECLUSTERDANGLINGNODES = (TNODE, "ECLUSTERDANGLINGNODES")
1444
  ECLUSTERDANGLINGINST = (TNODE, "ECLUSTERDANGLINGINST")
1445
  EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
1446
  EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
1447
  EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
1448
  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
1449
  EINSTANCEFAULTYDISK = (TINSTANCE, "EINSTANCEFAULTYDISK")
1450
  EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
1451
  EINSTANCESPLITGROUPS = (TINSTANCE, "EINSTANCESPLITGROUPS")
1452
  ENODEDRBD = (TNODE, "ENODEDRBD")
1453
  ENODEDRBDHELPER = (TNODE, "ENODEDRBDHELPER")
1454
  ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
1455
  ENODEHOOKS = (TNODE, "ENODEHOOKS")
1456
  ENODEHV = (TNODE, "ENODEHV")
1457
  ENODELVM = (TNODE, "ENODELVM")
1458
  ENODEN1 = (TNODE, "ENODEN1")
1459
  ENODENET = (TNODE, "ENODENET")
1460
  ENODEOS = (TNODE, "ENODEOS")
1461
  ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
1462
  ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
1463
  ENODERPC = (TNODE, "ENODERPC")
1464
  ENODESSH = (TNODE, "ENODESSH")
1465
  ENODEVERSION = (TNODE, "ENODEVERSION")
1466
  ENODESETUP = (TNODE, "ENODESETUP")
1467
  ENODETIME = (TNODE, "ENODETIME")
1468
  ENODEOOBPATH = (TNODE, "ENODEOOBPATH")
1469

    
1470
  ETYPE_FIELD = "code"
1471
  ETYPE_ERROR = "ERROR"
1472
  ETYPE_WARNING = "WARNING"
1473

    
1474
  def _Error(self, ecode, item, msg, *args, **kwargs):
1475
    """Format an error message.
1476

1477
    Based on the opcode's error_codes parameter, either format a
1478
    parseable error code, or a simpler error string.
1479

1480
    This must be called only from Exec and functions called from Exec.
1481

1482
    """
1483
    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
1484
    itype, etxt = ecode
1485
    # first complete the msg
1486
    if args:
1487
      msg = msg % args
1488
    # then format the whole message
1489
    if self.op.error_codes: # This is a mix-in. pylint: disable=E1101
1490
      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
1491
    else:
1492
      if item:
1493
        item = " " + item
1494
      else:
1495
        item = ""
1496
      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
1497
    # and finally report it via the feedback_fn
1498
    self._feedback_fn("  - %s" % msg) # Mix-in. pylint: disable=E1101
1499

    
1500
  def _ErrorIf(self, cond, *args, **kwargs):
1501
    """Log an error message if the passed condition is True.
1502

1503
    """
1504
    cond = (bool(cond)
1505
            or self.op.debug_simulate_errors) # pylint: disable=E1101
1506
    if cond:
1507
      self._Error(*args, **kwargs)
1508
    # do not mark the operation as failed for WARN cases only
1509
    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
1510
      self.bad = self.bad or cond
1511

    
1512

    
1513
class LUClusterVerify(NoHooksLU):
1514
  """Submits all jobs necessary to verify the cluster.
1515

1516
  """
1517
  REQ_BGL = False
1518

    
1519
  def ExpandNames(self):
1520
    self.needed_locks = {}
1521

    
1522
  def Exec(self, feedback_fn):
1523
    jobs = []
1524

    
1525
    if self.op.group_name:
1526
      groups = [self.op.group_name]
1527
      depends_fn = lambda: None
1528
    else:
1529
      groups = self.cfg.GetNodeGroupList()
1530

    
1531
      # Verify global configuration
1532
      jobs.append([opcodes.OpClusterVerifyConfig()])
1533

    
1534
      # Always depend on global verification
1535
      depends_fn = lambda: [(-len(jobs), [])]
1536

    
1537
    jobs.extend([opcodes.OpClusterVerifyGroup(group_name=group,
1538
                                              depends=depends_fn())]
1539
                for group in groups)
1540

    
1541
    # Fix up all parameters
1542
    for op in itertools.chain(*jobs): # pylint: disable=W0142
1543
      op.debug_simulate_errors = self.op.debug_simulate_errors
1544
      op.verbose = self.op.verbose
1545
      op.error_codes = self.op.error_codes
1546
      try:
1547
        op.skip_checks = self.op.skip_checks
1548
      except AttributeError:
1549
        assert not isinstance(op, opcodes.OpClusterVerifyGroup)
1550

    
1551
    return ResultWithJobs(jobs)
1552

    
1553

    
1554
class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
1555
  """Verifies the cluster config.
1556

1557
  """
1558
  REQ_BGL = True
1559

    
1560
  def _VerifyHVP(self, hvp_data):
1561
    """Verifies locally the syntax of the hypervisor parameters.
1562

1563
    """
1564
    for item, hv_name, hv_params in hvp_data:
1565
      msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
1566
             (item, hv_name))
1567
      try:
1568
        hv_class = hypervisor.GetHypervisor(hv_name)
1569
        utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1570
        hv_class.CheckParameterSyntax(hv_params)
1571
      except errors.GenericError, err:
1572
        self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
1573

    
1574
  def ExpandNames(self):
1575
    # Information can be safely retrieved as the BGL is acquired in exclusive
1576
    # mode
1577
    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER)
1578
    self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
1579
    self.all_node_info = self.cfg.GetAllNodesInfo()
1580
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1581
    self.needed_locks = {}
1582

    
1583
  def Exec(self, feedback_fn):
1584
    """Verify integrity of cluster, performing various test on nodes.
1585

1586
    """
1587
    self.bad = False
1588
    self._feedback_fn = feedback_fn
1589

    
1590
    feedback_fn("* Verifying cluster config")
1591

    
1592
    for msg in self.cfg.VerifyConfig():
1593
      self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
1594

    
1595
    feedback_fn("* Verifying cluster certificate files")
1596

    
1597
    for cert_filename in constants.ALL_CERT_FILES:
1598
      (errcode, msg) = _VerifyCertificate(cert_filename)
1599
      self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
1600

    
1601
    feedback_fn("* Verifying hypervisor parameters")
1602

    
1603
    self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
1604
                                                self.all_inst_info.values()))
1605

    
1606
    feedback_fn("* Verifying all nodes belong to an existing group")
1607

    
1608
    # We do this verification here because, should this bogus circumstance
1609
    # occur, it would never be caught by VerifyGroup, which only acts on
1610
    # nodes/instances reachable from existing node groups.
1611

    
1612
    dangling_nodes = set(node.name for node in self.all_node_info.values()
1613
                         if node.group not in self.all_group_info)
1614

    
1615
    dangling_instances = {}
1616
    no_node_instances = []
1617

    
1618
    for inst in self.all_inst_info.values():
1619
      if inst.primary_node in dangling_nodes:
1620
        dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
1621
      elif inst.primary_node not in self.all_node_info:
1622
        no_node_instances.append(inst.name)
1623

    
1624
    pretty_dangling = [
1625
        "%s (%s)" %
1626
        (node.name,
1627
         utils.CommaJoin(dangling_instances.get(node.name,
1628
                                                ["no instances"])))
1629
        for node in dangling_nodes]
1630

    
1631
    self._ErrorIf(bool(dangling_nodes), self.ECLUSTERDANGLINGNODES, None,
1632
                  "the following nodes (and their instances) belong to a non"
1633
                  " existing group: %s", utils.CommaJoin(pretty_dangling))
1634

    
1635
    self._ErrorIf(bool(no_node_instances), self.ECLUSTERDANGLINGINST, None,
1636
                  "the following instances have a non-existing primary-node:"
1637
                  " %s", utils.CommaJoin(no_node_instances))
1638

    
1639
    return not self.bad
1640

    
1641

    
1642
class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
1643
  """Verifies the status of a node group.
1644

1645
  """
1646
  HPATH = "cluster-verify"
1647
  HTYPE = constants.HTYPE_CLUSTER
1648
  REQ_BGL = False
1649

    
1650
  _HOOKS_INDENT_RE = re.compile("^", re.M)
1651

    
1652
  class NodeImage(object):
1653
    """A class representing the logical and physical status of a node.
1654

1655
    @type name: string
1656
    @ivar name: the node name to which this object refers
1657
    @ivar volumes: a structure as returned from
1658
        L{ganeti.backend.GetVolumeList} (runtime)
1659
    @ivar instances: a list of running instances (runtime)
1660
    @ivar pinst: list of configured primary instances (config)
1661
    @ivar sinst: list of configured secondary instances (config)
1662
    @ivar sbp: dictionary of {primary-node: list of instances} for all
1663
        instances for which this node is secondary (config)
1664
    @ivar mfree: free memory, as reported by hypervisor (runtime)
1665
    @ivar dfree: free disk, as reported by the node (runtime)
1666
    @ivar offline: the offline status (config)
1667
    @type rpc_fail: boolean
1668
    @ivar rpc_fail: whether the RPC verify call was successfull (overall,
1669
        not whether the individual keys were correct) (runtime)
1670
    @type lvm_fail: boolean
1671
    @ivar lvm_fail: whether the RPC call didn't return valid LVM data
1672
    @type hyp_fail: boolean
1673
    @ivar hyp_fail: whether the RPC call didn't return the instance list
1674
    @type ghost: boolean
1675
    @ivar ghost: whether this is a known node or not (config)
1676
    @type os_fail: boolean
1677
    @ivar os_fail: whether the RPC call didn't return valid OS data
1678
    @type oslist: list
1679
    @ivar oslist: list of OSes as diagnosed by DiagnoseOS
1680
    @type vm_capable: boolean
1681
    @ivar vm_capable: whether the node can host instances
1682

1683
    """
1684
    def __init__(self, offline=False, name=None, vm_capable=True):
1685
      self.name = name
1686
      self.volumes = {}
1687
      self.instances = []
1688
      self.pinst = []
1689
      self.sinst = []
1690
      self.sbp = {}
1691
      self.mfree = 0
1692
      self.dfree = 0
1693
      self.offline = offline
1694
      self.vm_capable = vm_capable
1695
      self.rpc_fail = False
1696
      self.lvm_fail = False
1697
      self.hyp_fail = False
1698
      self.ghost = False
1699
      self.os_fail = False
1700
      self.oslist = {}
1701

    
1702
  def ExpandNames(self):
1703
    # This raises errors.OpPrereqError on its own:
1704
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
1705

    
1706
    # Get instances in node group; this is unsafe and needs verification later
1707
    inst_names = self.cfg.GetNodeGroupInstances(self.group_uuid)
1708

    
1709
    self.needed_locks = {
1710
      locking.LEVEL_INSTANCE: inst_names,
1711
      locking.LEVEL_NODEGROUP: [self.group_uuid],
1712
      locking.LEVEL_NODE: [],
1713
      }
1714

    
1715
    self.share_locks = _ShareAll()
1716

    
1717
  def DeclareLocks(self, level):
1718
    if level == locking.LEVEL_NODE:
1719
      # Get members of node group; this is unsafe and needs verification later
1720
      nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
1721

    
1722
      all_inst_info = self.cfg.GetAllInstancesInfo()
1723

    
1724
      # In Exec(), we warn about mirrored instances that have primary and
1725
      # secondary living in separate node groups. To fully verify that
1726
      # volumes for these instances are healthy, we will need to do an
1727
      # extra call to their secondaries. We ensure here those nodes will
1728
      # be locked.
1729
      for inst in self.owned_locks(locking.LEVEL_INSTANCE):
1730
        # Important: access only the instances whose lock is owned
1731
        if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
1732
          nodes.update(all_inst_info[inst].secondary_nodes)
1733

    
1734
      self.needed_locks[locking.LEVEL_NODE] = nodes
1735

    
1736
  def CheckPrereq(self):
1737
    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
1738
    self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
1739

    
1740
    group_nodes = set(self.group_info.members)
1741
    group_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
1742

    
1743
    unlocked_nodes = \
1744
        group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1745

    
1746
    unlocked_instances = \
1747
        group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
1748

    
1749
    if unlocked_nodes:
1750
      raise errors.OpPrereqError("Missing lock for nodes: %s" %
1751
                                 utils.CommaJoin(unlocked_nodes))
1752

    
1753
    if unlocked_instances:
1754
      raise errors.OpPrereqError("Missing lock for instances: %s" %
1755
                                 utils.CommaJoin(unlocked_instances))
1756

    
1757
    self.all_node_info = self.cfg.GetAllNodesInfo()
1758
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1759

    
1760
    self.my_node_names = utils.NiceSort(group_nodes)
1761
    self.my_inst_names = utils.NiceSort(group_instances)
1762

    
1763
    self.my_node_info = dict((name, self.all_node_info[name])
1764
                             for name in self.my_node_names)
1765

    
1766
    self.my_inst_info = dict((name, self.all_inst_info[name])
1767
                             for name in self.my_inst_names)
1768

    
1769
    # We detect here the nodes that will need the extra RPC calls for verifying
1770
    # split LV volumes; they should be locked.
1771
    extra_lv_nodes = set()
1772

    
1773
    for inst in self.my_inst_info.values():
1774
      if inst.disk_template in constants.DTS_INT_MIRROR:
1775
        group = self.my_node_info[inst.primary_node].group
1776
        for nname in inst.secondary_nodes:
1777
          if self.all_node_info[nname].group != group:
1778
            extra_lv_nodes.add(nname)
1779

    
1780
    unlocked_lv_nodes = \
1781
        extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1782

    
1783
    if unlocked_lv_nodes:
1784
      raise errors.OpPrereqError("these nodes could be locked: %s" %
1785
                                 utils.CommaJoin(unlocked_lv_nodes))
1786
    self.extra_lv_nodes = list(extra_lv_nodes)
1787

    
1788
  def _VerifyNode(self, ninfo, nresult):
1789
    """Perform some basic validation on data returned from a node.
1790

1791
      - check the result data structure is well formed and has all the
1792
        mandatory fields
1793
      - check ganeti version
1794

1795
    @type ninfo: L{objects.Node}
1796
    @param ninfo: the node to check
1797
    @param nresult: the results from the node
1798
    @rtype: boolean
1799
    @return: whether overall this call was successful (and we can expect
1800
         reasonable values in the respose)
1801

1802
    """
1803
    node = ninfo.name
1804
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1805

    
1806
    # main result, nresult should be a non-empty dict
1807
    test = not nresult or not isinstance(nresult, dict)
1808
    _ErrorIf(test, self.ENODERPC, node,
1809
                  "unable to verify node: no data returned")
1810
    if test:
1811
      return False
1812

    
1813
    # compares ganeti version
1814
    local_version = constants.PROTOCOL_VERSION
1815
    remote_version = nresult.get("version", None)
1816
    test = not (remote_version and
1817
                isinstance(remote_version, (list, tuple)) and
1818
                len(remote_version) == 2)
1819
    _ErrorIf(test, self.ENODERPC, node,
1820
             "connection to node returned invalid data")
1821
    if test:
1822
      return False
1823

    
1824
    test = local_version != remote_version[0]
1825
    _ErrorIf(test, self.ENODEVERSION, node,
1826
             "incompatible protocol versions: master %s,"
1827
             " node %s", local_version, remote_version[0])
1828
    if test:
1829
      return False
1830

    
1831
    # node seems compatible, we can actually try to look into its results
1832

    
1833
    # full package version
1834
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1835
                  self.ENODEVERSION, node,
1836
                  "software version mismatch: master %s, node %s",
1837
                  constants.RELEASE_VERSION, remote_version[1],
1838
                  code=self.ETYPE_WARNING)
1839

    
1840
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1841
    if ninfo.vm_capable and isinstance(hyp_result, dict):
1842
      for hv_name, hv_result in hyp_result.iteritems():
1843
        test = hv_result is not None
1844
        _ErrorIf(test, self.ENODEHV, node,
1845
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1846

    
1847
    hvp_result = nresult.get(constants.NV_HVPARAMS, None)
1848
    if ninfo.vm_capable and isinstance(hvp_result, list):
1849
      for item, hv_name, hv_result in hvp_result:
1850
        _ErrorIf(True, self.ENODEHV, node,
1851
                 "hypervisor %s parameter verify failure (source %s): %s",
1852
                 hv_name, item, hv_result)
1853

    
1854
    test = nresult.get(constants.NV_NODESETUP,
1855
                       ["Missing NODESETUP results"])
1856
    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
1857
             "; ".join(test))
1858

    
1859
    return True
1860

    
1861
  def _VerifyNodeTime(self, ninfo, nresult,
1862
                      nvinfo_starttime, nvinfo_endtime):
1863
    """Check the node time.
1864

1865
    @type ninfo: L{objects.Node}
1866
    @param ninfo: the node to check
1867
    @param nresult: the remote results for the node
1868
    @param nvinfo_starttime: the start time of the RPC call
1869
    @param nvinfo_endtime: the end time of the RPC call
1870

1871
    """
1872
    node = ninfo.name
1873
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1874

    
1875
    ntime = nresult.get(constants.NV_TIME, None)
1876
    try:
1877
      ntime_merged = utils.MergeTime(ntime)
1878
    except (ValueError, TypeError):
1879
      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
1880
      return
1881

    
1882
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1883
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1884
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1885
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1886
    else:
1887
      ntime_diff = None
1888

    
1889
    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
1890
             "Node time diverges by at least %s from master node time",
1891
             ntime_diff)
1892

    
1893
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1894
    """Check the node LVM results.
1895

1896
    @type ninfo: L{objects.Node}
1897
    @param ninfo: the node to check
1898
    @param nresult: the remote results for the node
1899
    @param vg_name: the configured VG name
1900

1901
    """
1902
    if vg_name is None:
1903
      return
1904

    
1905
    node = ninfo.name
1906
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1907

    
1908
    # checks vg existence and size > 20G
1909
    vglist = nresult.get(constants.NV_VGLIST, None)
1910
    test = not vglist
1911
    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
1912
    if not test:
1913
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
1914
                                            constants.MIN_VG_SIZE)
1915
      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
1916

    
1917
    # check pv names
1918
    pvlist = nresult.get(constants.NV_PVLIST, None)
1919
    test = pvlist is None
1920
    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
1921
    if not test:
1922
      # check that ':' is not present in PV names, since it's a
1923
      # special character for lvcreate (denotes the range of PEs to
1924
      # use on the PV)
1925
      for _, pvname, owner_vg in pvlist:
1926
        test = ":" in pvname
1927
        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
1928
                 " '%s' of VG '%s'", pvname, owner_vg)
1929

    
1930
  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
1931
    """Check the node bridges.
1932

1933
    @type ninfo: L{objects.Node}
1934
    @param ninfo: the node to check
1935
    @param nresult: the remote results for the node
1936
    @param bridges: the expected list of bridges
1937

1938
    """
1939
    if not bridges:
1940
      return
1941

    
1942
    node = ninfo.name
1943
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1944

    
1945
    missing = nresult.get(constants.NV_BRIDGES, None)
1946
    test = not isinstance(missing, list)
1947
    _ErrorIf(test, self.ENODENET, node,
1948
             "did not return valid bridge information")
1949
    if not test:
1950
      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
1951
               utils.CommaJoin(sorted(missing)))
1952

    
1953
  def _VerifyNodeNetwork(self, ninfo, nresult):
1954
    """Check the node network connectivity results.
1955

1956
    @type ninfo: L{objects.Node}
1957
    @param ninfo: the node to check
1958
    @param nresult: the remote results for the node
1959

1960
    """
1961
    node = ninfo.name
1962
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1963

    
1964
    test = constants.NV_NODELIST not in nresult
1965
    _ErrorIf(test, self.ENODESSH, node,
1966
             "node hasn't returned node ssh connectivity data")
1967
    if not test:
1968
      if nresult[constants.NV_NODELIST]:
1969
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1970
          _ErrorIf(True, self.ENODESSH, node,
1971
                   "ssh communication with node '%s': %s", a_node, a_msg)
1972

    
1973
    test = constants.NV_NODENETTEST not in nresult
1974
    _ErrorIf(test, self.ENODENET, node,
1975
             "node hasn't returned node tcp connectivity data")
1976
    if not test:
1977
      if nresult[constants.NV_NODENETTEST]:
1978
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1979
        for anode in nlist:
1980
          _ErrorIf(True, self.ENODENET, node,
1981
                   "tcp communication with node '%s': %s",
1982
                   anode, nresult[constants.NV_NODENETTEST][anode])
1983

    
1984
    test = constants.NV_MASTERIP not in nresult
1985
    _ErrorIf(test, self.ENODENET, node,
1986
             "node hasn't returned node master IP reachability data")
1987
    if not test:
1988
      if not nresult[constants.NV_MASTERIP]:
1989
        if node == self.master_node:
1990
          msg = "the master node cannot reach the master IP (not configured?)"
1991
        else:
1992
          msg = "cannot reach the master IP"
1993
        _ErrorIf(True, self.ENODENET, node, msg)
1994

    
1995
  def _VerifyInstance(self, instance, instanceconfig, node_image,
1996
                      diskstatus):
1997
    """Verify an instance.
1998

1999
    This function checks to see if the required block devices are
2000
    available on the instance's node.
2001

2002
    """
2003
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2004
    node_current = instanceconfig.primary_node
2005

    
2006
    node_vol_should = {}
2007
    instanceconfig.MapLVsByNode(node_vol_should)
2008

    
2009
    for node in node_vol_should:
2010
      n_img = node_image[node]
2011
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2012
        # ignore missing volumes on offline or broken nodes
2013
        continue
2014
      for volume in node_vol_should[node]:
2015
        test = volume not in n_img.volumes
2016
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
2017
                 "volume %s missing on node %s", volume, node)
2018

    
2019
    if instanceconfig.admin_up:
2020
      pri_img = node_image[node_current]
2021
      test = instance not in pri_img.instances and not pri_img.offline
2022
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
2023
               "instance not running on its primary node %s",
2024
               node_current)
2025

    
2026
    diskdata = [(nname, success, status, idx)
2027
                for (nname, disks) in diskstatus.items()
2028
                for idx, (success, status) in enumerate(disks)]
2029

    
2030
    for nname, success, bdev_status, idx in diskdata:
2031
      # the 'ghost node' construction in Exec() ensures that we have a
2032
      # node here
2033
      snode = node_image[nname]
2034
      bad_snode = snode.ghost or snode.offline
2035
      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
2036
               self.EINSTANCEFAULTYDISK, instance,
2037
               "couldn't retrieve status for disk/%s on %s: %s",
2038
               idx, nname, bdev_status)
2039
      _ErrorIf((instanceconfig.admin_up and success and
2040
                bdev_status.ldisk_status == constants.LDS_FAULTY),
2041
               self.EINSTANCEFAULTYDISK, instance,
2042
               "disk/%s on %s is faulty", idx, nname)
2043

    
2044
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
2045
    """Verify if there are any unknown volumes in the cluster.
2046

2047
    The .os, .swap and backup volumes are ignored. All other volumes are
2048
    reported as unknown.
2049

2050
    @type reserved: L{ganeti.utils.FieldSet}
2051
    @param reserved: a FieldSet of reserved volume names
2052

2053
    """
2054
    for node, n_img in node_image.items():
2055
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2056
        # skip non-healthy nodes
2057
        continue
2058
      for volume in n_img.volumes:
2059
        test = ((node not in node_vol_should or
2060
                volume not in node_vol_should[node]) and
2061
                not reserved.Matches(volume))
2062
        self._ErrorIf(test, self.ENODEORPHANLV, node,
2063
                      "volume %s is unknown", volume)
2064

    
2065
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
2066
    """Verify N+1 Memory Resilience.
2067

2068
    Check that if one single node dies we can still start all the
2069
    instances it was primary for.
2070

2071
    """
2072
    cluster_info = self.cfg.GetClusterInfo()
2073
    for node, n_img in node_image.items():
2074
      # This code checks that every node which is now listed as
2075
      # secondary has enough memory to host all instances it is
2076
      # supposed to should a single other node in the cluster fail.
2077
      # FIXME: not ready for failover to an arbitrary node
2078
      # FIXME: does not support file-backed instances
2079
      # WARNING: we currently take into account down instances as well
2080
      # as up ones, considering that even if they're down someone
2081
      # might want to start them even in the event of a node failure.
2082
      if n_img.offline:
2083
        # we're skipping offline nodes from the N+1 warning, since
2084
        # most likely we don't have good memory infromation from them;
2085
        # we already list instances living on such nodes, and that's
2086
        # enough warning
2087
        continue
2088
      for prinode, instances in n_img.sbp.items():
2089
        needed_mem = 0
2090
        for instance in instances:
2091
          bep = cluster_info.FillBE(instance_cfg[instance])
2092
          if bep[constants.BE_AUTO_BALANCE]:
2093
            needed_mem += bep[constants.BE_MEMORY]
2094
        test = n_img.mfree < needed_mem
2095
        self._ErrorIf(test, self.ENODEN1, node,
2096
                      "not enough memory to accomodate instance failovers"
2097
                      " should node %s fail (%dMiB needed, %dMiB available)",
2098
                      prinode, needed_mem, n_img.mfree)
2099

    
2100
  @classmethod
2101
  def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
2102
                   (files_all, files_all_opt, files_mc, files_vm)):
2103
    """Verifies file checksums collected from all nodes.
2104

2105
    @param errorif: Callback for reporting errors
2106
    @param nodeinfo: List of L{objects.Node} objects
2107
    @param master_node: Name of master node
2108
    @param all_nvinfo: RPC results
2109

2110
    """
2111
    node_names = frozenset(node.name for node in nodeinfo if not node.offline)
2112

    
2113
    assert master_node in node_names
2114
    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
2115
            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
2116
           "Found file listed in more than one file list"
2117

    
2118
    # Define functions determining which nodes to consider for a file
2119
    file2nodefn = dict([(filename, fn)
2120
      for (files, fn) in [(files_all, None),
2121
                          (files_all_opt, None),
2122
                          (files_mc, lambda node: (node.master_candidate or
2123
                                                   node.name == master_node)),
2124
                          (files_vm, lambda node: node.vm_capable)]
2125
      for filename in files])
2126

    
2127
    fileinfo = dict((filename, {}) for filename in file2nodefn.keys())
2128

    
2129
    for node in nodeinfo:
2130
      if node.offline:
2131
        continue
2132

    
2133
      nresult = all_nvinfo[node.name]
2134

    
2135
      if nresult.fail_msg or not nresult.payload:
2136
        node_files = None
2137
      else:
2138
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
2139

    
2140
      test = not (node_files and isinstance(node_files, dict))
2141
      errorif(test, cls.ENODEFILECHECK, node.name,
2142
              "Node did not return file checksum data")
2143
      if test:
2144
        continue
2145

    
2146
      for (filename, checksum) in node_files.items():
2147
        # Check if the file should be considered for a node
2148
        fn = file2nodefn[filename]
2149
        if fn is None or fn(node):
2150
          fileinfo[filename].setdefault(checksum, set()).add(node.name)
2151

    
2152
    for (filename, checksums) in fileinfo.items():
2153
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2154

    
2155
      # Nodes having the file
2156
      with_file = frozenset(node_name
2157
                            for nodes in fileinfo[filename].values()
2158
                            for node_name in nodes)
2159

    
2160
      # Nodes missing file
2161
      missing_file = node_names - with_file
2162

    
2163
      if filename in files_all_opt:
2164
        # All or no nodes
2165
        errorif(missing_file and missing_file != node_names,
2166
                cls.ECLUSTERFILECHECK, None,
2167
                "File %s is optional, but it must exist on all or no"
2168
                " nodes (not found on %s)",
2169
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2170
      else:
2171
        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
2172
                "File %s is missing from node(s) %s", filename,
2173
                utils.CommaJoin(utils.NiceSort(missing_file)))
2174

    
2175
      # See if there are multiple versions of the file
2176
      test = len(checksums) > 1
2177
      if test:
2178
        variants = ["variant %s on %s" %
2179
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2180
                    for (idx, (checksum, nodes)) in
2181
                      enumerate(sorted(checksums.items()))]
2182
      else:
2183
        variants = []
2184

    
2185
      errorif(test, cls.ECLUSTERFILECHECK, None,
2186
              "File %s found with %s different checksums (%s)",
2187
              filename, len(checksums), "; ".join(variants))
2188

    
2189
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2190
                      drbd_map):
2191
    """Verifies and the node DRBD status.
2192

2193
    @type ninfo: L{objects.Node}
2194
    @param ninfo: the node to check
2195
    @param nresult: the remote results for the node
2196
    @param instanceinfo: the dict of instances
2197
    @param drbd_helper: the configured DRBD usermode helper
2198
    @param drbd_map: the DRBD map as returned by
2199
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2200

2201
    """
2202
    node = ninfo.name
2203
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2204

    
2205
    if drbd_helper:
2206
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2207
      test = (helper_result == None)
2208
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
2209
               "no drbd usermode helper returned")
2210
      if helper_result:
2211
        status, payload = helper_result
2212
        test = not status
2213
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2214
                 "drbd usermode helper check unsuccessful: %s", payload)
2215
        test = status and (payload != drbd_helper)
2216
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2217
                 "wrong drbd usermode helper: %s", payload)
2218

    
2219
    # compute the DRBD minors
2220
    node_drbd = {}
2221
    for minor, instance in drbd_map[node].items():
2222
      test = instance not in instanceinfo
2223
      _ErrorIf(test, self.ECLUSTERCFG, None,
2224
               "ghost instance '%s' in temporary DRBD map", instance)
2225
        # ghost instance should not be running, but otherwise we
2226
        # don't give double warnings (both ghost instance and
2227
        # unallocated minor in use)
2228
      if test:
2229
        node_drbd[minor] = (instance, False)
2230
      else:
2231
        instance = instanceinfo[instance]
2232
        node_drbd[minor] = (instance.name, instance.admin_up)
2233

    
2234
    # and now check them
2235
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2236
    test = not isinstance(used_minors, (tuple, list))
2237
    _ErrorIf(test, self.ENODEDRBD, node,
2238
             "cannot parse drbd status file: %s", str(used_minors))
2239
    if test:
2240
      # we cannot check drbd status
2241
      return
2242

    
2243
    for minor, (iname, must_exist) in node_drbd.items():
2244
      test = minor not in used_minors and must_exist
2245
      _ErrorIf(test, self.ENODEDRBD, node,
2246
               "drbd minor %d of instance %s is not active", minor, iname)
2247
    for minor in used_minors:
2248
      test = minor not in node_drbd
2249
      _ErrorIf(test, self.ENODEDRBD, node,
2250
               "unallocated drbd minor %d is in use", minor)
2251

    
2252
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2253
    """Builds the node OS structures.
2254

2255
    @type ninfo: L{objects.Node}
2256
    @param ninfo: the node to check
2257
    @param nresult: the remote results for the node
2258
    @param nimg: the node image object
2259

2260
    """
2261
    node = ninfo.name
2262
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2263

    
2264
    remote_os = nresult.get(constants.NV_OSLIST, None)
2265
    test = (not isinstance(remote_os, list) or
2266
            not compat.all(isinstance(v, list) and len(v) == 7
2267
                           for v in remote_os))
2268

    
2269
    _ErrorIf(test, self.ENODEOS, node,
2270
             "node hasn't returned valid OS data")
2271

    
2272
    nimg.os_fail = test
2273

    
2274
    if test:
2275
      return
2276

    
2277
    os_dict = {}
2278

    
2279
    for (name, os_path, status, diagnose,
2280
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2281

    
2282
      if name not in os_dict:
2283
        os_dict[name] = []
2284

    
2285
      # parameters is a list of lists instead of list of tuples due to
2286
      # JSON lacking a real tuple type, fix it:
2287
      parameters = [tuple(v) for v in parameters]
2288
      os_dict[name].append((os_path, status, diagnose,
2289
                            set(variants), set(parameters), set(api_ver)))
2290

    
2291
    nimg.oslist = os_dict
2292

    
2293
  def _VerifyNodeOS(self, ninfo, nimg, base):
2294
    """Verifies the node OS list.
2295

2296
    @type ninfo: L{objects.Node}
2297
    @param ninfo: the node to check
2298
    @param nimg: the node image object
2299
    @param base: the 'template' node we match against (e.g. from the master)
2300

2301
    """
2302
    node = ninfo.name
2303
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2304

    
2305
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2306

    
2307
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2308
    for os_name, os_data in nimg.oslist.items():
2309
      assert os_data, "Empty OS status for OS %s?!" % os_name
2310
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2311
      _ErrorIf(not f_status, self.ENODEOS, node,
2312
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2313
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
2314
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2315
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2316
      # comparisons with the 'base' image
2317
      test = os_name not in base.oslist
2318
      _ErrorIf(test, self.ENODEOS, node,
2319
               "Extra OS %s not present on reference node (%s)",
2320
               os_name, base.name)
2321
      if test:
2322
        continue
2323
      assert base.oslist[os_name], "Base node has empty OS status?"
2324
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2325
      if not b_status:
2326
        # base OS is invalid, skipping
2327
        continue
2328
      for kind, a, b in [("API version", f_api, b_api),
2329
                         ("variants list", f_var, b_var),
2330
                         ("parameters", beautify_params(f_param),
2331
                          beautify_params(b_param))]:
2332
        _ErrorIf(a != b, self.ENODEOS, node,
2333
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2334
                 kind, os_name, base.name,
2335
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2336

    
2337
    # check any missing OSes
2338
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2339
    _ErrorIf(missing, self.ENODEOS, node,
2340
             "OSes present on reference node %s but missing on this node: %s",
2341
             base.name, utils.CommaJoin(missing))
2342

    
2343
  def _VerifyOob(self, ninfo, nresult):
2344
    """Verifies out of band functionality of a node.
2345

2346
    @type ninfo: L{objects.Node}
2347
    @param ninfo: the node to check
2348
    @param nresult: the remote results for the node
2349

2350
    """
2351
    node = ninfo.name
2352
    # We just have to verify the paths on master and/or master candidates
2353
    # as the oob helper is invoked on the master
2354
    if ((ninfo.master_candidate or ninfo.master_capable) and
2355
        constants.NV_OOB_PATHS in nresult):
2356
      for path_result in nresult[constants.NV_OOB_PATHS]:
2357
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
2358

    
2359
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2360
    """Verifies and updates the node volume data.
2361

2362
    This function will update a L{NodeImage}'s internal structures
2363
    with data from the remote call.
2364

2365
    @type ninfo: L{objects.Node}
2366
    @param ninfo: the node to check
2367
    @param nresult: the remote results for the node
2368
    @param nimg: the node image object
2369
    @param vg_name: the configured VG name
2370

2371
    """
2372
    node = ninfo.name
2373
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2374

    
2375
    nimg.lvm_fail = True
2376
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2377
    if vg_name is None:
2378
      pass
2379
    elif isinstance(lvdata, basestring):
2380
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
2381
               utils.SafeEncode(lvdata))
2382
    elif not isinstance(lvdata, dict):
2383
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
2384
    else:
2385
      nimg.volumes = lvdata
2386
      nimg.lvm_fail = False
2387

    
2388
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2389
    """Verifies and updates the node instance list.
2390

2391
    If the listing was successful, then updates this node's instance
2392
    list. Otherwise, it marks the RPC call as failed for the instance
2393
    list key.
2394

2395
    @type ninfo: L{objects.Node}
2396
    @param ninfo: the node to check
2397
    @param nresult: the remote results for the node
2398
    @param nimg: the node image object
2399

2400
    """
2401
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2402
    test = not isinstance(idata, list)
2403
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
2404
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
2405
    if test:
2406
      nimg.hyp_fail = True
2407
    else:
2408
      nimg.instances = idata
2409

    
2410
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2411
    """Verifies and computes a node information map
2412

2413
    @type ninfo: L{objects.Node}
2414
    @param ninfo: the node to check
2415
    @param nresult: the remote results for the node
2416
    @param nimg: the node image object
2417
    @param vg_name: the configured VG name
2418

2419
    """
2420
    node = ninfo.name
2421
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2422

    
2423
    # try to read free memory (from the hypervisor)
2424
    hv_info = nresult.get(constants.NV_HVINFO, None)
2425
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2426
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
2427
    if not test:
2428
      try:
2429
        nimg.mfree = int(hv_info["memory_free"])
2430
      except (ValueError, TypeError):
2431
        _ErrorIf(True, self.ENODERPC, node,
2432
                 "node returned invalid nodeinfo, check hypervisor")
2433

    
2434
    # FIXME: devise a free space model for file based instances as well
2435
    if vg_name is not None:
2436
      test = (constants.NV_VGLIST not in nresult or
2437
              vg_name not in nresult[constants.NV_VGLIST])
2438
      _ErrorIf(test, self.ENODELVM, node,
2439
               "node didn't return data for the volume group '%s'"
2440
               " - it is either missing or broken", vg_name)
2441
      if not test:
2442
        try:
2443
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2444
        except (ValueError, TypeError):
2445
          _ErrorIf(True, self.ENODERPC, node,
2446
                   "node returned invalid LVM info, check LVM status")
2447

    
2448
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2449
    """Gets per-disk status information for all instances.
2450

2451
    @type nodelist: list of strings
2452
    @param nodelist: Node names
2453
    @type node_image: dict of (name, L{objects.Node})
2454
    @param node_image: Node objects
2455
    @type instanceinfo: dict of (name, L{objects.Instance})
2456
    @param instanceinfo: Instance objects
2457
    @rtype: {instance: {node: [(succes, payload)]}}
2458
    @return: a dictionary of per-instance dictionaries with nodes as
2459
        keys and disk information as values; the disk information is a
2460
        list of tuples (success, payload)
2461

2462
    """
2463
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2464

    
2465
    node_disks = {}
2466
    node_disks_devonly = {}
2467
    diskless_instances = set()
2468
    diskless = constants.DT_DISKLESS
2469

    
2470
    for nname in nodelist:
2471
      node_instances = list(itertools.chain(node_image[nname].pinst,
2472
                                            node_image[nname].sinst))
2473
      diskless_instances.update(inst for inst in node_instances
2474
                                if instanceinfo[inst].disk_template == diskless)
2475
      disks = [(inst, disk)
2476
               for inst in node_instances
2477
               for disk in instanceinfo[inst].disks]
2478

    
2479
      if not disks:
2480
        # No need to collect data
2481
        continue
2482

    
2483
      node_disks[nname] = disks
2484

    
2485
      # Creating copies as SetDiskID below will modify the objects and that can
2486
      # lead to incorrect data returned from nodes
2487
      devonly = [dev.Copy() for (_, dev) in disks]
2488

    
2489
      for dev in devonly:
2490
        self.cfg.SetDiskID(dev, nname)
2491

    
2492
      node_disks_devonly[nname] = devonly
2493

    
2494
    assert len(node_disks) == len(node_disks_devonly)
2495

    
2496
    # Collect data from all nodes with disks
2497
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2498
                                                          node_disks_devonly)
2499

    
2500
    assert len(result) == len(node_disks)
2501

    
2502
    instdisk = {}
2503

    
2504
    for (nname, nres) in result.items():
2505
      disks = node_disks[nname]
2506

    
2507
      if nres.offline:
2508
        # No data from this node
2509
        data = len(disks) * [(False, "node offline")]
2510
      else:
2511
        msg = nres.fail_msg
2512
        _ErrorIf(msg, self.ENODERPC, nname,
2513
                 "while getting disk information: %s", msg)
2514
        if msg:
2515
          # No data from this node
2516
          data = len(disks) * [(False, msg)]
2517
        else:
2518
          data = []
2519
          for idx, i in enumerate(nres.payload):
2520
            if isinstance(i, (tuple, list)) and len(i) == 2:
2521
              data.append(i)
2522
            else:
2523
              logging.warning("Invalid result from node %s, entry %d: %s",
2524
                              nname, idx, i)
2525
              data.append((False, "Invalid result from the remote node"))
2526

    
2527
      for ((inst, _), status) in zip(disks, data):
2528
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2529

    
2530
    # Add empty entries for diskless instances.
2531
    for inst in diskless_instances:
2532
      assert inst not in instdisk
2533
      instdisk[inst] = {}
2534

    
2535
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2536
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2537
                      compat.all(isinstance(s, (tuple, list)) and
2538
                                 len(s) == 2 for s in statuses)
2539
                      for inst, nnames in instdisk.items()
2540
                      for nname, statuses in nnames.items())
2541
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2542

    
2543
    return instdisk
2544

    
2545
  def BuildHooksEnv(self):
2546
    """Build hooks env.
2547

2548
    Cluster-Verify hooks just ran in the post phase and their failure makes
2549
    the output be logged in the verify output and the verification to fail.
2550

2551
    """
2552
    env = {
2553
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2554
      }
2555

    
2556
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2557
               for node in self.my_node_info.values())
2558

    
2559
    return env
2560

    
2561
  def BuildHooksNodes(self):
2562
    """Build hooks nodes.
2563

2564
    """
2565
    return ([], self.my_node_names)
2566

    
2567
  def Exec(self, feedback_fn):
2568
    """Verify integrity of the node group, performing various test on nodes.
2569

2570
    """
2571
    # This method has too many local variables. pylint: disable=R0914
2572
    feedback_fn("* Verifying group '%s'" % self.group_info.name)
2573

    
2574
    if not self.my_node_names:
2575
      # empty node group
2576
      feedback_fn("* Empty node group, skipping verification")
2577
      return True
2578

    
2579
    self.bad = False
2580
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2581
    verbose = self.op.verbose
2582
    self._feedback_fn = feedback_fn
2583

    
2584
    vg_name = self.cfg.GetVGName()
2585
    drbd_helper = self.cfg.GetDRBDHelper()
2586
    cluster = self.cfg.GetClusterInfo()
2587
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2588
    hypervisors = cluster.enabled_hypervisors
2589
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2590

    
2591
    i_non_redundant = [] # Non redundant instances
2592
    i_non_a_balanced = [] # Non auto-balanced instances
2593
    n_offline = 0 # Count of offline nodes
2594
    n_drained = 0 # Count of nodes being drained
2595
    node_vol_should = {}
2596

    
2597
    # FIXME: verify OS list
2598

    
2599
    # File verification
2600
    filemap = _ComputeAncillaryFiles(cluster, False)
2601

    
2602
    # do local checksums
2603
    master_node = self.master_node = self.cfg.GetMasterNode()
2604
    master_ip = self.cfg.GetMasterIP()
2605

    
2606
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2607

    
2608
    # We will make nodes contact all nodes in their group, and one node from
2609
    # every other group.
2610
    # TODO: should it be a *random* node, different every time?
2611
    online_nodes = [node.name for node in node_data_list if not node.offline]
2612
    other_group_nodes = {}
2613

    
2614
    for name in sorted(self.all_node_info):
2615
      node = self.all_node_info[name]
2616
      if (node.group not in other_group_nodes
2617
          and node.group != self.group_uuid
2618
          and not node.offline):
2619
        other_group_nodes[node.group] = node.name
2620

    
2621
    node_verify_param = {
2622
      constants.NV_FILELIST:
2623
        utils.UniqueSequence(filename
2624
                             for files in filemap
2625
                             for filename in files),
2626
      constants.NV_NODELIST: online_nodes + other_group_nodes.values(),
2627
      constants.NV_HYPERVISOR: hypervisors,
2628
      constants.NV_HVPARAMS:
2629
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2630
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2631
                                 for node in node_data_list
2632
                                 if not node.offline],
2633
      constants.NV_INSTANCELIST: hypervisors,
2634
      constants.NV_VERSION: None,
2635
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2636
      constants.NV_NODESETUP: None,
2637
      constants.NV_TIME: None,
2638
      constants.NV_MASTERIP: (master_node, master_ip),
2639
      constants.NV_OSLIST: None,
2640
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2641
      }
2642

    
2643
    if vg_name is not None:
2644
      node_verify_param[constants.NV_VGLIST] = None
2645
      node_verify_param[constants.NV_LVLIST] = vg_name
2646
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2647
      node_verify_param[constants.NV_DRBDLIST] = None
2648

    
2649
    if drbd_helper:
2650
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2651

    
2652
    # bridge checks
2653
    # FIXME: this needs to be changed per node-group, not cluster-wide
2654
    bridges = set()
2655
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2656
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2657
      bridges.add(default_nicpp[constants.NIC_LINK])
2658
    for instance in self.my_inst_info.values():
2659
      for nic in instance.nics:
2660
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2661
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2662
          bridges.add(full_nic[constants.NIC_LINK])
2663

    
2664
    if bridges:
2665
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2666

    
2667
    # Build our expected cluster state
2668
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2669
                                                 name=node.name,
2670
                                                 vm_capable=node.vm_capable))
2671
                      for node in node_data_list)
2672

    
2673
    # Gather OOB paths
2674
    oob_paths = []
2675
    for node in self.all_node_info.values():
2676
      path = _SupportsOob(self.cfg, node)
2677
      if path and path not in oob_paths:
2678
        oob_paths.append(path)
2679

    
2680
    if oob_paths:
2681
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2682

    
2683
    for instance in self.my_inst_names:
2684
      inst_config = self.my_inst_info[instance]
2685

    
2686
      for nname in inst_config.all_nodes:
2687
        if nname not in node_image:
2688
          gnode = self.NodeImage(name=nname)
2689
          gnode.ghost = (nname not in self.all_node_info)
2690
          node_image[nname] = gnode
2691

    
2692
      inst_config.MapLVsByNode(node_vol_should)
2693

    
2694
      pnode = inst_config.primary_node
2695
      node_image[pnode].pinst.append(instance)
2696

    
2697
      for snode in inst_config.secondary_nodes:
2698
        nimg = node_image[snode]
2699
        nimg.sinst.append(instance)
2700
        if pnode not in nimg.sbp:
2701
          nimg.sbp[pnode] = []
2702
        nimg.sbp[pnode].append(instance)
2703

    
2704
    # At this point, we have the in-memory data structures complete,
2705
    # except for the runtime information, which we'll gather next
2706

    
2707
    # Due to the way our RPC system works, exact response times cannot be
2708
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2709
    # time before and after executing the request, we can at least have a time
2710
    # window.
2711
    nvinfo_starttime = time.time()
2712
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2713
                                           node_verify_param,
2714
                                           self.cfg.GetClusterName())
2715
    nvinfo_endtime = time.time()
2716

    
2717
    if self.extra_lv_nodes and vg_name is not None:
2718
      extra_lv_nvinfo = \
2719
          self.rpc.call_node_verify(self.extra_lv_nodes,
2720
                                    {constants.NV_LVLIST: vg_name},
2721
                                    self.cfg.GetClusterName())
2722
    else:
2723
      extra_lv_nvinfo = {}
2724

    
2725
    all_drbd_map = self.cfg.ComputeDRBDMap()
2726

    
2727
    feedback_fn("* Gathering disk information (%s nodes)" %
2728
                len(self.my_node_names))
2729
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2730
                                     self.my_inst_info)
2731

    
2732
    feedback_fn("* Verifying configuration file consistency")
2733

    
2734
    # If not all nodes are being checked, we need to make sure the master node
2735
    # and a non-checked vm_capable node are in the list.
2736
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2737
    if absent_nodes:
2738
      vf_nvinfo = all_nvinfo.copy()
2739
      vf_node_info = list(self.my_node_info.values())
2740
      additional_nodes = []
2741
      if master_node not in self.my_node_info:
2742
        additional_nodes.append(master_node)
2743
        vf_node_info.append(self.all_node_info[master_node])
2744
      # Add the first vm_capable node we find which is not included
2745
      for node in absent_nodes:
2746
        nodeinfo = self.all_node_info[node]
2747
        if nodeinfo.vm_capable and not nodeinfo.offline:
2748
          additional_nodes.append(node)
2749
          vf_node_info.append(self.all_node_info[node])
2750
          break
2751
      key = constants.NV_FILELIST
2752
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2753
                                                 {key: node_verify_param[key]},
2754
                                                 self.cfg.GetClusterName()))
2755
    else:
2756
      vf_nvinfo = all_nvinfo
2757
      vf_node_info = self.my_node_info.values()
2758

    
2759
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2760

    
2761
    feedback_fn("* Verifying node status")
2762

    
2763
    refos_img = None
2764

    
2765
    for node_i in node_data_list:
2766
      node = node_i.name
2767
      nimg = node_image[node]
2768

    
2769
      if node_i.offline:
2770
        if verbose:
2771
          feedback_fn("* Skipping offline node %s" % (node,))
2772
        n_offline += 1
2773
        continue
2774

    
2775
      if node == master_node:
2776
        ntype = "master"
2777
      elif node_i.master_candidate:
2778
        ntype = "master candidate"
2779
      elif node_i.drained:
2780
        ntype = "drained"
2781
        n_drained += 1
2782
      else:
2783
        ntype = "regular"
2784
      if verbose:
2785
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2786

    
2787
      msg = all_nvinfo[node].fail_msg
2788
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2789
      if msg:
2790
        nimg.rpc_fail = True
2791
        continue
2792

    
2793
      nresult = all_nvinfo[node].payload
2794

    
2795
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2796
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2797
      self._VerifyNodeNetwork(node_i, nresult)
2798
      self._VerifyOob(node_i, nresult)
2799

    
2800
      if nimg.vm_capable:
2801
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2802
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2803
                             all_drbd_map)
2804

    
2805
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2806
        self._UpdateNodeInstances(node_i, nresult, nimg)
2807
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2808
        self._UpdateNodeOS(node_i, nresult, nimg)
2809

    
2810
        if not nimg.os_fail:
2811
          if refos_img is None:
2812
            refos_img = nimg
2813
          self._VerifyNodeOS(node_i, nimg, refos_img)
2814
        self._VerifyNodeBridges(node_i, nresult, bridges)
2815

    
2816
        # Check whether all running instancies are primary for the node. (This
2817
        # can no longer be done from _VerifyInstance below, since some of the
2818
        # wrong instances could be from other node groups.)
2819
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2820

    
2821
        for inst in non_primary_inst:
2822
          test = inst in self.all_inst_info
2823
          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
2824
                   "instance should not run on node %s", node_i.name)
2825
          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
2826
                   "node is running unknown instance %s", inst)
2827

    
2828
    for node, result in extra_lv_nvinfo.items():
2829
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
2830
                              node_image[node], vg_name)
2831

    
2832
    feedback_fn("* Verifying instance status")
2833
    for instance in self.my_inst_names:
2834
      if verbose:
2835
        feedback_fn("* Verifying instance %s" % instance)
2836
      inst_config = self.my_inst_info[instance]
2837
      self._VerifyInstance(instance, inst_config, node_image,
2838
                           instdisk[instance])
2839
      inst_nodes_offline = []
2840

    
2841
      pnode = inst_config.primary_node
2842
      pnode_img = node_image[pnode]
2843
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2844
               self.ENODERPC, pnode, "instance %s, connection to"
2845
               " primary node failed", instance)
2846

    
2847
      _ErrorIf(inst_config.admin_up and pnode_img.offline,
2848
               self.EINSTANCEBADNODE, instance,
2849
               "instance is marked as running and lives on offline node %s",
2850
               inst_config.primary_node)
2851

    
2852
      # If the instance is non-redundant we cannot survive losing its primary
2853
      # node, so we are not N+1 compliant. On the other hand we have no disk
2854
      # templates with more than one secondary so that situation is not well
2855
      # supported either.
2856
      # FIXME: does not support file-backed instances
2857
      if not inst_config.secondary_nodes:
2858
        i_non_redundant.append(instance)
2859

    
2860
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2861
               instance, "instance has multiple secondary nodes: %s",
2862
               utils.CommaJoin(inst_config.secondary_nodes),
2863
               code=self.ETYPE_WARNING)
2864

    
2865
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2866
        pnode = inst_config.primary_node
2867
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2868
        instance_groups = {}
2869

    
2870
        for node in instance_nodes:
2871
          instance_groups.setdefault(self.all_node_info[node].group,
2872
                                     []).append(node)
2873

    
2874
        pretty_list = [
2875
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2876
          # Sort so that we always list the primary node first.
2877
          for group, nodes in sorted(instance_groups.items(),
2878
                                     key=lambda (_, nodes): pnode in nodes,
2879
                                     reverse=True)]
2880

    
2881
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2882
                      instance, "instance has primary and secondary nodes in"
2883
                      " different groups: %s", utils.CommaJoin(pretty_list),
2884
                      code=self.ETYPE_WARNING)
2885

    
2886
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2887
        i_non_a_balanced.append(instance)
2888

    
2889
      for snode in inst_config.secondary_nodes:
2890
        s_img = node_image[snode]
2891
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2892
                 "instance %s, connection to secondary node failed", instance)
2893

    
2894
        if s_img.offline:
2895
          inst_nodes_offline.append(snode)
2896

    
2897
      # warn that the instance lives on offline nodes
2898
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2899
               "instance has offline secondary node(s) %s",
2900
               utils.CommaJoin(inst_nodes_offline))
2901
      # ... or ghost/non-vm_capable nodes
2902
      for node in inst_config.all_nodes:
2903
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2904
                 "instance lives on ghost node %s", node)
2905
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2906
                 instance, "instance lives on non-vm_capable node %s", node)
2907

    
2908
    feedback_fn("* Verifying orphan volumes")
2909
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2910

    
2911
    # We will get spurious "unknown volume" warnings if any node of this group
2912
    # is secondary for an instance whose primary is in another group. To avoid
2913
    # them, we find these instances and add their volumes to node_vol_should.
2914
    for inst in self.all_inst_info.values():
2915
      for secondary in inst.secondary_nodes:
2916
        if (secondary in self.my_node_info
2917
            and inst.name not in self.my_inst_info):
2918
          inst.MapLVsByNode(node_vol_should)
2919
          break
2920

    
2921
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2922

    
2923
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2924
      feedback_fn("* Verifying N+1 Memory redundancy")
2925
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
2926

    
2927
    feedback_fn("* Other Notes")
2928
    if i_non_redundant:
2929
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2930
                  % len(i_non_redundant))
2931

    
2932
    if i_non_a_balanced:
2933
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2934
                  % len(i_non_a_balanced))
2935

    
2936
    if n_offline:
2937
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2938

    
2939
    if n_drained:
2940
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2941

    
2942
    return not self.bad
2943

    
2944
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2945
    """Analyze the post-hooks' result
2946

2947
    This method analyses the hook result, handles it, and sends some
2948
    nicely-formatted feedback back to the user.
2949

2950
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
2951
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
2952
    @param hooks_results: the results of the multi-node hooks rpc call
2953
    @param feedback_fn: function used send feedback back to the caller
2954
    @param lu_result: previous Exec result
2955
    @return: the new Exec result, based on the previous result
2956
        and hook results
2957

2958
    """
2959
    # We only really run POST phase hooks, only for non-empty groups,
2960
    # and are only interested in their results
2961
    if not self.my_node_names:
2962
      # empty node group
2963
      pass
2964
    elif phase == constants.HOOKS_PHASE_POST:
2965
      # Used to change hooks' output to proper indentation
2966
      feedback_fn("* Hooks Results")
2967
      assert hooks_results, "invalid result from hooks"
2968

    
2969
      for node_name in hooks_results:
2970
        res = hooks_results[node_name]
2971
        msg = res.fail_msg
2972
        test = msg and not res.offline
2973
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
2974
                      "Communication failure in hooks execution: %s", msg)
2975
        if res.offline or msg:
2976
          # No need to investigate payload if node is offline or gave an error.
2977
          # override manually lu_result here as _ErrorIf only
2978
          # overrides self.bad
2979
          lu_result = 1
2980
          continue
2981
        for script, hkr, output in res.payload:
2982
          test = hkr == constants.HKR_FAIL
2983
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
2984
                        "Script %s failed, output:", script)
2985
          if test:
2986
            output = self._HOOKS_INDENT_RE.sub("      ", output)
2987
            feedback_fn("%s" % output)
2988
            lu_result = 0
2989

    
2990
    return lu_result
2991

    
2992

    
2993
class LUClusterVerifyDisks(NoHooksLU):
2994
  """Verifies the cluster disks status.
2995

2996
  """
2997
  REQ_BGL = False
2998

    
2999
  def ExpandNames(self):
3000
    self.share_locks = _ShareAll()
3001
    self.needed_locks = {
3002
      locking.LEVEL_NODEGROUP: locking.ALL_SET,
3003
      }
3004

    
3005
  def Exec(self, feedback_fn):
3006
    group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3007

    
3008
    # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3009
    return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3010
                           for group in group_names])
3011

    
3012

    
3013
class LUGroupVerifyDisks(NoHooksLU):
3014
  """Verifies the status of all disks in a node group.
3015

3016
  """
3017
  REQ_BGL = False
3018

    
3019
  def ExpandNames(self):
3020
    # Raises errors.OpPrereqError on its own if group can't be found
3021
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3022

    
3023
    self.share_locks = _ShareAll()
3024
    self.needed_locks = {
3025
      locking.LEVEL_INSTANCE: [],
3026
      locking.LEVEL_NODEGROUP: [],
3027
      locking.LEVEL_NODE: [],
3028
      }
3029

    
3030
  def DeclareLocks(self, level):
3031
    if level == locking.LEVEL_INSTANCE:
3032
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
3033

    
3034
      # Lock instances optimistically, needs verification once node and group
3035
      # locks have been acquired
3036
      self.needed_locks[locking.LEVEL_INSTANCE] = \
3037
        self.cfg.GetNodeGroupInstances(self.group_uuid)
3038

    
3039
    elif level == locking.LEVEL_NODEGROUP:
3040
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3041

    
3042
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
3043
        set([self.group_uuid] +
3044
            # Lock all groups used by instances optimistically; this requires
3045
            # going via the node before it's locked, requiring verification
3046
            # later on
3047
            [group_uuid
3048
             for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3049
             for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3050

    
3051
    elif level == locking.LEVEL_NODE:
3052
      # This will only lock the nodes in the group to be verified which contain
3053
      # actual instances
3054
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3055
      self._LockInstancesNodes()
3056

    
3057
      # Lock all nodes in group to be verified
3058
      assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3059
      member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3060
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3061

    
3062
  def CheckPrereq(self):
3063
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3064
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3065
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3066

    
3067
    assert self.group_uuid in owned_groups
3068

    
3069
    # Check if locked instances are still correct
3070
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3071

    
3072
    # Get instance information
3073
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3074

    
3075
    # Check if node groups for locked instances are still correct
3076
    for (instance_name, inst) in self.instances.items():
3077
      assert owned_nodes.issuperset(inst.all_nodes), \
3078
        "Instance %s's nodes changed while we kept the lock" % instance_name
3079

    
3080
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
3081
                                             owned_groups)
3082

    
3083
      assert self.group_uuid in inst_groups, \
3084
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
3085

    
3086
  def Exec(self, feedback_fn):
3087
    """Verify integrity of cluster disks.
3088

3089
    @rtype: tuple of three items
3090
    @return: a tuple of (dict of node-to-node_error, list of instances
3091
        which need activate-disks, dict of instance: (node, volume) for
3092
        missing volumes
3093

3094
    """
3095
    res_nodes = {}
3096
    res_instances = set()
3097
    res_missing = {}
3098

    
3099
    nv_dict = _MapInstanceDisksToNodes([inst
3100
                                        for inst in self.instances.values()
3101
                                        if inst.admin_up])
3102

    
3103
    if nv_dict:
3104
      nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3105
                             set(self.cfg.GetVmCapableNodeList()))
3106

    
3107
      node_lvs = self.rpc.call_lv_list(nodes, [])
3108

    
3109
      for (node, node_res) in node_lvs.items():
3110
        if node_res.offline:
3111
          continue
3112

    
3113
        msg = node_res.fail_msg
3114
        if msg:
3115
          logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3116
          res_nodes[node] = msg
3117
          continue
3118

    
3119
        for lv_name, (_, _, lv_online) in node_res.payload.items():
3120
          inst = nv_dict.pop((node, lv_name), None)
3121
          if not (lv_online or inst is None):
3122
            res_instances.add(inst)
3123

    
3124
      # any leftover items in nv_dict are missing LVs, let's arrange the data
3125
      # better
3126
      for key, inst in nv_dict.iteritems():
3127
        res_missing.setdefault(inst, []).append(key)
3128

    
3129
    return (res_nodes, list(res_instances), res_missing)
3130

    
3131

    
3132
class LUClusterRepairDiskSizes(NoHooksLU):
3133
  """Verifies the cluster disks sizes.
3134

3135
  """
3136
  REQ_BGL = False
3137

    
3138
  def ExpandNames(self):
3139
    if self.op.instances:
3140
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
3141
      self.needed_locks = {
3142
        locking.LEVEL_NODE: [],
3143
        locking.LEVEL_INSTANCE: self.wanted_names,
3144
        }
3145
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3146
    else:
3147
      self.wanted_names = None
3148
      self.needed_locks = {
3149
        locking.LEVEL_NODE: locking.ALL_SET,
3150
        locking.LEVEL_INSTANCE: locking.ALL_SET,
3151
        }
3152
    self.share_locks = _ShareAll()
3153

    
3154
  def DeclareLocks(self, level):
3155
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
3156
      self._LockInstancesNodes(primary_only=True)
3157

    
3158
  def CheckPrereq(self):
3159
    """Check prerequisites.
3160

3161
    This only checks the optional instance list against the existing names.
3162

3163
    """
3164
    if self.wanted_names is None:
3165
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3166

    
3167
    self.wanted_instances = \
3168
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3169

    
3170
  def _EnsureChildSizes(self, disk):
3171
    """Ensure children of the disk have the needed disk size.
3172

3173
    This is valid mainly for DRBD8 and fixes an issue where the
3174
    children have smaller disk size.
3175

3176
    @param disk: an L{ganeti.objects.Disk} object
3177

3178
    """
3179
    if disk.dev_type == constants.LD_DRBD8:
3180
      assert disk.children, "Empty children for DRBD8?"
3181
      fchild = disk.children[0]
3182
      mismatch = fchild.size < disk.size
3183
      if mismatch:
3184
        self.LogInfo("Child disk has size %d, parent %d, fixing",
3185
                     fchild.size, disk.size)
3186
        fchild.size = disk.size
3187

    
3188
      # and we recurse on this child only, not on the metadev
3189
      return self._EnsureChildSizes(fchild) or mismatch
3190
    else:
3191
      return False
3192

    
3193
  def Exec(self, feedback_fn):
3194
    """Verify the size of cluster disks.
3195

3196
    """
3197
    # TODO: check child disks too
3198
    # TODO: check differences in size between primary/secondary nodes
3199
    per_node_disks = {}
3200
    for instance in self.wanted_instances:
3201
      pnode = instance.primary_node
3202
      if pnode not in per_node_disks:
3203
        per_node_disks[pnode] = []
3204
      for idx, disk in enumerate(instance.disks):
3205
        per_node_disks[pnode].append((instance, idx, disk))
3206

    
3207
    changed = []
3208
    for node, dskl in per_node_disks.items():
3209
      newl = [v[2].Copy() for v in dskl]
3210
      for dsk in newl:
3211
        self.cfg.SetDiskID(dsk, node)
3212
      result = self.rpc.call_blockdev_getsize(node, newl)
3213
      if result.fail_msg:
3214
        self.LogWarning("Failure in blockdev_getsize call to node"
3215
                        " %s, ignoring", node)
3216
        continue
3217
      if len(result.payload) != len(dskl):
3218
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
3219
                        " result.payload=%s", node, len(dskl), result.payload)
3220
        self.LogWarning("Invalid result from node %s, ignoring node results",
3221
                        node)
3222
        continue
3223
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
3224
        if size is None:
3225
          self.LogWarning("Disk %d of instance %s did not return size"
3226
                          " information, ignoring", idx, instance.name)
3227
          continue
3228
        if not isinstance(size, (int, long)):
3229
          self.LogWarning("Disk %d of instance %s did not return valid"
3230
                          " size information, ignoring", idx, instance.name)
3231
          continue
3232
        size = size >> 20
3233
        if size != disk.size:
3234
          self.LogInfo("Disk %d of instance %s has mismatched size,"
3235
                       " correcting: recorded %d, actual %d", idx,
3236
                       instance.name, disk.size, size)
3237
          disk.size = size
3238
          self.cfg.Update(instance, feedback_fn)
3239
          changed.append((instance.name, idx, size))
3240
        if self._EnsureChildSizes(disk):
3241
          self.cfg.Update(instance, feedback_fn)
3242
          changed.append((instance.name, idx, disk.size))
3243
    return changed
3244

    
3245

    
3246
class LUClusterRename(LogicalUnit):
3247
  """Rename the cluster.
3248

3249
  """
3250
  HPATH = "cluster-rename"
3251
  HTYPE = constants.HTYPE_CLUSTER
3252

    
3253
  def BuildHooksEnv(self):
3254
    """Build hooks env.
3255

3256
    """
3257
    return {
3258
      "OP_TARGET": self.cfg.GetClusterName(),
3259
      "NEW_NAME": self.op.name,
3260
      }
3261

    
3262
  def BuildHooksNodes(self):
3263
    """Build hooks nodes.
3264

3265
    """
3266
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3267

    
3268
  def CheckPrereq(self):
3269
    """Verify that the passed name is a valid one.
3270

3271
    """
3272
    hostname = netutils.GetHostname(name=self.op.name,
3273
                                    family=self.cfg.GetPrimaryIPFamily())
3274

    
3275
    new_name = hostname.name
3276
    self.ip = new_ip = hostname.ip
3277
    old_name = self.cfg.GetClusterName()
3278
    old_ip = self.cfg.GetMasterIP()
3279
    if new_name == old_name and new_ip == old_ip:
3280
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3281
                                 " cluster has changed",
3282
                                 errors.ECODE_INVAL)
3283
    if new_ip != old_ip:
3284
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3285
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3286
                                   " reachable on the network" %
3287
                                   new_ip, errors.ECODE_NOTUNIQUE)
3288

    
3289
    self.op.name = new_name
3290

    
3291
  def Exec(self, feedback_fn):
3292
    """Rename the cluster.
3293

3294
    """
3295
    clustername = self.op.name
3296
    ip = self.ip
3297

    
3298
    # shutdown the master IP
3299
    master = self.cfg.GetMasterNode()
3300
    result = self.rpc.call_node_stop_master(master, False)
3301
    result.Raise("Could not disable the master role")
3302

    
3303
    try:
3304
      cluster = self.cfg.GetClusterInfo()
3305
      cluster.cluster_name = clustername
3306
      cluster.master_ip = ip
3307
      self.cfg.Update(cluster, feedback_fn)
3308

    
3309
      # update the known hosts file
3310
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3311
      node_list = self.cfg.GetOnlineNodeList()
3312
      try:
3313
        node_list.remove(master)
3314
      except ValueError:
3315
        pass
3316
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3317
    finally:
3318
      result = self.rpc.call_node_start_master(master, False, False)
3319
      msg = result.fail_msg
3320
      if msg:
3321
        self.LogWarning("Could not re-enable the master role on"
3322
                        " the master, please restart manually: %s", msg)
3323

    
3324
    return clustername
3325

    
3326

    
3327
class LUClusterSetParams(LogicalUnit):
3328
  """Change the parameters of the cluster.
3329

3330
  """
3331
  HPATH = "cluster-modify"
3332
  HTYPE = constants.HTYPE_CLUSTER
3333
  REQ_BGL = False
3334

    
3335
  def CheckArguments(self):
3336
    """Check parameters
3337

3338
    """
3339
    if self.op.uid_pool:
3340
      uidpool.CheckUidPool(self.op.uid_pool)
3341

    
3342
    if self.op.add_uids:
3343
      uidpool.CheckUidPool(self.op.add_uids)
3344

    
3345
    if self.op.remove_uids:
3346
      uidpool.CheckUidPool(self.op.remove_uids)
3347

    
3348
  def ExpandNames(self):
3349
    # FIXME: in the future maybe other cluster params won't require checking on
3350
    # all nodes to be modified.
3351
    self.needed_locks = {
3352
      locking.LEVEL_NODE: locking.ALL_SET,
3353
    }
3354
    self.share_locks[locking.LEVEL_NODE] = 1
3355

    
3356
  def BuildHooksEnv(self):
3357
    """Build hooks env.
3358

3359
    """
3360
    return {
3361
      "OP_TARGET": self.cfg.GetClusterName(),
3362
      "NEW_VG_NAME": self.op.vg_name,
3363
      }
3364

    
3365
  def BuildHooksNodes(self):
3366
    """Build hooks nodes.
3367

3368
    """
3369
    mn = self.cfg.GetMasterNode()
3370
    return ([mn], [mn])
3371

    
3372
  def CheckPrereq(self):
3373
    """Check prerequisites.
3374

3375
    This checks whether the given params don't conflict and
3376
    if the given volume group is valid.
3377

3378
    """
3379
    if self.op.vg_name is not None and not self.op.vg_name:
3380
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3381
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3382
                                   " instances exist", errors.ECODE_INVAL)
3383

    
3384
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3385
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3386
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3387
                                   " drbd-based instances exist",
3388
                                   errors.ECODE_INVAL)
3389

    
3390
    node_list = self.owned_locks(locking.LEVEL_NODE)
3391

    
3392
    # if vg_name not None, checks given volume group on all nodes
3393
    if self.op.vg_name:
3394
      vglist = self.rpc.call_vg_list(node_list)
3395
      for node in node_list:
3396
        msg = vglist[node].fail_msg
3397
        if msg:
3398
          # ignoring down node
3399
          self.LogWarning("Error while gathering data on node %s"
3400
                          " (ignoring node): %s", node, msg)
3401
          continue
3402
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3403
                                              self.op.vg_name,
3404
                                              constants.MIN_VG_SIZE)
3405
        if vgstatus:
3406
          raise errors.OpPrereqError("Error on node '%s': %s" %
3407
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3408

    
3409
    if self.op.drbd_helper:
3410
      # checks given drbd helper on all nodes
3411
      helpers = self.rpc.call_drbd_helper(node_list)
3412
      for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3413
        if ninfo.offline:
3414
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3415
          continue
3416
        msg = helpers[node].fail_msg
3417
        if msg:
3418
          raise errors.OpPrereqError("Error checking drbd helper on node"
3419
                                     " '%s': %s" % (node, msg),
3420
                                     errors.ECODE_ENVIRON)
3421
        node_helper = helpers[node].payload
3422
        if node_helper != self.op.drbd_helper:
3423
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3424
                                     (node, node_helper), errors.ECODE_ENVIRON)
3425

    
3426
    self.cluster = cluster = self.cfg.GetClusterInfo()
3427
    # validate params changes
3428
    if self.op.beparams:
3429
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3430
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3431

    
3432
    if self.op.ndparams:
3433
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3434
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3435

    
3436
      # TODO: we need a more general way to handle resetting
3437
      # cluster-level parameters to default values
3438
      if self.new_ndparams["oob_program"] == "":
3439
        self.new_ndparams["oob_program"] = \
3440
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3441

    
3442
    if self.op.nicparams:
3443
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3444
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3445
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3446
      nic_errors = []
3447

    
3448
      # check all instances for consistency
3449
      for instance in self.cfg.GetAllInstancesInfo().values():
3450
        for nic_idx, nic in enumerate(instance.nics):
3451
          params_copy = copy.deepcopy(nic.nicparams)
3452
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3453

    
3454
          # check parameter syntax
3455
          try:
3456
            objects.NIC.CheckParameterSyntax(params_filled)
3457
          except errors.ConfigurationError, err:
3458
            nic_errors.append("Instance %s, nic/%d: %s" %
3459
                              (instance.name, nic_idx, err))
3460

    
3461
          # if we're moving instances to routed, check that they have an ip
3462
          target_mode = params_filled[constants.NIC_MODE]
3463
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3464
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3465
                              " address" % (instance.name, nic_idx))
3466
      if nic_errors:
3467
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3468
                                   "\n".join(nic_errors))
3469

    
3470
    # hypervisor list/parameters
3471
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3472
    if self.op.hvparams:
3473
      for hv_name, hv_dict in self.op.hvparams.items():
3474
        if hv_name not in self.new_hvparams:
3475
          self.new_hvparams[hv_name] = hv_dict
3476
        else:
3477
          self.new_hvparams[hv_name].update(hv_dict)
3478

    
3479
    # os hypervisor parameters
3480
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3481
    if self.op.os_hvp:
3482
      for os_name, hvs in self.op.os_hvp.items():
3483
        if os_name not in self.new_os_hvp:
3484
          self.new_os_hvp[os_name] = hvs
3485
        else:
3486
          for hv_name, hv_dict in hvs.items():
3487
            if hv_name not in self.new_os_hvp[os_name]:
3488
              self.new_os_hvp[os_name][hv_name] = hv_dict
3489
            else:
3490
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3491

    
3492
    # os parameters
3493
    self.new_osp = objects.FillDict(cluster.osparams, {})
3494
    if self.op.osparams:
3495
      for os_name, osp in self.op.osparams.items():
3496
        if os_name not in self.new_osp:
3497
          self.new_osp[os_name] = {}
3498

    
3499
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3500
                                                  use_none=True)
3501

    
3502
        if not self.new_osp[os_name]:
3503
          # we removed all parameters
3504
          del self.new_osp[os_name]
3505
        else:
3506
          # check the parameter validity (remote check)
3507
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3508
                         os_name, self.new_osp[os_name])
3509

    
3510
    # changes to the hypervisor list
3511
    if self.op.enabled_hypervisors is not None:
3512
      self.hv_list = self.op.enabled_hypervisors
3513
      for hv in self.hv_list:
3514
        # if the hypervisor doesn't already exist in the cluster
3515
        # hvparams, we initialize it to empty, and then (in both
3516
        # cases) we make sure to fill the defaults, as we might not
3517
        # have a complete defaults list if the hypervisor wasn't
3518
        # enabled before
3519
        if hv not in new_hvp:
3520
          new_hvp[hv] = {}
3521
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3522
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3523
    else:
3524
      self.hv_list = cluster.enabled_hypervisors
3525

    
3526
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3527
      # either the enabled list has changed, or the parameters have, validate
3528
      for hv_name, hv_params in self.new_hvparams.items():
3529
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3530
            (self.op.enabled_hypervisors and
3531
             hv_name in self.op.enabled_hypervisors)):
3532
          # either this is a new hypervisor, or its parameters have changed
3533
          hv_class = hypervisor.GetHypervisor(hv_name)
3534
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3535
          hv_class.CheckParameterSyntax(hv_params)
3536
          _CheckHVParams(self, node_list, hv_name, hv_params)
3537

    
3538
    if self.op.os_hvp:
3539
      # no need to check any newly-enabled hypervisors, since the
3540
      # defaults have already been checked in the above code-block
3541
      for os_name, os_hvp in self.new_os_hvp.items():
3542
        for hv_name, hv_params in os_hvp.items():
3543
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3544
          # we need to fill in the new os_hvp on top of the actual hv_p
3545
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3546
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3547
          hv_class = hypervisor.GetHypervisor(hv_name)
3548
          hv_class.CheckParameterSyntax(new_osp)
3549
          _CheckHVParams(self, node_list, hv_name, new_osp)
3550

    
3551
    if self.op.default_iallocator:
3552
      alloc_script = utils.FindFile(self.op.default_iallocator,
3553
                                    constants.IALLOCATOR_SEARCH_PATH,
3554
                                    os.path.isfile)
3555
      if alloc_script is None:
3556
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3557
                                   " specified" % self.op.default_iallocator,
3558
                                   errors.ECODE_INVAL)
3559

    
3560
  def Exec(self, feedback_fn):
3561
    """Change the parameters of the cluster.
3562

3563
    """
3564
    if self.op.vg_name is not None:
3565
      new_volume = self.op.vg_name
3566
      if not new_volume:
3567
        new_volume = None
3568
      if new_volume != self.cfg.GetVGName():
3569
        self.cfg.SetVGName(new_volume)
3570
      else:
3571
        feedback_fn("Cluster LVM configuration already in desired"
3572
                    " state, not changing")
3573
    if self.op.drbd_helper is not None:
3574
      new_helper = self.op.drbd_helper
3575
      if not new_helper:
3576
        new_helper = None
3577
      if new_helper != self.cfg.GetDRBDHelper():
3578
        self.cfg.SetDRBDHelper(new_helper)
3579
      else:
3580
        feedback_fn("Cluster DRBD helper already in desired state,"
3581
                    " not changing")
3582
    if self.op.hvparams:
3583
      self.cluster.hvparams = self.new_hvparams
3584
    if self.op.os_hvp:
3585
      self.cluster.os_hvp = self.new_os_hvp
3586
    if self.op.enabled_hypervisors is not None:
3587
      self.cluster.hvparams = self.new_hvparams
3588
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3589
    if self.op.beparams:
3590
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3591
    if self.op.nicparams:
3592
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3593
    if self.op.osparams:
3594
      self.cluster.osparams = self.new_osp
3595
    if self.op.ndparams:
3596
      self.cluster.ndparams = self.new_ndparams
3597

    
3598
    if self.op.candidate_pool_size is not None:
3599
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3600
      # we need to update the pool size here, otherwise the save will fail
3601
      _AdjustCandidatePool(self, [])
3602

    
3603
    if self.op.maintain_node_health is not None:
3604
      self.cluster.maintain_node_health = self.op.maintain_node_health
3605

    
3606
    if self.op.prealloc_wipe_disks is not None:
3607
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3608

    
3609
    if self.op.add_uids is not None:
3610
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3611

    
3612
    if self.op.remove_uids is not None:
3613
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3614

    
3615
    if self.op.uid_pool is not None:
3616
      self.cluster.uid_pool = self.op.uid_pool
3617

    
3618
    if self.op.default_iallocator is not None:
3619
      self.cluster.default_iallocator = self.op.default_iallocator
3620

    
3621
    if self.op.reserved_lvs is not None:
3622
      self.cluster.reserved_lvs = self.op.reserved_lvs
3623

    
3624
    def helper_os(aname, mods, desc):
3625
      desc += " OS list"
3626
      lst = getattr(self.cluster, aname)
3627
      for key, val in mods:
3628
        if key == constants.DDM_ADD:
3629
          if val in lst:
3630
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3631
          else:
3632
            lst.append(val)
3633
        elif key == constants.DDM_REMOVE:
3634
          if val in lst:
3635
            lst.remove(val)
3636
          else:
3637
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3638
        else:
3639
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3640

    
3641
    if self.op.hidden_os:
3642
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3643

    
3644
    if self.op.blacklisted_os:
3645
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3646

    
3647
    if self.op.master_netdev:
3648
      master = self.cfg.GetMasterNode()
3649
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3650
                  self.cluster.master_netdev)
3651
      result = self.rpc.call_node_stop_master(master, False)
3652
      result.Raise("Could not disable the master ip")
3653
      feedback_fn("Changing master_netdev from %s to %s" %
3654
                  (self.cluster.master_netdev, self.op.master_netdev))
3655
      self.cluster.master_netdev = self.op.master_netdev
3656

    
3657
    self.cfg.Update(self.cluster, feedback_fn)
3658

    
3659
    if self.op.master_netdev:
3660
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3661
                  self.op.master_netdev)
3662
      result = self.rpc.call_node_start_master(master, False, False)
3663
      if result.fail_msg:
3664
        self.LogWarning("Could not re-enable the master ip on"
3665
                        " the master, please restart manually: %s",
3666
                        result.fail_msg)
3667

    
3668

    
3669
def _UploadHelper(lu, nodes, fname):
3670
  """Helper for uploading a file and showing warnings.
3671

3672
  """
3673
  if os.path.exists(fname):
3674
    result = lu.rpc.call_upload_file(nodes, fname)
3675
    for to_node, to_result in result.items():
3676
      msg = to_result.fail_msg
3677
      if msg:
3678
        msg = ("Copy of file %s to node %s failed: %s" %
3679
               (fname, to_node, msg))
3680
        lu.proc.LogWarning(msg)
3681

    
3682

    
3683
def _ComputeAncillaryFiles(cluster, redist):
3684
  """Compute files external to Ganeti which need to be consistent.
3685

3686
  @type redist: boolean
3687
  @param redist: Whether to include files which need to be redistributed
3688

3689
  """
3690
  # Compute files for all nodes
3691
  files_all = set([
3692
    constants.SSH_KNOWN_HOSTS_FILE,
3693
    constants.CONFD_HMAC_KEY,
3694
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3695
    ])
3696

    
3697
  if not redist:
3698
    files_all.update(constants.ALL_CERT_FILES)
3699
    files_all.update(ssconf.SimpleStore().GetFileList())
3700

    
3701
  if cluster.modify_etc_hosts:
3702
    files_all.add(constants.ETC_HOSTS)
3703

    
3704
  # Files which must either exist on all nodes or on none
3705
  files_all_opt = set([
3706
    constants.RAPI_USERS_FILE,
3707
    ])
3708

    
3709
  # Files which should only be on master candidates
3710
  files_mc = set()
3711
  if not redist:
3712
    files_mc.add(constants.CLUSTER_CONF_FILE)
3713

    
3714
  # Files which should only be on VM-capable nodes
3715
  files_vm = set(filename
3716
    for hv_name in cluster.enabled_hypervisors
3717
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
3718

    
3719
  # Filenames must be unique
3720
  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
3721
          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
3722
         "Found file listed in more than one file list"
3723

    
3724
  return (files_all, files_all_opt, files_mc, files_vm)
3725

    
3726

    
3727
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3728
  """Distribute additional files which are part of the cluster configuration.
3729

3730
  ConfigWriter takes care of distributing the config and ssconf files, but
3731
  there are more files which should be distributed to all nodes. This function
3732
  makes sure those are copied.
3733

3734
  @param lu: calling logical unit
3735
  @param additional_nodes: list of nodes not in the config to distribute to
3736
  @type additional_vm: boolean
3737
  @param additional_vm: whether the additional nodes are vm-capable or not
3738

3739
  """
3740
  # Gather target nodes
3741
  cluster = lu.cfg.GetClusterInfo()
3742
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3743

    
3744
  online_nodes = lu.cfg.GetOnlineNodeList()
3745
  vm_nodes = lu.cfg.GetVmCapableNodeList()
3746

    
3747
  if additional_nodes is not None:
3748
    online_nodes.extend(additional_nodes)
3749
    if additional_vm:
3750
      vm_nodes.extend(additional_nodes)
3751

    
3752
  # Never distribute to master node
3753
  for nodelist in [online_nodes, vm_nodes]:
3754
    if master_info.name in nodelist:
3755
      nodelist.remove(master_info.name)
3756

    
3757
  # Gather file lists
3758
  (files_all, files_all_opt, files_mc, files_vm) = \
3759
    _ComputeAncillaryFiles(cluster, True)
3760

    
3761
  # Never re-distribute configuration file from here
3762
  assert not (constants.CLUSTER_CONF_FILE in files_all or
3763
              constants.CLUSTER_CONF_FILE in files_vm)
3764
  assert not files_mc, "Master candidates not handled in this function"
3765

    
3766
  filemap = [
3767
    (online_nodes, files_all),
3768
    (online_nodes, files_all_opt),
3769
    (vm_nodes, files_vm),
3770
    ]
3771

    
3772
  # Upload the files
3773
  for (node_list, files) in filemap:
3774
    for fname in files:
3775
      _UploadHelper(lu, node_list, fname)
3776

    
3777

    
3778
class LUClusterRedistConf(NoHooksLU):
3779
  """Force the redistribution of cluster configuration.
3780

3781
  This is a very simple LU.
3782

3783
  """
3784
  REQ_BGL = False
3785

    
3786
  def ExpandNames(self):
3787
    self.needed_locks = {
3788
      locking.LEVEL_NODE: locking.ALL_SET,
3789
    }
3790
    self.share_locks[locking.LEVEL_NODE] = 1
3791

    
3792
  def Exec(self, feedback_fn):
3793
    """Redistribute the configuration.
3794

3795
    """
3796
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3797
    _RedistributeAncillaryFiles(self)
3798

    
3799

    
3800
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3801
  """Sleep and poll for an instance's disk to sync.
3802

3803
  """
3804
  if not instance.disks or disks is not None and not disks:
3805
    return True
3806

    
3807
  disks = _ExpandCheckDisks(instance, disks)
3808

    
3809
  if not oneshot:
3810
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3811

    
3812
  node = instance.primary_node
3813

    
3814
  for dev in disks:
3815
    lu.cfg.SetDiskID(dev, node)
3816

    
3817
  # TODO: Convert to utils.Retry
3818

    
3819
  retries = 0
3820
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3821
  while True:
3822
    max_time = 0
3823
    done = True
3824
    cumul_degraded = False
3825
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3826
    msg = rstats.fail_msg
3827
    if msg:
3828
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3829
      retries += 1
3830
      if retries >= 10:
3831
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3832
                                 " aborting." % node)
3833
      time.sleep(6)
3834
      continue
3835
    rstats = rstats.payload
3836
    retries = 0
3837
    for i, mstat in enumerate(rstats):
3838
      if mstat is None:
3839
        lu.LogWarning("Can't compute data for node %s/%s",
3840
                           node, disks[i].iv_name)
3841
        continue
3842

    
3843
      cumul_degraded = (cumul_degraded or
3844
                        (mstat.is_degraded and mstat.sync_percent is None))
3845
      if mstat.sync_percent is not None:
3846
        done = False
3847
        if mstat.estimated_time is not None:
3848
          rem_time = ("%s remaining (estimated)" %
3849
                      utils.FormatSeconds(mstat.estimated_time))
3850
          max_time = mstat.estimated_time
3851
        else:
3852
          rem_time = "no time estimate"
3853
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3854
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3855

    
3856
    # if we're done but degraded, let's do a few small retries, to
3857
    # make sure we see a stable and not transient situation; therefore
3858
    # we force restart of the loop
3859
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3860
      logging.info("Degraded disks found, %d retries left", degr_retries)
3861
      degr_retries -= 1
3862
      time.sleep(1)
3863
      continue
3864

    
3865
    if done or oneshot:
3866
      break
3867

    
3868
    time.sleep(min(60, max_time))
3869

    
3870
  if done:
3871
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3872
  return not cumul_degraded
3873

    
3874

    
3875
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3876
  """Check that mirrors are not degraded.
3877

3878
  The ldisk parameter, if True, will change the test from the
3879
  is_degraded attribute (which represents overall non-ok status for
3880
  the device(s)) to the ldisk (representing the local storage status).
3881

3882
  """
3883
  lu.cfg.SetDiskID(dev, node)
3884

    
3885
  result = True
3886

    
3887
  if on_primary or dev.AssembleOnSecondary():
3888
    rstats = lu.rpc.call_blockdev_find(node, dev)
3889
    msg = rstats.fail_msg
3890
    if msg:
3891
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3892
      result = False
3893
    elif not rstats.payload:
3894
      lu.LogWarning("Can't find disk on node %s", node)
3895
      result = False
3896
    else:
3897
      if ldisk:
3898
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3899
      else:
3900
        result = result and not rstats.payload.is_degraded
3901

    
3902
  if dev.children:
3903
    for child in dev.children:
3904
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3905

    
3906
  return result
3907

    
3908

    
3909
class LUOobCommand(NoHooksLU):
3910
  """Logical unit for OOB handling.
3911

3912
  """
3913
  REG_BGL = False
3914
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
3915

    
3916
  def ExpandNames(self):
3917
    """Gather locks we need.
3918

3919
    """
3920
    if self.op.node_names:
3921
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
3922
      lock_names = self.op.node_names
3923
    else:
3924
      lock_names = locking.ALL_SET
3925

    
3926
    self.needed_locks = {
3927
      locking.LEVEL_NODE: lock_names,
3928
      }
3929

    
3930
  def CheckPrereq(self):
3931
    """Check prerequisites.
3932

3933
    This checks:
3934
     - the node exists in the configuration
3935
     - OOB is supported
3936

3937
    Any errors are signaled by raising errors.OpPrereqError.
3938

3939
    """
3940
    self.nodes = []
3941
    self.master_node = self.cfg.GetMasterNode()
3942

    
3943
    assert self.op.power_delay >= 0.0
3944

    
3945
    if self.op.node_names:
3946
      if (self.op.command in self._SKIP_MASTER and
3947
          self.master_node in self.op.node_names):
3948
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
3949
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
3950

    
3951
        if master_oob_handler:
3952
          additional_text = ("run '%s %s %s' if you want to operate on the"
3953
                             " master regardless") % (master_oob_handler,
3954
                                                      self.op.command,
3955
                                                      self.master_node)
3956
        else:
3957
          additional_text = "it does not support out-of-band operations"
3958

    
3959
        raise errors.OpPrereqError(("Operating on the master node %s is not"
3960
                                    " allowed for %s; %s") %
3961
                                   (self.master_node, self.op.command,
3962
                                    additional_text), errors.ECODE_INVAL)
3963
    else:
3964
      self.op.node_names = self.cfg.GetNodeList()
3965
      if self.op.command in self._SKIP_MASTER:
3966
        self.op.node_names.remove(self.master_node)
3967

    
3968
    if self.op.command in self._SKIP_MASTER:
3969
      assert self.master_node not in self.op.node_names
3970

    
3971
    for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
3972
      if node is None:
3973
        raise errors.OpPrereqError("Node %s not found" % node_name,
3974
                                   errors.ECODE_NOENT)
3975
      else:
3976
        self.nodes.append(node)
3977

    
3978
      if (not self.op.ignore_status and
3979
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
3980
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
3981
                                    " not marked offline") % node_name,
3982
                                   errors.ECODE_STATE)
3983

    
3984
  def Exec(self, feedback_fn):
3985
    """Execute OOB and return result if we expect any.
3986

3987
    """
3988
    master_node = self.master_node
3989
    ret = []
3990

    
3991
    for idx, node in enumerate(utils.NiceSort(self.nodes,
3992
                                              key=lambda node: node.name)):
3993
      node_entry = [(constants.RS_NORMAL, node.name)]
3994
      ret.append(node_entry)
3995

    
3996
      oob_program = _SupportsOob(self.cfg, node)
3997

    
3998
      if not oob_program:
3999
        node_entry.append((constants.RS_UNAVAIL, None))
4000
        continue
4001

    
4002
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
4003
                   self.op.command, oob_program, node.name)
4004
      result = self.rpc.call_run_oob(master_node, oob_program,
4005
                                     self.op.command, node.name,
4006
                                     self.op.timeout)
4007

    
4008
      if result.fail_msg:
4009
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4010
                        node.name, result.fail_msg)
4011
        node_entry.append((constants.RS_NODATA, None))
4012
      else:
4013
        try:
4014
          self._CheckPayload(result)
4015
        except errors.OpExecError, err:
4016
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
4017
                          node.name, err)
4018
          node_entry.append((constants.RS_NODATA, None))
4019
        else:
4020
          if self.op.command == constants.OOB_HEALTH:
4021
            # For health we should log important events
4022
            for item, status in result.payload:
4023
              if status in [constants.OOB_STATUS_WARNING,
4024
                            constants.OOB_STATUS_CRITICAL]:
4025
                self.LogWarning("Item '%s' on node '%s' has status '%s'",
4026
                                item, node.name, status)
4027

    
4028
          if self.op.command == constants.OOB_POWER_ON:
4029
            node.powered = True
4030
          elif self.op.command == constants.OOB_POWER_OFF:
4031
            node.powered = False
4032
          elif self.op.command == constants.OOB_POWER_STATUS:
4033
            powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
4034
            if powered != node.powered:
4035
              logging.warning(("Recorded power state (%s) of node '%s' does not"
4036
                               " match actual power state (%s)"), node.powered,
4037
                              node.name, powered)
4038

    
4039
          # For configuration changing commands we should update the node
4040
          if self.op.command in (constants.OOB_POWER_ON,
4041
                                 constants.OOB_POWER_OFF):
4042
            self.cfg.Update(node, feedback_fn)
4043

    
4044
          node_entry.append((constants.RS_NORMAL, result.payload))
4045

    
4046
          if (self.op.command == constants.OOB_POWER_ON and
4047
              idx < len(self.nodes) - 1):
4048
            time.sleep(self.op.power_delay)
4049

    
4050
    return ret
4051

    
4052
  def _CheckPayload(self, result):
4053
    """Checks if the payload is valid.
4054

4055
    @param result: RPC result
4056
    @raises errors.OpExecError: If payload is not valid
4057

4058
    """
4059
    errs = []
4060
    if self.op.command == constants.OOB_HEALTH:
4061
      if not isinstance(result.payload, list):
4062
        errs.append("command 'health' is expected to return a list but got %s" %
4063
                    type(result.payload))
4064
      else:
4065
        for item, status in result.payload:
4066
          if status not in constants.OOB_STATUSES:
4067
            errs.append("health item '%s' has invalid status '%s'" %
4068
                        (item, status))
4069

    
4070
    if self.op.command == constants.OOB_POWER_STATUS:
4071
      if not isinstance(result.payload, dict):
4072
        errs.append("power-status is expected to return a dict but got %s" %
4073
                    type(result.payload))
4074

    
4075
    if self.op.command in [
4076
        constants.OOB_POWER_ON,
4077
        constants.OOB_POWER_OFF,
4078
        constants.OOB_POWER_CYCLE,
4079
        ]:
4080
      if result.payload is not None:
4081
        errs.append("%s is expected to not return payload but got '%s'" %
4082
                    (self.op.command, result.payload))
4083

    
4084
    if errs:
4085
      raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
4086
                               utils.CommaJoin(errs))
4087

    
4088

    
4089
class _OsQuery(_QueryBase):
4090
  FIELDS = query.OS_FIELDS
4091

    
4092
  def ExpandNames(self, lu):
4093
    # Lock all nodes in shared mode
4094
    # Temporary removal of locks, should be reverted later
4095
    # TODO: reintroduce locks when they are lighter-weight
4096
    lu.needed_locks = {}
4097
    #self.share_locks[locking.LEVEL_NODE] = 1
4098
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4099

    
4100
    # The following variables interact with _QueryBase._GetNames
4101
    if self.names:
4102
      self.wanted = self.names
4103
    else:
4104
      self.wanted = locking.ALL_SET
4105

    
4106
    self.do_locking = self.use_locking
4107

    
4108
  def DeclareLocks(self, lu, level):
4109
    pass
4110

    
4111
  @staticmethod
4112
  def _DiagnoseByOS(rlist):
4113
    """Remaps a per-node return list into an a per-os per-node dictionary
4114

4115
    @param rlist: a map with node names as keys and OS objects as values
4116

4117
    @rtype: dict
4118
    @return: a dictionary with osnames as keys and as value another
4119
        map, with nodes as keys and tuples of (path, status, diagnose,
4120
        variants, parameters, api_versions) as values, eg::
4121

4122
          {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
4123
                                     (/srv/..., False, "invalid api")],
4124
                           "node2": [(/srv/..., True, "", [], [])]}
4125
          }
4126

4127
    """
4128
    all_os = {}
4129
    # we build here the list of nodes that didn't fail the RPC (at RPC
4130
    # level), so that nodes with a non-responding node daemon don't
4131
    # make all OSes invalid
4132
    good_nodes = [node_name for node_name in rlist
4133
                  if not rlist[node_name].fail_msg]
4134
    for node_name, nr in rlist.items():
4135
      if nr.fail_msg or not nr.payload:
4136
        continue
4137
      for (name, path, status, diagnose, variants,
4138
           params, api_versions) in nr.payload:
4139
        if name not in all_os:
4140
          # build a list of nodes for this os containing empty lists
4141
          # for each node in node_list
4142
          all_os[name] = {}
4143
          for nname in good_nodes:
4144
            all_os[name][nname] = []
4145
        # convert params from [name, help] to (name, help)
4146
        params = [tuple(v) for v in params]
4147
        all_os[name][node_name].append((path, status, diagnose,
4148
                                        variants, params, api_versions))
4149
    return all_os
4150

    
4151
  def _GetQueryData(self, lu):
4152
    """Computes the list of nodes and their attributes.
4153

4154
    """
4155
    # Locking is not used
4156
    assert not (compat.any(lu.glm.is_owned(level)
4157
                           for level in locking.LEVELS
4158
                           if level != locking.LEVEL_CLUSTER) or
4159
                self.do_locking or self.use_locking)
4160

    
4161
    valid_nodes = [node.name
4162
                   for node in lu.cfg.GetAllNodesInfo().values()
4163
                   if not node.offline and node.vm_capable]
4164
    pol = self._DiagnoseByOS(lu.rpc.call_os_diagnose(valid_nodes))
4165
    cluster = lu.cfg.GetClusterInfo()
4166

    
4167
    data = {}
4168

    
4169
    for (os_name, os_data) in pol.items():
4170
      info = query.OsInfo(name=os_name, valid=True, node_status=os_data,
4171
                          hidden=(os_name in cluster.hidden_os),
4172
                          blacklisted=(os_name in cluster.blacklisted_os))
4173

    
4174
      variants = set()
4175
      parameters = set()
4176
      api_versions = set()
4177

    
4178
      for idx, osl in enumerate(os_data.values()):
4179
        info.valid = bool(info.valid and osl and osl[0][1])
4180
        if not info.valid:
4181
          break
4182

    
4183
        (node_variants, node_params, node_api) = osl[0][3:6]
4184
        if idx == 0:
4185
          # First entry
4186
          variants.update(node_variants)
4187
          parameters.update(node_params)
4188
          api_versions.update(node_api)
4189
        else:
4190
          # Filter out inconsistent values
4191
          variants.intersection_update(node_variants)
4192
          parameters.intersection_update(node_params)
4193
          api_versions.intersection_update(node_api)
4194

    
4195
      info.variants = list(variants)
4196
      info.parameters = list(parameters)
4197
      info.api_versions = list(api_versions)
4198

    
4199
      data[os_name] = info
4200

    
4201
    # Prepare data in requested order
4202
    return [data[name] for name in self._GetNames(lu, pol.keys(), None)
4203
            if name in data]
4204

    
4205

    
4206
class LUOsDiagnose(NoHooksLU):
4207
  """Logical unit for OS diagnose/query.
4208

4209
  """
4210
  REQ_BGL = False
4211

    
4212
  @staticmethod
4213
  def _BuildFilter(fields, names):
4214
    """Builds a filter for querying OSes.
4215

4216
    """
4217
    name_filter = qlang.MakeSimpleFilter("name", names)
4218

    
4219
    # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the
4220
    # respective field is not requested
4221
    status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]
4222
                     for fname in ["hidden", "blacklisted"]
4223
                     if fname not in fields]
4224
    if "valid" not in fields:
4225
      status_filter.append([qlang.OP_TRUE, "valid"])
4226

    
4227
    if status_filter:
4228
      status_filter.insert(0, qlang.OP_AND)
4229
    else:
4230
      status_filter = None
4231

    
4232
    if name_filter and status_filter:
4233
      return [qlang.OP_AND, name_filter, status_filter]
4234
    elif name_filter:
4235
      return name_filter
4236
    else:
4237
      return status_filter
4238

    
4239
  def CheckArguments(self):
4240
    self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),
4241
                       self.op.output_fields, False)
4242

    
4243
  def ExpandNames(self):
4244
    self.oq.ExpandNames(self)
4245

    
4246
  def Exec(self, feedback_fn):
4247
    return self.oq.OldStyleQuery(self)
4248

    
4249

    
4250
class LUNodeRemove(LogicalUnit):
4251
  """Logical unit for removing a node.
4252

4253
  """
4254
  HPATH = "node-remove"
4255
  HTYPE = constants.HTYPE_NODE
4256

    
4257
  def BuildHooksEnv(self):
4258
    """Build hooks env.
4259

4260
    This doesn't run on the target node in the pre phase as a failed
4261
    node would then be impossible to remove.
4262

4263
    """
4264
    return {
4265
      "OP_TARGET": self.op.node_name,
4266
      "NODE_NAME": self.op.node_name,
4267
      }
4268

    
4269
  def BuildHooksNodes(self):
4270
    """Build hooks nodes.
4271

4272
    """
4273
    all_nodes = self.cfg.GetNodeList()
4274
    try:
4275
      all_nodes.remove(self.op.node_name)
4276
    except ValueError:
4277
      logging.warning("Node '%s', which is about to be removed, was not found"
4278
                      " in the list of all nodes", self.op.node_name)
4279
    return (all_nodes, all_nodes)
4280

    
4281
  def CheckPrereq(self):
4282
    """Check prerequisites.
4283

4284
    This checks:
4285
     - the node exists in the configuration
4286
     - it does not have primary or secondary instances
4287
     - it's not the master
4288

4289
    Any errors are signaled by raising errors.OpPrereqError.
4290

4291
    """
4292
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4293
    node = self.cfg.GetNodeInfo(self.op.node_name)
4294
    assert node is not None
4295

    
4296
    masternode = self.cfg.GetMasterNode()
4297
    if node.name == masternode:
4298
      raise errors.OpPrereqError("Node is the master node, failover to another"
4299
                                 " node is required", errors.ECODE_INVAL)
4300

    
4301
    for instance_name, instance in self.cfg.GetAllInstancesInfo():
4302
      if node.name in instance.all_nodes:
4303
        raise errors.OpPrereqError("Instance %s is still running on the node,"
4304
                                   " please remove first" % instance_name,
4305
                                   errors.ECODE_INVAL)
4306
    self.op.node_name = node.name
4307
    self.node = node
4308

    
4309
  def Exec(self, feedback_fn):
4310
    """Removes the node from the cluster.
4311

4312
    """
4313
    node = self.node
4314
    logging.info("Stopping the node daemon and removing configs from node %s",
4315
                 node.name)
4316

    
4317
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
4318

    
4319
    # Promote nodes to master candidate as needed
4320
    _AdjustCandidatePool(self, exceptions=[node.name])
4321
    self.context.RemoveNode(node.name)
4322

    
4323
    # Run post hooks on the node before it's removed
4324
    _RunPostHook(self, node.name)
4325

    
4326
    result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
4327
    msg = result.fail_msg
4328
    if msg:
4329
      self.LogWarning("Errors encountered on the remote node while leaving"
4330
                      " the cluster: %s", msg)
4331

    
4332
    # Remove node from our /etc/hosts
4333
    if self.cfg.GetClusterInfo().modify_etc_hosts:
4334
      master_node = self.cfg.GetMasterNode()
4335
      result = self.rpc.call_etc_hosts_modify(master_node,
4336
                                              constants.ETC_HOSTS_REMOVE,
4337
                                              node.name, None)
4338
      result.Raise("Can't update hosts file with new host data")
4339
      _RedistributeAncillaryFiles(self)
4340

    
4341

    
4342
class _NodeQuery(_QueryBase):
4343
  FIELDS = query.NODE_FIELDS
4344

    
4345
  def ExpandNames(self, lu):
4346
    lu.needed_locks = {}
4347
    lu.share_locks = _ShareAll()
4348

    
4349
    if self.names:
4350
      self.wanted = _GetWantedNodes(lu, self.names)
4351
    else:
4352
      self.wanted = locking.ALL_SET
4353

    
4354
    self.do_locking = (self.use_locking and
4355
                       query.NQ_LIVE in self.requested_data)
4356

    
4357
    if self.do_locking:
4358
      # If any non-static field is requested we need to lock the nodes
4359
      lu.needed_locks[locking.LEVEL_NODE] = self.wanted
4360

    
4361
  def DeclareLocks(self, lu, level):
4362
    pass
4363

    
4364
  def _GetQueryData(self, lu):
4365
    """Computes the list of nodes and their attributes.
4366

4367
    """
4368
    all_info = lu.cfg.GetAllNodesInfo()
4369

    
4370
    nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
4371

    
4372
    # Gather data as requested
4373
    if query.NQ_LIVE in self.requested_data:
4374
      # filter out non-vm_capable nodes
4375
      toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
4376

    
4377
      node_data = lu.rpc.call_node_info(toquery_nodes, lu.cfg.GetVGName(),
4378
                                        lu.cfg.GetHypervisorType())
4379
      live_data = dict((name, nresult.payload)
4380
                       for (name, nresult) in node_data.items()
4381
                       if not nresult.fail_msg and nresult.payload)
4382
    else:
4383
      live_data = None
4384

    
4385
    if query.NQ_INST in self.requested_data:
4386
      node_to_primary = dict([(name, set()) for name in nodenames])
4387
      node_to_secondary = dict([(name, set()) for name in nodenames])
4388

    
4389
      inst_data = lu.cfg.GetAllInstancesInfo()
4390

    
4391
      for inst in inst_data.values():
4392
        if inst.primary_node in node_to_primary:
4393
          node_to_primary[inst.primary_node].add(inst.name)
4394
        for secnode in inst.secondary_nodes:
4395
          if secnode in node_to_secondary:
4396
            node_to_secondary[secnode].add(inst.name)
4397
    else:
4398
      node_to_primary = None
4399
      node_to_secondary = None
4400

    
4401
    if query.NQ_OOB in self.requested_data:
4402
      oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
4403
                         for name, node in all_info.iteritems())
4404
    else:
4405
      oob_support = None
4406

    
4407
    if query.NQ_GROUP in self.requested_data:
4408
      groups = lu.cfg.GetAllNodeGroupsInfo()
4409
    else:
4410
      groups = {}
4411

    
4412
    return query.NodeQueryData([all_info[name] for name in nodenames],
4413
                               live_data, lu.cfg.GetMasterNode(),
4414
                               node_to_primary, node_to_secondary, groups,
4415
                               oob_support, lu.cfg.GetClusterInfo())
4416

    
4417

    
4418
class LUNodeQuery(NoHooksLU):
4419
  """Logical unit for querying nodes.
4420

4421
  """
4422
  # pylint: disable=W0142
4423
  REQ_BGL = False
4424

    
4425
  def CheckArguments(self):
4426
    self.nq = _NodeQuery(qlang.MakeSimpleFilter("name", self.op.names),
4427
                         self.op.output_fields, self.op.use_locking)
4428

    
4429
  def ExpandNames(self):
4430
    self.nq.ExpandNames(self)
4431

    
4432
  def Exec(self, feedback_fn):
4433
    return self.nq.OldStyleQuery(self)
4434

    
4435

    
4436
class LUNodeQueryvols(NoHooksLU):
4437
  """Logical unit for getting volumes on node(s).
4438

4439
  """
4440
  REQ_BGL = False
4441
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
4442
  _FIELDS_STATIC = utils.FieldSet("node")
4443

    
4444
  def CheckArguments(self):
4445
    _CheckOutputFields(static=self._FIELDS_STATIC,
4446
                       dynamic=self._FIELDS_DYNAMIC,
4447
                       selected=self.op.output_fields)
4448

    
4449
  def ExpandNames(self):
4450
    self.needed_locks = {}
4451
    self.share_locks[locking.LEVEL_NODE] = 1
4452
    if not self.op.nodes:
4453
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4454
    else:
4455
      self.needed_locks[locking.LEVEL_NODE] = \
4456
        _GetWantedNodes(self, self.op.nodes)
4457

    
4458
  def Exec(self, feedback_fn):
4459
    """Computes the list of nodes and their attributes.
4460

4461
    """
4462
    nodenames = self.owned_locks(locking.LEVEL_NODE)
4463
    volumes = self.rpc.call_node_volumes(nodenames)
4464

    
4465
    ilist = self.cfg.GetAllInstancesInfo()
4466
    vol2inst = _MapInstanceDisksToNodes(ilist.values())
4467

    
4468
    output = []
4469
    for node in nodenames:
4470
      nresult = volumes[node]
4471
      if nresult.offline:
4472
        continue
4473
      msg = nresult.fail_msg
4474
      if msg:
4475
        self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
4476
        continue
4477

    
4478
      node_vols = sorted(nresult.payload,
4479
                         key=operator.itemgetter("dev"))
4480

    
4481
      for vol in node_vols:
4482
        node_output = []
4483
        for field in self.op.output_fields:
4484
          if field == "node":
4485
            val = node
4486
          elif field == "phys":
4487
            val = vol["dev"]
4488
          elif field == "vg":
4489
            val = vol["vg"]
4490
          elif field == "name":
4491
            val = vol["name"]
4492
          elif field == "size":
4493
            val = int(float(vol["size"]))
4494
          elif field == "instance":
4495
            val = vol2inst.get((node, vol["vg"] + "/" + vol["name"]), "-")
4496
          else:
4497
            raise errors.ParameterError(field)
4498
          node_output.append(str(val))
4499

    
4500
        output.append(node_output)
4501

    
4502
    return output
4503

    
4504

    
4505
class LUNodeQueryStorage(NoHooksLU):
4506
  """Logical unit for getting information on storage units on node(s).
4507

4508
  """
4509
  _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
4510
  REQ_BGL = False
4511

    
4512
  def CheckArguments(self):
4513
    _CheckOutputFields(static=self._FIELDS_STATIC,
4514
                       dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
4515
                       selected=self.op.output_fields)
4516

    
4517
  def ExpandNames(self):
4518
    self.needed_locks = {}
4519
    self.share_locks[locking.LEVEL_NODE] = 1
4520

    
4521
    if self.op.nodes:
4522
      self.needed_locks[locking.LEVEL_NODE] = \
4523
        _GetWantedNodes(self, self.op.nodes)
4524
    else:
4525
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4526

    
4527
  def Exec(self, feedback_fn):
4528
    """Computes the list of nodes and their attributes.
4529

4530
    """
4531
    self.nodes = self.owned_locks(locking.LEVEL_NODE)
4532

    
4533
    # Always get name to sort by
4534
    if constants.SF_NAME in self.op.output_fields:
4535
      fields = self.op.output_fields[:]
4536
    else:
4537
      fields = [constants.SF_NAME] + self.op.output_fields
4538

    
4539
    # Never ask for node or type as it's only known to the LU
4540
    for extra in [constants.SF_NODE, constants.SF_TYPE]:
4541
      while extra in fields:
4542
        fields.remove(extra)
4543

    
4544
    field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
4545
    name_idx = field_idx[constants.SF_NAME]
4546

    
4547
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4548
    data = self.rpc.call_storage_list(self.nodes,
4549
                                      self.op.storage_type, st_args,
4550
                                      self.op.name, fields)
4551

    
4552
    result = []
4553

    
4554
    for node in utils.NiceSort(self.nodes):
4555
      nresult = data[node]
4556
      if nresult.offline:
4557
        continue
4558

    
4559
      msg = nresult.fail_msg
4560
      if msg:
4561
        self.LogWarning("Can't get storage data from node %s: %s", node, msg)
4562
        continue
4563

    
4564
      rows = dict([(row[name_idx], row) for row in nresult.payload])
4565

    
4566
      for name in utils.NiceSort(rows.keys()):
4567
        row = rows[name]
4568

    
4569
        out = []
4570

    
4571
        for field in self.op.output_fields:
4572
          if field == constants.SF_NODE:
4573
            val = node
4574
          elif field == constants.SF_TYPE:
4575
            val = self.op.storage_type
4576
          elif field in field_idx:
4577
            val = row[field_idx[field]]
4578
          else:
4579
            raise errors.ParameterError(field)
4580

    
4581
          out.append(val)
4582

    
4583
        result.append(out)
4584

    
4585
    return result
4586

    
4587

    
4588
class _InstanceQuery(_QueryBase):
4589
  FIELDS = query.INSTANCE_FIELDS
4590

    
4591
  def ExpandNames(self, lu):
4592
    lu.needed_locks = {}
4593
    lu.share_locks = _ShareAll()
4594

    
4595
    if self.names:
4596
      self.wanted = _GetWantedInstances(lu, self.names)
4597
    else:
4598
      self.wanted = locking.ALL_SET
4599

    
4600
    self.do_locking = (self.use_locking and
4601
                       query.IQ_LIVE in self.requested_data)
4602
    if self.do_locking:
4603
      lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
4604
      lu.needed_locks[locking.LEVEL_NODEGROUP] = []
4605
      lu.needed_locks[locking.LEVEL_NODE] = []
4606
      lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4607

    
4608
    self.do_grouplocks = (self.do_locking and
4609
                          query.IQ_NODES in self.requested_data)
4610

    
4611
  def DeclareLocks(self, lu, level):
4612
    if self.do_locking:
4613
      if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:
4614
        assert not lu.needed_locks[locking.LEVEL_NODEGROUP]
4615

    
4616
        # Lock all groups used by instances optimistically; this requires going
4617
        # via the node before it's locked, requiring verification later on
4618
        lu.needed_locks[locking.LEVEL_NODEGROUP] = \
4619
          set(group_uuid
4620
              for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
4621
              for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
4622
      elif level == locking.LEVEL_NODE:
4623
        lu._LockInstancesNodes() # pylint: disable=W0212
4624

    
4625
  @staticmethod
4626
  def _CheckGroupLocks(lu):
4627
    owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))
4628
    owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))
4629

    
4630
    # Check if node groups for locked instances are still correct
4631
    for instance_name in owned_instances:
4632
      _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
4633

    
4634
  def _GetQueryData(self, lu):
4635
    """Computes the list of instances and their attributes.
4636

4637
    """
4638
    if self.do_grouplocks:
4639
      self._CheckGroupLocks(lu)
4640

    
4641
    cluster = lu.cfg.GetClusterInfo()
4642
    all_info = lu.cfg.GetAllInstancesInfo()
4643

    
4644
    instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
4645

    
4646
    instance_list = [all_info[name] for name in instance_names]
4647
    nodes = frozenset(itertools.chain(*(inst.all_nodes
4648
                                        for inst in instance_list)))
4649
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
4650
    bad_nodes = []
4651
    offline_nodes = []
4652
    wrongnode_inst = set()
4653

    
4654
    # Gather data as requested
4655
    if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
4656
      live_data = {}
4657
      node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
4658
      for name in nodes:
4659
        result = node_data[name]
4660
        if result.offline:
4661
          # offline nodes will be in both lists
4662
          assert result.fail_msg
4663
          offline_nodes.append(name)
4664
        if result.fail_msg:
4665
          bad_nodes.append(name)
4666
        elif result.payload:
4667
          for inst in result.payload:
4668
            if inst in all_info:
4669
              if all_info[inst].primary_node == name:
4670
                live_data.update(result.payload)
4671
              else:
4672
                wrongnode_inst.add(inst)
4673
            else:
4674
              # orphan instance; we don't list it here as we don't
4675
              # handle this case yet in the output of instance listing
4676
              logging.warning("Orphan instance '%s' found on node %s",
4677
                              inst, name)
4678
        # else no instance is alive
4679
    else:
4680
      live_data = {}
4681

    
4682
    if query.IQ_DISKUSAGE in self.requested_data:
4683
      disk_usage = dict((inst.name,
4684
                         _ComputeDiskSize(inst.disk_template,
4685
                                          [{constants.IDISK_SIZE: disk.size}
4686
                                           for disk in inst.disks]))
4687
                        for inst in instance_list)
4688
    else:
4689
      disk_usage = None
4690

    
4691
    if query.IQ_CONSOLE in self.requested_data:
4692
      consinfo = {}
4693
      for inst in instance_list:
4694
        if inst.name in live_data:
4695
          # Instance is running
4696
          consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
4697
        else:
4698
          consinfo[inst.name] = None
4699
      assert set(consinfo.keys()) == set(instance_names)
4700
    else:
4701
      consinfo = None
4702

    
4703
    if query.IQ_NODES in self.requested_data:
4704
      node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),
4705
                                            instance_list)))
4706
      nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))
4707
      groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))
4708
                    for uuid in set(map(operator.attrgetter("group"),
4709
                                        nodes.values())))
4710
    else:
4711
      nodes = None
4712
      groups = None
4713

    
4714
    return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
4715
                                   disk_usage, offline_nodes, bad_nodes,
4716
                                   live_data, wrongnode_inst, consinfo,
4717
                                   nodes, groups)
4718

    
4719

    
4720
class LUQuery(NoHooksLU):
4721
  """Query for resources/items of a certain kind.
4722

4723
  """
4724
  # pylint: disable=W0142
4725
  REQ_BGL = False
4726

    
4727
  def CheckArguments(self):
4728
    qcls = _GetQueryImplementation(self.op.what)
4729

    
4730
    self.impl = qcls(self.op.filter, self.op.fields, self.op.use_locking)
4731

    
4732
  def ExpandNames(self):
4733
    self.impl.ExpandNames(self)
4734

    
4735
  def DeclareLocks(self, level):
4736
    self.impl.DeclareLocks(self, level)
4737

    
4738
  def Exec(self, feedback_fn):
4739
    return self.impl.NewStyleQuery(self)
4740

    
4741

    
4742
class LUQueryFields(NoHooksLU):
4743
  """Query for resources/items of a certain kind.
4744

4745
  """
4746
  # pylint: disable=W0142
4747
  REQ_BGL = False
4748

    
4749
  def CheckArguments(self):
4750
    self.qcls = _GetQueryImplementation(self.op.what)
4751

    
4752
  def ExpandNames(self):
4753
    self.needed_locks = {}
4754

    
4755
  def Exec(self, feedback_fn):
4756
    return query.QueryFields(self.qcls.FIELDS, self.op.fields)
4757

    
4758

    
4759
class LUNodeModifyStorage(NoHooksLU):
4760
  """Logical unit for modifying a storage volume on a node.
4761

4762
  """
4763
  REQ_BGL = False
4764

    
4765
  def CheckArguments(self):
4766
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4767

    
4768
    storage_type = self.op.storage_type
4769

    
4770
    try:
4771
      modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
4772
    except KeyError:
4773
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
4774
                                 " modified" % storage_type,
4775
                                 errors.ECODE_INVAL)
4776

    
4777
    diff = set(self.op.changes.keys()) - modifiable
4778
    if diff:
4779
      raise errors.OpPrereqError("The following fields can not be modified for"
4780
                                 " storage units of type '%s': %r" %
4781
                                 (storage_type, list(diff)),
4782
                                 errors.ECODE_INVAL)
4783

    
4784
  def ExpandNames(self):
4785
    self.needed_locks = {
4786
      locking.LEVEL_NODE: self.op.node_name,
4787
      }
4788

    
4789
  def Exec(self, feedback_fn):
4790
    """Computes the list of nodes and their attributes.
4791

4792
    """
4793
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4794
    result = self.rpc.call_storage_modify(self.op.node_name,
4795
                                          self.op.storage_type, st_args,
4796
                                          self.op.name, self.op.changes)
4797
    result.Raise("Failed to modify storage unit '%s' on %s" %
4798
                 (self.op.name, self.op.node_name))
4799

    
4800

    
4801
class LUNodeAdd(LogicalUnit):
4802
  """Logical unit for adding node to the cluster.
4803

4804
  """
4805
  HPATH = "node-add"
4806
  HTYPE = constants.HTYPE_NODE
4807
  _NFLAGS = ["master_capable", "vm_capable"]
4808

    
4809
  def CheckArguments(self):
4810
    self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
4811
    # validate/normalize the node name
4812
    self.hostname = netutils.GetHostname(name=self.op.node_name,
4813
                                         family=self.primary_ip_family)
4814
    self.op.node_name = self.hostname.name
4815

    
4816
    if self.op.readd and self.op.node_name == self.cfg.GetMasterNode():
4817
      raise errors.OpPrereqError("Cannot readd the master node",
4818
                                 errors.ECODE_STATE)
4819

    
4820
    if self.op.readd and self.op.group:
4821
      raise errors.OpPrereqError("Cannot pass a node group when a node is"
4822
                                 " being readded", errors.ECODE_INVAL)
4823

    
4824
  def BuildHooksEnv(self):
4825
    """Build hooks env.
4826

4827
    This will run on all nodes before, and on all nodes + the new node after.
4828

4829
    """
4830
    return {
4831
      "OP_TARGET": self.op.node_name,
4832
      "NODE_NAME": self.op.node_name,
4833
      "NODE_PIP": self.op.primary_ip,
4834
      "NODE_SIP": self.op.secondary_ip,
4835
      "MASTER_CAPABLE": str(self.op.master_capable),
4836
      "VM_CAPABLE": str(self.op.vm_capable),
4837
      }
4838

    
4839
  def BuildHooksNodes(self):
4840
    """Build hooks nodes.
4841

4842
    """
4843
    # Exclude added node
4844
    pre_nodes = list(set(self.cfg.GetNodeList()) - set([self.op.node_name]))
4845
    post_nodes = pre_nodes + [self.op.node_name, ]
4846

    
4847
    return (pre_nodes, post_nodes)
4848

    
4849
  def CheckPrereq(self):
4850
    """Check prerequisites.
4851

4852
    This checks:
4853
     - the new node is not already in the config
4854
     - it is resolvable
4855
     - its parameters (single/dual homed) matches the cluster
4856

4857
    Any errors are signaled by raising errors.OpPrereqError.
4858

4859
    """
4860
    cfg = self.cfg
4861
    hostname = self.hostname
4862
    node = hostname.name
4863
    primary_ip = self.op.primary_ip = hostname.ip
4864
    if self.op.secondary_ip is None:
4865
      if self.primary_ip_family == netutils.IP6Address.family:
4866
        raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
4867
                                   " IPv4 address must be given as secondary",
4868
                                   errors.ECODE_INVAL)
4869
      self.op.secondary_ip = primary_ip
4870

    
4871
    secondary_ip = self.op.secondary_ip
4872
    if not netutils.IP4Address.IsValid(secondary_ip):
4873
      raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
4874
                                 " address" % secondary_ip, errors.ECODE_INVAL)
4875

    
4876
    node_list = cfg.GetNodeList()
4877
    if not self.op.readd and node in node_list:
4878
      raise errors.OpPrereqError("Node %s is already in the configuration" %
4879
                                 node, errors.ECODE_EXISTS)
4880
    elif self.op.readd and node not in node_list:
4881
      raise errors.OpPrereqError("Node %s is not in the configuration" % node,
4882
                                 errors.ECODE_NOENT)
4883

    
4884
    self.changed_primary_ip = False
4885

    
4886
    for existing_node_name, existing_node in cfg.GetMultiNodeInfo(node_list):
4887
      if self.op.readd and node == existing_node_name:
4888
        if existing_node.secondary_ip != secondary_ip:
4889
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
4890
                                     " address configuration as before",
4891
                                     errors.ECODE_INVAL)
4892
        if existing_node.primary_ip != primary_ip:
4893
          self.changed_primary_ip = True
4894

    
4895
        continue
4896

    
4897
      if (existing_node.primary_ip == primary_ip or
4898
          existing_node.secondary_ip == primary_ip or
4899
          existing_node.primary_ip == secondary_ip or
4900
          existing_node.secondary_ip == secondary_ip):
4901
        raise errors.OpPrereqError("New node ip address(es) conflict with"
4902
                                   " existing node %s" % existing_node.name,
4903
                                   errors.ECODE_NOTUNIQUE)
4904

    
4905
    # After this 'if' block, None is no longer a valid value for the
4906
    # _capable op attributes
4907
    if self.op.readd:
4908
      old_node = self.cfg.GetNodeInfo(node)
4909
      assert old_node is not None, "Can't retrieve locked node %s" % node
4910
      for attr in self._NFLAGS:
4911
        if getattr(self.op, attr) is None:
4912
          setattr(self.op, attr, getattr(old_node, attr))
4913
    else:
4914
      for attr in self._NFLAGS:
4915
        if getattr(self.op, attr) is None:
4916
          setattr(self.op, attr, True)
4917

    
4918
    if self.op.readd and not self.op.vm_capable:
4919
      pri, sec = cfg.GetNodeInstances(node)
4920
      if pri or sec:
4921
        raise errors.OpPrereqError("Node %s being re-added with vm_capable"
4922
                                   " flag set to false, but it already holds"
4923
                                   " instances" % node,
4924
                                   errors.ECODE_STATE)
4925

    
4926
    # check that the type of the node (single versus dual homed) is the
4927
    # same as for the master
4928
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
4929
    master_singlehomed = myself.secondary_ip == myself.primary_ip
4930
    newbie_singlehomed = secondary_ip == primary_ip
4931
    if master_singlehomed != newbie_singlehomed:
4932
      if master_singlehomed:
4933
        raise errors.OpPrereqError("The master has no secondary ip but the"
4934
                                   " new node has one",
4935
                                   errors.ECODE_INVAL)
4936
      else:
4937
        raise errors.OpPrereqError("The master has a secondary ip but the"
4938
                                   " new node doesn't have one",
4939
                                   errors.ECODE_INVAL)
4940

    
4941
    # checks reachability
4942
    if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
4943
      raise errors.OpPrereqError("Node not reachable by ping",
4944
                                 errors.ECODE_ENVIRON)
4945

    
4946
    if not newbie_singlehomed:
4947
      # check reachability from my secondary ip to newbie's secondary ip
4948
      if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
4949
                           source=myself.secondary_ip):
4950
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
4951
                                   " based ping to node daemon port",
4952
                                   errors.ECODE_ENVIRON)
4953

    
4954
    if self.op.readd:
4955
      exceptions = [node]
4956
    else:
4957
      exceptions = []
4958

    
4959
    if self.op.master_capable:
4960
      self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
4961
    else:
4962
      self.master_candidate = False
4963

    
4964
    if self.op.readd:
4965
      self.new_node = old_node
4966
    else:
4967
      node_group = cfg.LookupNodeGroup(self.op.group)
4968
      self.new_node = objects.Node(name=node,
4969
                                   primary_ip=primary_ip,
4970
                                   secondary_ip=secondary_ip,
4971
                                   master_candidate=self.master_candidate,
4972
                                   offline=False, drained=False,
4973
                                   group=node_group)
4974

    
4975
    if self.op.ndparams:
4976
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
4977

    
4978
  def Exec(self, feedback_fn):
4979
    """Adds the new node to the cluster.
4980

4981
    """
4982
    new_node = self.new_node
4983
    node = new_node.name
4984

    
4985
    # We adding a new node so we assume it's powered
4986
    new_node.powered = True
4987

    
4988
    # for re-adds, reset the offline/drained/master-candidate flags;
4989
    # we need to reset here, otherwise offline would prevent RPC calls
4990
    # later in the procedure; this also means that if the re-add
4991
    # fails, we are left with a non-offlined, broken node
4992
    if self.op.readd:
4993
      new_node.drained = new_node.offline = False # pylint: disable=W0201
4994
      self.LogInfo("Readding a node, the offline/drained flags were reset")
4995
      # if we demote the node, we do cleanup later in the procedure
4996
      new_node.master_candidate = self.master_candidate
4997
      if self.changed_primary_ip:
4998
        new_node.primary_ip = self.op.primary_ip
4999

    
5000
    # copy the master/vm_capable flags
5001
    for attr in self._NFLAGS:
5002
      setattr(new_node, attr, getattr(self.op, attr))
5003

    
5004
    # notify the user about any possible mc promotion
5005
    if new_node.master_candidate:
5006
      self.LogInfo("Node will be a master candidate")
5007

    
5008
    if self.op.ndparams:
5009
      new_node.ndparams = self.op.ndparams
5010
    else:
5011
      new_node.ndparams = {}
5012

    
5013
    # check connectivity
5014
    result = self.rpc.call_version([node])[node]
5015
    result.Raise("Can't get version information from node %s" % node)
5016
    if constants.PROTOCOL_VERSION == result.payload:
5017
      logging.info("Communication to node %s fine, sw version %s match",
5018
                   node, result.payload)
5019
    else:
5020
      raise errors.OpExecError("Version mismatch master version %s,"
5021
                               " node version %s" %
5022
                               (constants.PROTOCOL_VERSION, result.payload))
5023

    
5024
    # Add node to our /etc/hosts, and add key to known_hosts
5025
    if self.cfg.GetClusterInfo().modify_etc_hosts:
5026
      master_node = self.cfg.GetMasterNode()
5027
      result = self.rpc.call_etc_hosts_modify(master_node,
5028
                                              constants.ETC_HOSTS_ADD,
5029
                                              self.hostname.name,
5030
                                              self.hostname.ip)
5031
      result.Raise("Can't update hosts file with new host data")
5032

    
5033
    if new_node.secondary_ip != new_node.primary_ip:
5034
      _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
5035
                               False)
5036

    
5037
    node_verify_list = [self.cfg.GetMasterNode()]
5038
    node_verify_param = {
5039
      constants.NV_NODELIST: [node],
5040
      # TODO: do a node-net-test as well?
5041
    }
5042

    
5043
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
5044
                                       self.cfg.GetClusterName())
5045
    for verifier in node_verify_list:
5046
      result[verifier].Raise("Cannot communicate with node %s" % verifier)
5047
      nl_payload = result[verifier].payload[constants.NV_NODELIST]
5048
      if nl_payload:
5049
        for failed in nl_payload:
5050
          feedback_fn("ssh/hostname verification failed"
5051
                      " (checking from %s): %s" %
5052
                      (verifier, nl_payload[failed]))
5053
        raise errors.OpExecError("ssh/hostname verification failed")
5054

    
5055
    if self.op.readd:
5056
      _RedistributeAncillaryFiles(self)
5057
      self.context.ReaddNode(new_node)
5058
      # make sure we redistribute the config
5059
      self.cfg.Update(new_node, feedback_fn)
5060
      # and make sure the new node will not have old files around
5061
      if not new_node.master_candidate:
5062
        result = self.rpc.call_node_demote_from_mc(new_node.name)
5063
        msg = result.fail_msg
5064
        if msg:
5065
          self.LogWarning("Node failed to demote itself from master"
5066
                          " candidate status: %s" % msg)
5067
    else:
5068
      _RedistributeAncillaryFiles(self, additional_nodes=[node],
5069
                                  additional_vm=self.op.vm_capable)
5070
      self.context.AddNode(new_node, self.proc.GetECId())
5071

    
5072

    
5073
class LUNodeSetParams(LogicalUnit):
5074
  """Modifies the parameters of a node.
5075

5076
  @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
5077
      to the node role (as _ROLE_*)
5078
  @cvar _R2F: a dictionary from node role to tuples of flags
5079
  @cvar _FLAGS: a list of attribute names corresponding to the flags
5080

5081
  """
5082
  HPATH = "node-modify"
5083
  HTYPE = constants.HTYPE_NODE
5084
  REQ_BGL = False
5085
  (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
5086
  _F2R = {
5087
    (True, False, False): _ROLE_CANDIDATE,
5088
    (False, True, False): _ROLE_DRAINED,
5089
    (False, False, True): _ROLE_OFFLINE,
5090
    (False, False, False): _ROLE_REGULAR,
5091
    }
5092
  _R2F = dict((v, k) for k, v in _F2R.items())
5093
  _FLAGS = ["master_candidate", "drained", "offline"]
5094

    
5095
  def CheckArguments(self):
5096
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5097
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
5098
                self.op.master_capable, self.op.vm_capable,
5099
                self.op.secondary_ip, self.op.ndparams]
5100
    if all_mods.count(None) == len(all_mods):
5101
      raise errors.OpPrereqError("Please pass at least one modification",
5102
                                 errors.ECODE_INVAL)
5103
    if all_mods.count(True) > 1:
5104
      raise errors.OpPrereqError("Can't set the node into more than one"
5105
                                 " state at the same time",
5106
                                 errors.ECODE_INVAL)
5107

    
5108
    # Boolean value that tells us whether we might be demoting from MC
5109
    self.might_demote = (self.op.master_candidate == False or
5110
                         self.op.offline == True or
5111
                         self.op.drained == True or
5112
                         self.op.master_capable == False)
5113

    
5114
    if self.op.secondary_ip:
5115
      if not netutils.IP4Address.IsValid(self.op.secondary_ip):
5116
        raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5117
                                   " address" % self.op.secondary_ip,
5118
                                   errors.ECODE_INVAL)
5119

    
5120
    self.lock_all = self.op.auto_promote and self.might_demote
5121
    self.lock_instances = self.op.secondary_ip is not None
5122

    
5123
  def ExpandNames(self):
5124
    if self.lock_all:
5125
      self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
5126
    else:
5127
      self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
5128

    
5129
    if self.lock_instances:
5130
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5131

    
5132
  def DeclareLocks(self, level):
5133
    # If we have locked all instances, before waiting to lock nodes, release
5134
    # all the ones living on nodes unrelated to the current operation.
5135
    if level == locking.LEVEL_NODE and self.lock_instances:
5136
      self.affected_instances = []
5137
      if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
5138
        instances_keep = []
5139

    
5140
        # Build list of instances to release
5141
        locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
5142
        for instance_name, instance in self.cfg.GetMultiInstanceInfo(locked_i):
5143
          if (instance.disk_template in constants.DTS_INT_MIRROR and
5144
              self.op.node_name in instance.all_nodes):
5145
            instances_keep.append(instance_name)
5146
            self.affected_instances.append(instance)
5147

    
5148
        _ReleaseLocks(self, locking.LEVEL_INSTANCE, keep=instances_keep)
5149

    
5150
        assert (set(self.owned_locks(locking.LEVEL_INSTANCE)) ==
5151
                set(instances_keep))
5152

    
5153
  def BuildHooksEnv(self):
5154
    """Build hooks env.
5155

5156
    This runs on the master node.
5157

5158
    """
5159
    return {
5160
      "OP_TARGET": self.op.node_name,
5161
      "MASTER_CANDIDATE": str(self.op.master_candidate),
5162
      "OFFLINE": str(self.op.offline),
5163
      "DRAINED": str(self.op.drained),
5164
      "MASTER_CAPABLE": str(self.op.master_capable),
5165
      "VM_CAPABLE": str(self.op.vm_capable),
5166
      }
5167

    
5168
  def BuildHooksNodes(self):
5169
    """Build hooks nodes.
5170

5171
    """
5172
    nl = [self.cfg.GetMasterNode(), self.op.node_name]
5173
    return (nl, nl)
5174

    
5175
  def CheckPrereq(self):
5176
    """Check prerequisites.
5177

5178
    This only checks the instance list against the existing names.
5179

5180
    """
5181
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
5182

    
5183
    if (self.op.master_candidate is not None or
5184
        self.op.drained is not None or
5185
        self.op.offline is not None):
5186
      # we can't change the master's node flags
5187
      if self.op.node_name == self.cfg.GetMasterNode():
5188
        raise errors.OpPrereqError("The master role can be changed"
5189
                                   " only via master-failover",
5190
                                   errors.ECODE_INVAL)
5191

    
5192
    if self.op.master_candidate and not node.master_capable:
5193
      raise errors.OpPrereqError("Node %s is not master capable, cannot make"
5194
                                 " it a master candidate" % node.name,
5195
                                 errors.ECODE_STATE)
5196

    
5197
    if self.op.vm_capable == False:
5198
      (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
5199
      if ipri or isec:
5200
        raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
5201
                                   " the vm_capable flag" % node.name,
5202
                                   errors.ECODE_STATE)
5203

    
5204
    if node.master_candidate and self.might_demote and not self.lock_all:
5205
      assert not self.op.auto_promote, "auto_promote set but lock_all not"
5206
      # check if after removing the current node, we're missing master
5207
      # candidates
5208
      (mc_remaining, mc_should, _) = \
5209
          self.cfg.GetMasterCandidateStats(exceptions=[node.name])
5210
      if mc_remaining < mc_should:
5211
        raise errors.OpPrereqError("Not enough master candidates, please"
5212
                                   " pass auto promote option to allow"
5213
                                   " promotion", errors.ECODE_STATE)
5214

    
5215
    self.old_flags = old_flags = (node.master_candidate,
5216
                                  node.drained, node.offline)
5217
    assert old_flags in self._F2R, "Un-handled old flags %s" % str(old_flags)
5218
    self.old_role = old_role = self._F2R[old_flags]
5219

    
5220
    # Check for ineffective changes
5221
    for attr in self._FLAGS:
5222
      if (getattr(self.op, attr) == False and getattr(node, attr) == False):
5223
        self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
5224
        setattr(self.op, attr, None)
5225

    
5226
    # Past this point, any flag change to False means a transition
5227
    # away from the respective state, as only real changes are kept
5228

    
5229
    # TODO: We might query the real power state if it supports OOB
5230
    if _SupportsOob(self.cfg, node):
5231
      if self.op.offline is False and not (node.powered or
5232
                                           self.op.powered == True):
5233
        raise errors.OpPrereqError(("Node %s needs to be turned on before its"
5234
                                    " offline status can be reset") %
5235
                                   self.op.node_name)
5236
    elif self.op.powered is not None:
5237
      raise errors.OpPrereqError(("Unable to change powered state for node %s"
5238
                                  " as it does not support out-of-band"
5239
                                  " handling") % self.op.node_name)
5240

    
5241
    # If we're being deofflined/drained, we'll MC ourself if needed
5242
    if (self.op.drained == False or self.op.offline == False or
5243
        (self.op.master_capable and not node.master_capable)):
5244
      if _DecideSelfPromotion(self):
5245
        self.op.master_candidate = True
5246
        self.LogInfo("Auto-promoting node to master candidate")
5247

    
5248
    # If we're no longer master capable, we'll demote ourselves from MC
5249
    if self.op.master_capable == False and node.master_candidate:
5250
      self.LogInfo("Demoting from master candidate")
5251
      self.op.master_candidate = False
5252

    
5253
    # Compute new role
5254
    assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
5255
    if self.op.master_candidate:
5256
      new_role = self._ROLE_CANDIDATE
5257
    elif self.op.drained:
5258
      new_role = self._ROLE_DRAINED
5259
    elif self.op.offline:
5260
      new_role = self._ROLE_OFFLINE
5261
    elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
5262
      # False is still in new flags, which means we're un-setting (the
5263
      # only) True flag
5264
      new_role = self._ROLE_REGULAR
5265
    else: # no new flags, nothing, keep old role
5266
      new_role = old_role
5267

    
5268
    self.new_role = new_role
5269

    
5270
    if old_role == self._ROLE_OFFLINE and new_role != old_role:
5271
      # Trying to transition out of offline status
5272
      result = self.rpc.call_version([node.name])[node.name]
5273
      if result.fail_msg:
5274
        raise errors.OpPrereqError("Node %s is being de-offlined but fails"
5275
                                   " to report its version: %s" %
5276
                                   (node.name, result.fail_msg),
5277
                                   errors.ECODE_STATE)
5278
      else:
5279
        self.LogWarning("Transitioning node from offline to online state"
5280
                        " without using re-add. Please make sure the node"
5281
                        " is healthy!")
5282

    
5283
    if self.op.secondary_ip:
5284
      # Ok even without locking, because this can't be changed by any LU
5285
      master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
5286
      master_singlehomed = master.secondary_ip == master.primary_ip
5287
      if master_singlehomed and self.op.secondary_ip:
5288
        raise errors.OpPrereqError("Cannot change the secondary ip on a single"
5289
                                   " homed cluster", errors.ECODE_INVAL)
5290

    
5291
      if node.offline:
5292
        if self.affected_instances:
5293
          raise errors.OpPrereqError("Cannot change secondary ip: offline"
5294
                                     " node has instances (%s) configured"
5295
                                     " to use it" % self.affected_instances)
5296
      else:
5297
        # On online nodes, check that no instances are running, and that
5298
        # the node has the new ip and we can reach it.
5299
        for instance in self.affected_instances:
5300
          _CheckInstanceDown(self, instance, "cannot change secondary ip")
5301

    
5302
        _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
5303
        if master.name != node.name:
5304
          # check reachability from master secondary ip to new secondary ip
5305
          if not netutils.TcpPing(self.op.secondary_ip,
5306
                                  constants.DEFAULT_NODED_PORT,
5307
                                  source=master.secondary_ip):
5308
            raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5309
                                       " based ping to node daemon port",
5310
                                       errors.ECODE_ENVIRON)
5311

    
5312
    if self.op.ndparams:
5313
      new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
5314
      utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
5315
      self.new_ndparams = new_ndparams
5316

    
5317
  def Exec(self, feedback_fn):
5318
    """Modifies a node.
5319

5320
    """
5321
    node = self.node
5322
    old_role = self.old_role
5323
    new_role = self.new_role
5324

    
5325
    result = []
5326

    
5327
    if self.op.ndparams:
5328
      node.ndparams = self.new_ndparams
5329

    
5330
    if self.op.powered is not None:
5331
      node.powered = self.op.powered
5332

    
5333
    for attr in ["master_capable", "vm_capable"]:
5334
      val = getattr(self.op, attr)
5335
      if val is not None:
5336
        setattr(node, attr, val)
5337
        result.append((attr, str(val)))
5338

    
5339
    if new_role != old_role:
5340
      # Tell the node to demote itself, if no longer MC and not offline
5341
      if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
5342
        msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
5343
        if msg:
5344
          self.LogWarning("Node failed to demote itself: %s", msg)
5345

    
5346
      new_flags = self._R2F[new_role]
5347
      for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
5348
        if of != nf:
5349
          result.append((desc, str(nf)))
5350
      (node.master_candidate, node.drained, node.offline) = new_flags
5351

    
5352
      # we locked all nodes, we adjust the CP before updating this node
5353
      if self.lock_all:
5354
        _AdjustCandidatePool(self, [node.name])
5355

    
5356
    if self.op.secondary_ip:
5357
      node.secondary_ip = self.op.secondary_ip
5358
      result.append(("secondary_ip", self.op.secondary_ip))
5359

    
5360
    # this will trigger configuration file update, if needed
5361
    self.cfg.Update(node, feedback_fn)
5362

    
5363
    # this will trigger job queue propagation or cleanup if the mc
5364
    # flag changed
5365
    if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
5366
      self.context.ReaddNode(node)
5367

    
5368
    return result
5369

    
5370

    
5371
class LUNodePowercycle(NoHooksLU):
5372
  """Powercycles a node.
5373

5374
  """
5375
  REQ_BGL = False
5376

    
5377
  def CheckArguments(self):
5378
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5379
    if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
5380
      raise errors.OpPrereqError("The node is the master and the force"
5381
                                 " parameter was not set",
5382
                                 errors.ECODE_INVAL)
5383

    
5384
  def ExpandNames(self):
5385
    """Locking for PowercycleNode.
5386

5387
    This is a last-resort option and shouldn't block on other
5388
    jobs. Therefore, we grab no locks.
5389

5390
    """
5391
    self.needed_locks = {}
5392

    
5393
  def Exec(self, feedback_fn):
5394
    """Reboots a node.
5395

5396
    """
5397
    result = self.rpc.call_node_powercycle(self.op.node_name,
5398
                                           self.cfg.GetHypervisorType())
5399
    result.Raise("Failed to schedule the reboot")
5400
    return result.payload
5401

    
5402

    
5403
class LUClusterQuery(NoHooksLU):
5404
  """Query cluster configuration.
5405

5406
  """
5407
  REQ_BGL = False
5408

    
5409
  def ExpandNames(self):
5410
    self.needed_locks = {}
5411

    
5412
  def Exec(self, feedback_fn):
5413
    """Return cluster config.
5414

5415
    """
5416
    cluster = self.cfg.GetClusterInfo()
5417
    os_hvp = {}
5418

    
5419
    # Filter just for enabled hypervisors
5420
    for os_name, hv_dict in cluster.os_hvp.items():
5421
      os_hvp[os_name] = {}
5422
      for hv_name, hv_params in hv_dict.items():
5423
        if hv_name in cluster.enabled_hypervisors:
5424
          os_hvp[os_name][hv_name] = hv_params
5425

    
5426
    # Convert ip_family to ip_version
5427
    primary_ip_version = constants.IP4_VERSION
5428
    if cluster.primary_ip_family == netutils.IP6Address.family:
5429
      primary_ip_version = constants.IP6_VERSION
5430

    
5431
    result = {
5432
      "software_version": constants.RELEASE_VERSION,
5433
      "protocol_version": constants.PROTOCOL_VERSION,
5434
      "config_version": constants.CONFIG_VERSION,
5435
      "os_api_version": max(constants.OS_API_VERSIONS),
5436
      "export_version": constants.EXPORT_VERSION,
5437
      "architecture": (platform.architecture()[0], platform.machine()),
5438
      "name": cluster.cluster_name,
5439
      "master": cluster.master_node,
5440
      "default_hypervisor": cluster.enabled_hypervisors[0],
5441
      "enabled_hypervisors": cluster.enabled_hypervisors,
5442
      "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
5443
                        for hypervisor_name in cluster.enabled_hypervisors]),
5444
      "os_hvp": os_hvp,
5445
      "beparams": cluster.beparams,
5446
      "osparams": cluster.osparams,
5447
      "nicparams": cluster.nicparams,
5448
      "ndparams": cluster.ndparams,
5449
      "candidate_pool_size": cluster.candidate_pool_size,
5450
      "master_netdev": cluster.master_netdev,
5451
      "volume_group_name": cluster.volume_group_name,
5452
      "drbd_usermode_helper": cluster.drbd_usermode_helper,
5453
      "file_storage_dir": cluster.file_storage_dir,
5454
      "shared_file_storage_dir": cluster.shared_file_storage_dir,
5455
      "maintain_node_health": cluster.maintain_node_health,
5456
      "ctime": cluster.ctime,
5457
      "mtime": cluster.mtime,
5458
      "uuid": cluster.uuid,
5459
      "tags": list(cluster.GetTags()),
5460
      "uid_pool": cluster.uid_pool,
5461
      "default_iallocator": cluster.default_iallocator,
5462
      "reserved_lvs": cluster.reserved_lvs,
5463
      "primary_ip_version": primary_ip_version,
5464
      "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
5465
      "hidden_os": cluster.hidden_os,
5466
      "blacklisted_os": cluster.blacklisted_os,
5467
      }
5468

    
5469
    return result
5470

    
5471

    
5472
class LUClusterConfigQuery(NoHooksLU):
5473
  """Return configuration values.
5474

5475
  """
5476
  REQ_BGL = False
5477
  _FIELDS_DYNAMIC = utils.FieldSet()
5478
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag",
5479
                                  "watcher_pause", "volume_group_name")
5480

    
5481
  def CheckArguments(self):
5482
    _CheckOutputFields(static=self._FIELDS_STATIC,
5483
                       dynamic=self._FIELDS_DYNAMIC,
5484
                       selected=self.op.output_fields)
5485

    
5486
  def ExpandNames(self):
5487
    self.needed_locks = {}
5488

    
5489
  def Exec(self, feedback_fn):
5490
    """Dump a representation of the cluster config to the standard output.
5491

5492
    """
5493
    values = []
5494
    for field in self.op.output_fields:
5495
      if field == "cluster_name":
5496
        entry = self.cfg.GetClusterName()
5497
      elif field == "master_node":
5498
        entry = self.cfg.GetMasterNode()
5499
      elif field == "drain_flag":
5500
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
5501
      elif field == "watcher_pause":
5502
        entry = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
5503
      elif field == "volume_group_name":
5504
        entry = self.cfg.GetVGName()
5505
      else:
5506
        raise errors.ParameterError(field)
5507
      values.append(entry)
5508
    return values
5509

    
5510

    
5511
class LUInstanceActivateDisks(NoHooksLU):
5512
  """Bring up an instance's disks.
5513

5514
  """
5515
  REQ_BGL = False
5516

    
5517
  def ExpandNames(self):
5518
    self._ExpandAndLockInstance()
5519
    self.needed_locks[locking.LEVEL_NODE] = []
5520
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5521

    
5522
  def DeclareLocks(self, level):
5523
    if level == locking.LEVEL_NODE:
5524
      self._LockInstancesNodes()
5525

    
5526
  def CheckPrereq(self):
5527
    """Check prerequisites.
5528

5529
    This checks that the instance is in the cluster.
5530

5531
    """
5532
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5533
    assert self.instance is not None, \
5534
      "Cannot retrieve locked instance %s" % self.op.instance_name
5535
    _CheckNodeOnline(self, self.instance.primary_node)
5536

    
5537
  def Exec(self, feedback_fn):
5538
    """Activate the disks.
5539

5540
    """
5541
    disks_ok, disks_info = \
5542
              _AssembleInstanceDisks(self, self.instance,
5543
                                     ignore_size=self.op.ignore_size)
5544
    if not disks_ok:
5545
      raise errors.OpExecError("Cannot activate block devices")
5546

    
5547
    return disks_info
5548

    
5549

    
5550
def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
5551
                           ignore_size=False):
5552
  """Prepare the block devices for an instance.
5553

5554
  This sets up the block devices on all nodes.
5555

5556
  @type lu: L{LogicalUnit}
5557
  @param lu: the logical unit on whose behalf we execute
5558
  @type instance: L{objects.Instance}
5559
  @param instance: the instance for whose disks we assemble
5560
  @type disks: list of L{objects.Disk} or None
5561
  @param disks: which disks to assemble (or all, if None)
5562
  @type ignore_secondaries: boolean
5563
  @param ignore_secondaries: if true, errors on secondary nodes
5564
      won't result in an error return from the function
5565
  @type ignore_size: boolean
5566
  @param ignore_size: if true, the current known size of the disk
5567
      will not be used during the disk activation, useful for cases
5568
      when the size is wrong
5569
  @return: False if the operation failed, otherwise a list of
5570
      (host, instance_visible_name, node_visible_name)
5571
      with the mapping from node devices to instance devices
5572

5573
  """
5574
  device_info = []
5575
  disks_ok = True
5576
  iname = instance.name
5577
  disks = _ExpandCheckDisks(instance, disks)
5578

    
5579
  # With the two passes mechanism we try to reduce the window of
5580
  # opportunity for the race condition of switching DRBD to primary
5581
  # before handshaking occured, but we do not eliminate it
5582

    
5583
  # The proper fix would be to wait (with some limits) until the
5584
  # connection has been made and drbd transitions from WFConnection
5585
  # into any other network-connected state (Connected, SyncTarget,
5586
  # SyncSource, etc.)
5587

    
5588
  # 1st pass, assemble on all nodes in secondary mode
5589
  for idx, inst_disk in enumerate(disks):
5590
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
5591
      if ignore_size:
5592
        node_disk = node_disk.Copy()
5593
        node_disk.UnsetSize()
5594
      lu.cfg.SetDiskID(node_disk, node)
5595
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False, idx)
5596
      msg = result.fail_msg
5597
      if msg:
5598
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
5599
                           " (is_primary=False, pass=1): %s",
5600
                           inst_disk.iv_name, node, msg)
5601
        if not ignore_secondaries:
5602
          disks_ok = False
5603

    
5604
  # FIXME: race condition on drbd migration to primary
5605

    
5606
  # 2nd pass, do only the primary node
5607
  for idx, inst_disk in enumerate(disks):
5608
    dev_path = None
5609

    
5610
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
5611
      if node != instance.primary_node:
5612
        continue
5613
      if ignore_size:
5614
        node_disk = node_disk.Copy()
5615
        node_disk.UnsetSize()
5616
      lu.cfg.SetDiskID(node_disk, node)
5617
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True, idx)
5618
      msg = result.fail_msg
5619
      if msg:
5620
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
5621
                           " (is_primary=True, pass=2): %s",
5622
                           inst_disk.iv_name, node, msg)
5623
        disks_ok = False
5624
      else:
5625
        dev_path = result.payload
5626

    
5627
    device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
5628

    
5629
  # leave the disks configured for the primary node
5630
  # this is a workaround that would be fixed better by
5631
  # improving the logical/physical id handling
5632
  for disk in disks:
5633
    lu.cfg.SetDiskID(disk, instance.primary_node)
5634

    
5635
  return disks_ok, device_info
5636

    
5637

    
5638
def _StartInstanceDisks(lu, instance, force):
5639
  """Start the disks of an instance.
5640

5641
  """
5642
  disks_ok, _ = _AssembleInstanceDisks(lu, instance,
5643
                                           ignore_secondaries=force)
5644
  if not disks_ok:
5645
    _ShutdownInstanceDisks(lu, instance)
5646
    if force is not None and not force:
5647
      lu.proc.LogWarning("", hint="If the message above refers to a"
5648
                         " secondary node,"
5649
                         " you can retry the operation using '--force'.")
5650
    raise errors.OpExecError("Disk consistency error")
5651

    
5652

    
5653
class LUInstanceDeactivateDisks(NoHooksLU):
5654
  """Shutdown an instance's disks.
5655

5656
  """
5657
  REQ_BGL = False
5658

    
5659
  def ExpandNames(self):
5660
    self._ExpandAndLockInstance()
5661
    self.needed_locks[locking.LEVEL_NODE] = []
5662
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5663

    
5664
  def DeclareLocks(self, level):
5665
    if level == locking.LEVEL_NODE:
5666
      self._LockInstancesNodes()
5667

    
5668
  def CheckPrereq(self):
5669
    """Check prerequisites.
5670

5671
    This checks that the instance is in the cluster.
5672

5673
    """
5674
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5675
    assert self.instance is not None, \
5676
      "Cannot retrieve locked instance %s" % self.op.instance_name
5677

    
5678
  def Exec(self, feedback_fn):
5679
    """Deactivate the disks
5680

5681
    """
5682
    instance = self.instance
5683
    if self.op.force:
5684
      _ShutdownInstanceDisks(self, instance)
5685
    else:
5686
      _SafeShutdownInstanceDisks(self, instance)
5687

    
5688

    
5689
def _SafeShutdownInstanceDisks(lu, instance, disks=None):
5690
  """Shutdown block devices of an instance.
5691

5692
  This function checks if an instance is running, before calling
5693
  _ShutdownInstanceDisks.
5694

5695
  """
5696
  _CheckInstanceDown(lu, instance, "cannot shutdown disks")
5697
  _ShutdownInstanceDisks(lu, instance, disks=disks)
5698

    
5699

    
5700
def _ExpandCheckDisks(instance, disks):
5701
  """Return the instance disks selected by the disks list
5702

5703
  @type disks: list of L{objects.Disk} or None
5704
  @param disks: selected disks
5705
  @rtype: list of L{objects.Disk}
5706
  @return: selected instance disks to act on
5707

5708
  """
5709
  if disks is None:
5710
    return instance.disks
5711
  else:
5712
    if not set(disks).issubset(instance.disks):
5713
      raise errors.ProgrammerError("Can only act on disks belonging to the"
5714
                                   " target instance")
5715
    return disks
5716

    
5717

    
5718
def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
5719
  """Shutdown block devices of an instance.
5720

5721
  This does the shutdown on all nodes of the instance.
5722

5723
  If the ignore_primary is false, errors on the primary node are
5724
  ignored.
5725

5726
  """
5727
  all_result = True
5728
  disks = _ExpandCheckDisks(instance, disks)
5729

    
5730
  for disk in disks:
5731
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
5732
      lu.cfg.SetDiskID(top_disk, node)
5733
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
5734
      msg = result.fail_msg
5735
      if msg:
5736
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
5737
                      disk.iv_name, node, msg)
5738
        if ((node == instance.primary_node and not ignore_primary) or
5739
            (node != instance.primary_node and not result.offline)):
5740
          all_result = False
5741
  return all_result
5742

    
5743

    
5744
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
5745
  """Checks if a node has enough free memory.
5746

5747
  This function check if a given node has the needed amount of free
5748
  memory. In case the node has less memory or we cannot get the
5749
  information from the node, this function raise an OpPrereqError
5750
  exception.
5751

5752
  @type lu: C{LogicalUnit}
5753
  @param lu: a logical unit from which we get configuration data
5754
  @type node: C{str}
5755
  @param node: the node to check
5756
  @type reason: C{str}
5757
  @param reason: string to use in the error message
5758
  @type requested: C{int}
5759
  @param requested: the amount of memory in MiB to check for
5760
  @type hypervisor_name: C{str}
5761
  @param hypervisor_name: the hypervisor to ask for memory stats
5762
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
5763
      we cannot check the node
5764

5765
  """
5766
  nodeinfo = lu.rpc.call_node_info([node], None, hypervisor_name)
5767
  nodeinfo[node].Raise("Can't get data from node %s" % node,
5768
                       prereq=True, ecode=errors.ECODE_ENVIRON)
5769
  free_mem = nodeinfo[node].payload.get("memory_free", None)
5770
  if not isinstance(free_mem, int):
5771
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
5772
                               " was '%s'" % (node, free_mem),
5773
                               errors.ECODE_ENVIRON)
5774
  if requested > free_mem:
5775
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
5776
                               " needed %s MiB, available %s MiB" %
5777
                               (node, reason, requested, free_mem),
5778
                               errors.ECODE_NORES)
5779

    
5780

    
5781
def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
5782
  """Checks if nodes have enough free disk space in the all VGs.
5783

5784
  This function check if all given nodes have the needed amount of
5785
  free disk. In case any node has less disk or we cannot get the
5786
  information from the node, this function raise an OpPrereqError
5787
  exception.
5788

5789
  @type lu: C{LogicalUnit}
5790
  @param lu: a logical unit from which we get configuration data
5791
  @type nodenames: C{list}
5792
  @param nodenames: the list of node names to check
5793
  @type req_sizes: C{dict}
5794
  @param req_sizes: the hash of vg and corresponding amount of disk in
5795
      MiB to check for
5796
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5797
      or we cannot check the node
5798

5799
  """
5800
  for vg, req_size in req_sizes.items():
5801
    _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
5802

    
5803

    
5804
def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
5805
  """Checks if nodes have enough free disk space in the specified VG.
5806

5807
  This function check if all given nodes have the needed amount of
5808
  free disk. In case any node has less disk or we cannot get the
5809
  information from the node, this function raise an OpPrereqError
5810
  exception.
5811

5812
  @type lu: C{LogicalUnit}
5813
  @param lu: a logical unit from which we get configuration data
5814
  @type nodenames: C{list}
5815
  @param nodenames: the list of node names to check
5816
  @type vg: C{str}
5817
  @param vg: the volume group to check
5818
  @type requested: C{int}
5819
  @param requested: the amount of disk in MiB to check for
5820
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5821
      or we cannot check the node
5822

5823
  """
5824
  nodeinfo = lu.rpc.call_node_info(nodenames, vg, None)
5825
  for node in nodenames:
5826
    info = nodeinfo[node]
5827
    info.Raise("Cannot get current information from node %s" % node,
5828
               prereq=True, ecode=errors.ECODE_ENVIRON)
5829
    vg_free = info.payload.get("vg_free", None)
5830
    if not isinstance(vg_free, int):
5831
      raise errors.OpPrereqError("Can't compute free disk space on node"
5832
                                 " %s for vg %s, result was '%s'" %
5833
                                 (node, vg, vg_free), errors.ECODE_ENVIRON)
5834
    if requested > vg_free:
5835
      raise errors.OpPrereqError("Not enough disk space on target node %s"
5836
                                 " vg %s: required %d MiB, available %d MiB" %
5837
                                 (node, vg, requested, vg_free),
5838
                                 errors.ECODE_NORES)
5839

    
5840

    
5841
def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):
5842
  """Checks if nodes have enough physical CPUs
5843

5844
  This function checks if all given nodes have the needed number of
5845
  physical CPUs. In case any node has less CPUs or we cannot get the
5846
  information from the node, this function raises an OpPrereqError
5847
  exception.
5848

5849
  @type lu: C{LogicalUnit}
5850
  @param lu: a logical unit from which we get configuration data
5851
  @type nodenames: C{list}
5852
  @param nodenames: the list of node names to check
5853
  @type requested: C{int}
5854
  @param requested: the minimum acceptable number of physical CPUs
5855
  @raise errors.OpPrereqError: if the node doesn't have enough CPUs,
5856
      or we cannot check the node
5857

5858
  """
5859
  nodeinfo = lu.rpc.call_node_info(nodenames, None, hypervisor_name)
5860
  for node in nodenames:
5861
    info = nodeinfo[node]
5862
    info.Raise("Cannot get current information from node %s" % node,
5863
               prereq=True, ecode=errors.ECODE_ENVIRON)
5864
    num_cpus = info.payload.get("cpu_total", None)
5865
    if not isinstance(num_cpus, int):
5866
      raise errors.OpPrereqError("Can't compute the number of physical CPUs"
5867
                                 " on node %s, result was '%s'" %
5868
                                 (node, num_cpus), errors.ECODE_ENVIRON)
5869
    if requested > num_cpus:
5870
      raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "
5871
                                 "required" % (node, num_cpus, requested),
5872
                                 errors.ECODE_NORES)
5873

    
5874

    
5875
class LUInstanceStartup(LogicalUnit):
5876
  """Starts an instance.
5877

5878
  """
5879
  HPATH = "instance-start"
5880
  HTYPE = constants.HTYPE_INSTANCE
5881
  REQ_BGL = False
5882

    
5883
  def CheckArguments(self):
5884
    # extra beparams
5885
    if self.op.beparams:
5886
      # fill the beparams dict
5887
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
5888

    
5889
  def ExpandNames(self):
5890
    self._ExpandAndLockInstance()
5891

    
5892
  def BuildHooksEnv(self):
5893
    """Build hooks env.
5894

5895
    This runs on master, primary and secondary nodes of the instance.
5896

5897
    """
5898
    env = {
5899
      "FORCE": self.op.force,
5900
      }
5901

    
5902
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5903

    
5904
    return env
5905

    
5906
  def BuildHooksNodes(self):
5907
    """Build hooks nodes.
5908

5909
    """
5910
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5911
    return (nl, nl)
5912

    
5913
  def CheckPrereq(self):
5914
    """Check prerequisites.
5915

5916
    This checks that the instance is in the cluster.
5917

5918
    """
5919
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5920
    assert self.instance is not None, \
5921
      "Cannot retrieve locked instance %s" % self.op.instance_name
5922

    
5923
    # extra hvparams
5924
    if self.op.hvparams:
5925
      # check hypervisor parameter syntax (locally)
5926
      cluster = self.cfg.GetClusterInfo()
5927
      utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
5928
      filled_hvp = cluster.FillHV(instance)
5929
      filled_hvp.update(self.op.hvparams)
5930
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
5931
      hv_type.CheckParameterSyntax(filled_hvp)
5932
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
5933

    
5934
    self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
5935

    
5936
    if self.primary_offline and self.op.ignore_offline_nodes:
5937
      self.proc.LogWarning("Ignoring offline primary node")
5938

    
5939
      if self.op.hvparams or self.op.beparams:
5940
        self.proc.LogWarning("Overridden parameters are ignored")
5941
    else:
5942
      _CheckNodeOnline(self, instance.primary_node)
5943

    
5944
      bep = self.cfg.GetClusterInfo().FillBE(instance)
5945

    
5946
      # check bridges existence
5947
      _CheckInstanceBridgesExist(self, instance)
5948

    
5949
      remote_info = self.rpc.call_instance_info(instance.primary_node,
5950
                                                instance.name,
5951
                                                instance.hypervisor)
5952
      remote_info.Raise("Error checking node %s" % instance.primary_node,
5953
                        prereq=True, ecode=errors.ECODE_ENVIRON)
5954
      if not remote_info.payload: # not running already
5955
        _CheckNodeFreeMemory(self, instance.primary_node,
5956
                             "starting instance %s" % instance.name,
5957
                             bep[constants.BE_MEMORY], instance.hypervisor)
5958

    
5959
  def Exec(self, feedback_fn):
5960
    """Start the instance.
5961

5962
    """
5963
    instance = self.instance
5964
    force = self.op.force
5965

    
5966
    if not self.op.no_remember:
5967
      self.cfg.MarkInstanceUp(instance.name)
5968

    
5969
    if self.primary_offline:
5970
      assert self.op.ignore_offline_nodes
5971
      self.proc.LogInfo("Primary node offline, marked instance as started")
5972
    else:
5973
      node_current = instance.primary_node
5974

    
5975
      _StartInstanceDisks(self, instance, force)
5976

    
5977
      result = self.rpc.call_instance_start(node_current, instance,
5978
                                            self.op.hvparams, self.op.beparams,
5979
                                            self.op.startup_paused)
5980
      msg = result.fail_msg
5981
      if msg:
5982
        _ShutdownInstanceDisks(self, instance)
5983
        raise errors.OpExecError("Could not start instance: %s" % msg)
5984

    
5985

    
5986
class LUInstanceReboot(LogicalUnit):
5987
  """Reboot an instance.
5988

5989
  """
5990
  HPATH = "instance-reboot"
5991
  HTYPE = constants.HTYPE_INSTANCE
5992
  REQ_BGL = False
5993

    
5994
  def ExpandNames(self):
5995
    self._ExpandAndLockInstance()
5996

    
5997
  def BuildHooksEnv(self):
5998
    """Build hooks env.
5999

6000
    This runs on master, primary and secondary nodes of the instance.
6001

6002
    """
6003
    env = {
6004
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
6005
      "REBOOT_TYPE": self.op.reboot_type,
6006
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6007
      }
6008

    
6009
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6010

    
6011
    return env
6012

    
6013
  def BuildHooksNodes(self):
6014
    """Build hooks nodes.
6015

6016
    """
6017
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6018
    return (nl, nl)
6019

    
6020
  def CheckPrereq(self):
6021
    """Check prerequisites.
6022

6023
    This checks that the instance is in the cluster.
6024

6025
    """
6026
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6027
    assert self.instance is not None, \
6028
      "Cannot retrieve locked instance %s" % self.op.instance_name
6029

    
6030
    _CheckNodeOnline(self, instance.primary_node)
6031

    
6032
    # check bridges existence
6033
    _CheckInstanceBridgesExist(self, instance)
6034

    
6035
  def Exec(self, feedback_fn):
6036
    """Reboot the instance.
6037

6038
    """
6039
    instance = self.instance
6040
    ignore_secondaries = self.op.ignore_secondaries
6041
    reboot_type = self.op.reboot_type
6042

    
6043
    remote_info = self.rpc.call_instance_info(instance.primary_node,
6044
                                              instance.name,
6045
                                              instance.hypervisor)
6046
    remote_info.Raise("Error checking node %s" % instance.primary_node)
6047
    instance_running = bool(remote_info.payload)
6048

    
6049
    node_current = instance.primary_node
6050

    
6051
    if instance_running and reboot_type in [constants.INSTANCE_REBOOT_SOFT,
6052
                                            constants.INSTANCE_REBOOT_HARD]:
6053
      for disk in instance.disks:
6054
        self.cfg.SetDiskID(disk, node_current)
6055
      result = self.rpc.call_instance_reboot(node_current, instance,
6056
                                             reboot_type,
6057
                                             self.op.shutdown_timeout)
6058
      result.Raise("Could not reboot instance")
6059
    else:
6060
      if instance_running:
6061
        result = self.rpc.call_instance_shutdown(node_current, instance,
6062
                                                 self.op.shutdown_timeout)
6063
        result.Raise("Could not shutdown instance for full reboot")
6064
        _ShutdownInstanceDisks(self, instance)
6065
      else:
6066
        self.LogInfo("Instance %s was already stopped, starting now",
6067
                     instance.name)
6068
      _StartInstanceDisks(self, instance, ignore_secondaries)
6069
      result = self.rpc.call_instance_start(node_current, instance,
6070
                                            None, None, False)
6071
      msg = result.fail_msg
6072
      if msg:
6073
        _ShutdownInstanceDisks(self, instance)
6074
        raise errors.OpExecError("Could not start instance for"
6075
                                 " full reboot: %s" % msg)
6076

    
6077
    self.cfg.MarkInstanceUp(instance.name)
6078

    
6079

    
6080
class LUInstanceShutdown(LogicalUnit):
6081
  """Shutdown an instance.
6082

6083
  """
6084
  HPATH = "instance-stop"
6085
  HTYPE = constants.HTYPE_INSTANCE
6086
  REQ_BGL = False
6087

    
6088
  def ExpandNames(self):
6089
    self._ExpandAndLockInstance()
6090

    
6091
  def BuildHooksEnv(self):
6092
    """Build hooks env.
6093

6094
    This runs on master, primary and secondary nodes of the instance.
6095

6096
    """
6097
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6098
    env["TIMEOUT"] = self.op.timeout
6099
    return env
6100

    
6101
  def BuildHooksNodes(self):
6102
    """Build hooks nodes.
6103

6104
    """
6105
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6106
    return (nl, nl)
6107

    
6108
  def CheckPrereq(self):
6109
    """Check prerequisites.
6110

6111
    This checks that the instance is in the cluster.
6112

6113
    """
6114
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6115
    assert self.instance is not None, \
6116
      "Cannot retrieve locked instance %s" % self.op.instance_name
6117

    
6118
    self.primary_offline = \
6119
      self.cfg.GetNodeInfo(self.instance.primary_node).offline
6120

    
6121
    if self.primary_offline and self.op.ignore_offline_nodes:
6122
      self.proc.LogWarning("Ignoring offline primary node")
6123
    else:
6124
      _CheckNodeOnline(self, self.instance.primary_node)
6125

    
6126
  def Exec(self, feedback_fn):
6127
    """Shutdown the instance.
6128

6129
    """
6130
    instance = self.instance
6131
    node_current = instance.primary_node
6132
    timeout = self.op.timeout
6133

    
6134
    if not self.op.no_remember:
6135
      self.cfg.MarkInstanceDown(instance.name)
6136

    
6137
    if self.primary_offline:
6138
      assert self.op.ignore_offline_nodes
6139
      self.proc.LogInfo("Primary node offline, marked instance as stopped")
6140
    else:
6141
      result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
6142
      msg = result.fail_msg
6143
      if msg:
6144
        self.proc.LogWarning("Could not shutdown instance: %s" % msg)
6145

    
6146
      _ShutdownInstanceDisks(self, instance)
6147

    
6148

    
6149
class LUInstanceReinstall(LogicalUnit):
6150
  """Reinstall an instance.
6151

6152
  """
6153
  HPATH = "instance-reinstall"
6154
  HTYPE = constants.HTYPE_INSTANCE
6155
  REQ_BGL = False
6156

    
6157
  def ExpandNames(self):
6158
    self._ExpandAndLockInstance()
6159

    
6160
  def BuildHooksEnv(self):
6161
    """Build hooks env.
6162

6163
    This runs on master, primary and secondary nodes of the instance.
6164

6165
    """
6166
    return _BuildInstanceHookEnvByObject(self, self.instance)
6167

    
6168
  def BuildHooksNodes(self):
6169
    """Build hooks nodes.
6170

6171
    """
6172
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6173
    return (nl, nl)
6174

    
6175
  def CheckPrereq(self):
6176
    """Check prerequisites.
6177

6178
    This checks that the instance is in the cluster and is not running.
6179

6180
    """
6181
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6182
    assert instance is not None, \
6183
      "Cannot retrieve locked instance %s" % self.op.instance_name
6184
    _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
6185
                     " offline, cannot reinstall")
6186
    for node in instance.secondary_nodes:
6187
      _CheckNodeOnline(self, node, "Instance secondary node offline,"
6188
                       " cannot reinstall")
6189

    
6190
    if instance.disk_template == constants.DT_DISKLESS:
6191
      raise errors.OpPrereqError("Instance '%s' has no disks" %
6192
                                 self.op.instance_name,
6193
                                 errors.ECODE_INVAL)
6194
    _CheckInstanceDown(self, instance, "cannot reinstall")
6195

    
6196
    if self.op.os_type is not None:
6197
      # OS verification
6198
      pnode = _ExpandNodeName(self.cfg, instance.primary_node)
6199
      _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
6200
      instance_os = self.op.os_type
6201
    else:
6202
      instance_os = instance.os
6203

    
6204
    nodelist = list(instance.all_nodes)
6205

    
6206
    if self.op.osparams:
6207
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
6208
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
6209
      self.os_inst = i_osdict # the new dict (without defaults)
6210
    else:
6211
      self.os_inst = None
6212

    
6213
    self.instance = instance
6214

    
6215
  def Exec(self, feedback_fn):
6216
    """Reinstall the instance.
6217

6218
    """
6219
    inst = self.instance
6220

    
6221
    if self.op.os_type is not None:
6222
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
6223
      inst.os = self.op.os_type
6224
      # Write to configuration
6225
      self.cfg.Update(inst, feedback_fn)
6226

    
6227
    _StartInstanceDisks(self, inst, None)
6228
    try:
6229
      feedback_fn("Running the instance OS create scripts...")
6230
      # FIXME: pass debug option from opcode to backend
6231
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True,
6232
                                             self.op.debug_level,
6233
                                             osparams=self.os_inst)
6234
      result.Raise("Could not install OS for instance %s on node %s" %
6235
                   (inst.name, inst.primary_node))
6236
    finally:
6237
      _ShutdownInstanceDisks(self, inst)
6238

    
6239

    
6240
class LUInstanceRecreateDisks(LogicalUnit):
6241
  """Recreate an instance's missing disks.
6242

6243
  """
6244
  HPATH = "instance-recreate-disks"
6245
  HTYPE = constants.HTYPE_INSTANCE
6246
  REQ_BGL = False
6247

    
6248
  def CheckArguments(self):
6249
    # normalise the disk list
6250
    self.op.disks = sorted(frozenset(self.op.disks))
6251

    
6252
  def ExpandNames(self):
6253
    self._ExpandAndLockInstance()
6254
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6255
    if self.op.nodes:
6256
      self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
6257
      self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
6258
    else:
6259
      self.needed_locks[locking.LEVEL_NODE] = []
6260

    
6261
  def DeclareLocks(self, level):
6262
    if level == locking.LEVEL_NODE:
6263
      # if we replace the nodes, we only need to lock the old primary,
6264
      # otherwise we need to lock all nodes for disk re-creation
6265
      primary_only = bool(self.op.nodes)
6266
      self._LockInstancesNodes(primary_only=primary_only)
6267

    
6268
  def BuildHooksEnv(self):
6269
    """Build hooks env.
6270

6271
    This runs on master, primary and secondary nodes of the instance.
6272

6273
    """
6274
    return _BuildInstanceHookEnvByObject(self, self.instance)
6275

    
6276
  def BuildHooksNodes(self):
6277
    """Build hooks nodes.
6278

6279
    """
6280
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6281
    return (nl, nl)
6282

    
6283
  def CheckPrereq(self):
6284
    """Check prerequisites.
6285

6286
    This checks that the instance is in the cluster and is not running.
6287

6288
    """
6289
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6290
    assert instance is not None, \
6291
      "Cannot retrieve locked instance %s" % self.op.instance_name
6292
    if self.op.nodes:
6293
      if len(self.op.nodes) != len(instance.all_nodes):
6294
        raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
6295
                                   " %d replacement nodes were specified" %
6296
                                   (instance.name, len(instance.all_nodes),
6297
                                    len(self.op.nodes)),
6298
                                   errors.ECODE_INVAL)
6299
      assert instance.disk_template != constants.DT_DRBD8 or \
6300
          len(self.op.nodes) == 2
6301
      assert instance.disk_template != constants.DT_PLAIN or \
6302
          len(self.op.nodes) == 1
6303
      primary_node = self.op.nodes[0]
6304
    else:
6305
      primary_node = instance.primary_node
6306
    _CheckNodeOnline(self, primary_node)
6307

    
6308
    if instance.disk_template == constants.DT_DISKLESS:
6309
      raise errors.OpPrereqError("Instance '%s' has no disks" %
6310
                                 self.op.instance_name, errors.ECODE_INVAL)
6311
    # if we replace nodes *and* the old primary is offline, we don't
6312
    # check
6313
    assert instance.primary_node in self.needed_locks[locking.LEVEL_NODE]
6314
    old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
6315
    if not (self.op.nodes and old_pnode.offline):
6316
      _CheckInstanceDown(self, instance, "cannot recreate disks")
6317

    
6318
    if not self.op.disks:
6319
      self.op.disks = range(len(instance.disks))
6320
    else:
6321
      for idx in self.op.disks:
6322
        if idx >= len(instance.disks):
6323
          raise errors.OpPrereqError("Invalid disk index '%s'" % idx,
6324
                                     errors.ECODE_INVAL)
6325
    if self.op.disks != range(len(instance.disks)) and self.op.nodes:
6326
      raise errors.OpPrereqError("Can't recreate disks partially and"
6327
                                 " change the nodes at the same time",
6328
                                 errors.ECODE_INVAL)
6329
    self.instance = instance
6330

    
6331
  def Exec(self, feedback_fn):
6332
    """Recreate the disks.
6333

6334
    """
6335
    instance = self.instance
6336

    
6337
    to_skip = []
6338
    mods = [] # keeps track of needed logical_id changes
6339

    
6340
    for idx, disk in enumerate(instance.disks):
6341
      if idx not in self.op.disks: # disk idx has not been passed in
6342
        to_skip.append(idx)
6343
        continue
6344
      # update secondaries for disks, if needed
6345
      if self.op.nodes:
6346
        if disk.dev_type == constants.LD_DRBD8:
6347
          # need to update the nodes and minors
6348
          assert len(self.op.nodes) == 2
6349
          assert len(disk.logical_id) == 6 # otherwise disk internals
6350
                                           # have changed
6351
          (_, _, old_port, _, _, old_secret) = disk.logical_id
6352
          new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
6353
          new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
6354
                    new_minors[0], new_minors[1], old_secret)
6355
          assert len(disk.logical_id) == len(new_id)
6356
          mods.append((idx, new_id))
6357

    
6358
    # now that we have passed all asserts above, we can apply the mods
6359
    # in a single run (to avoid partial changes)
6360
    for idx, new_id in mods:
6361
      instance.disks[idx].logical_id = new_id
6362

    
6363
    # change primary node, if needed
6364
    if self.op.nodes:
6365
      instance.primary_node = self.op.nodes[0]
6366
      self.LogWarning("Changing the instance's nodes, you will have to"
6367
                      " remove any disks left on the older nodes manually")
6368

    
6369
    if self.op.nodes:
6370
      self.cfg.Update(instance, feedback_fn)
6371

    
6372
    _CreateDisks(self, instance, to_skip=to_skip)
6373

    
6374

    
6375
class LUInstanceRename(LogicalUnit):
6376
  """Rename an instance.
6377

6378
  """
6379
  HPATH = "instance-rename"
6380
  HTYPE = constants.HTYPE_INSTANCE
6381

    
6382
  def CheckArguments(self):
6383
    """Check arguments.
6384

6385
    """
6386
    if self.op.ip_check and not self.op.name_check:
6387
      # TODO: make the ip check more flexible and not depend on the name check
6388
      raise errors.OpPrereqError("IP address check requires a name check",
6389
                                 errors.ECODE_INVAL)
6390

    
6391
  def BuildHooksEnv(self):
6392
    """Build hooks env.
6393

6394
    This runs on master, primary and secondary nodes of the instance.
6395

6396
    """
6397
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6398
    env["INSTANCE_NEW_NAME"] = self.op.new_name
6399
    return env
6400

    
6401
  def BuildHooksNodes(self):
6402
    """Build hooks nodes.
6403

6404
    """
6405
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6406
    return (nl, nl)
6407

    
6408
  def CheckPrereq(self):
6409
    """Check prerequisites.
6410

6411
    This checks that the instance is in the cluster and is not running.
6412

6413
    """
6414
    self.op.instance_name = _ExpandInstanceName(self.cfg,
6415
                                                self.op.instance_name)
6416
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6417
    assert instance is not None
6418
    _CheckNodeOnline(self, instance.primary_node)
6419
    _CheckInstanceDown(self, instance, "cannot rename")
6420
    self.instance = instance
6421

    
6422
    new_name = self.op.new_name
6423
    if self.op.name_check:
6424
      hostname = netutils.GetHostname(name=new_name)
6425
      if hostname != new_name:
6426
        self.LogInfo("Resolved given name '%s' to '%s'", new_name,
6427
                     hostname.name)
6428
      if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
6429
        raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
6430
                                    " same as given hostname '%s'") %
6431
                                    (hostname.name, self.op.new_name),
6432
                                    errors.ECODE_INVAL)
6433
      new_name = self.op.new_name = hostname.name
6434
      if (self.op.ip_check and
6435
          netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
6436
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
6437
                                   (hostname.ip, new_name),
6438
                                   errors.ECODE_NOTUNIQUE)
6439

    
6440
    instance_list = self.cfg.GetInstanceList()
6441
    if new_name in instance_list and new_name != instance.name:
6442
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
6443
                                 new_name, errors.ECODE_EXISTS)
6444

    
6445
  def Exec(self, feedback_fn):
6446
    """Rename the instance.
6447

6448
    """
6449
    inst = self.instance
6450
    old_name = inst.name
6451

    
6452
    rename_file_storage = False
6453
    if (inst.disk_template in constants.DTS_FILEBASED and
6454
        self.op.new_name != inst.name):
6455
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
6456
      rename_file_storage = True
6457

    
6458
    self.cfg.RenameInstance(inst.name, self.op.new_name)
6459
    # Change the instance lock. This is definitely safe while we hold the BGL.
6460
    # Otherwise the new lock would have to be added in acquired mode.
6461
    assert self.REQ_BGL
6462
    self.glm.remove(locking.LEVEL_INSTANCE, old_name)
6463
    self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
6464

    
6465
    # re-read the instance from the configuration after rename
6466
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
6467

    
6468
    if rename_file_storage:
6469
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
6470
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
6471
                                                     old_file_storage_dir,
6472
                                                     new_file_storage_dir)
6473
      result.Raise("Could not rename on node %s directory '%s' to '%s'"
6474
                   " (but the instance has been renamed in Ganeti)" %
6475
                   (inst.primary_node, old_file_storage_dir,
6476
                    new_file_storage_dir))
6477

    
6478
    _StartInstanceDisks(self, inst, None)
6479
    try:
6480
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
6481
                                                 old_name, self.op.debug_level)
6482
      msg = result.fail_msg
6483
      if msg:
6484
        msg = ("Could not run OS rename script for instance %s on node %s"
6485
               " (but the instance has been renamed in Ganeti): %s" %
6486
               (inst.name, inst.primary_node, msg))
6487
        self.proc.LogWarning(msg)
6488
    finally:
6489
      _ShutdownInstanceDisks(self, inst)
6490

    
6491
    return inst.name
6492

    
6493

    
6494
class LUInstanceRemove(LogicalUnit):
6495
  """Remove an instance.
6496

6497
  """
6498
  HPATH = "instance-remove"
6499
  HTYPE = constants.HTYPE_INSTANCE
6500
  REQ_BGL = False
6501

    
6502
  def ExpandNames(self):
6503
    self._ExpandAndLockInstance()
6504
    self.needed_locks[locking.LEVEL_NODE] = []
6505
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6506

    
6507
  def DeclareLocks(self, level):
6508
    if level == locking.LEVEL_NODE:
6509
      self._LockInstancesNodes()
6510

    
6511
  def BuildHooksEnv(self):
6512
    """Build hooks env.
6513

6514
    This runs on master, primary and secondary nodes of the instance.
6515

6516
    """
6517
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6518
    env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
6519
    return env
6520

    
6521
  def BuildHooksNodes(self):
6522
    """Build hooks nodes.
6523

6524
    """
6525
    nl = [self.cfg.GetMasterNode()]
6526
    nl_post = list(self.instance.all_nodes) + nl
6527
    return (nl, nl_post)
6528

    
6529
  def CheckPrereq(self):
6530
    """Check prerequisites.
6531

6532
    This checks that the instance is in the cluster.
6533

6534
    """
6535
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6536
    assert self.instance is not None, \
6537
      "Cannot retrieve locked instance %s" % self.op.instance_name
6538

    
6539
  def Exec(self, feedback_fn):
6540
    """Remove the instance.
6541

6542
    """
6543
    instance = self.instance
6544
    logging.info("Shutting down instance %s on node %s",
6545
                 instance.name, instance.primary_node)
6546

    
6547
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
6548
                                             self.op.shutdown_timeout)
6549
    msg = result.fail_msg
6550
    if msg:
6551
      if self.op.ignore_failures:
6552
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
6553
      else:
6554
        raise errors.OpExecError("Could not shutdown instance %s on"
6555
                                 " node %s: %s" %
6556
                                 (instance.name, instance.primary_node, msg))
6557

    
6558
    _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
6559

    
6560

    
6561
def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
6562
  """Utility function to remove an instance.
6563

6564
  """
6565
  logging.info("Removing block devices for instance %s", instance.name)
6566

    
6567
  if not _RemoveDisks(lu, instance):
6568
    if not ignore_failures:
6569
      raise errors.OpExecError("Can't remove instance's disks")
6570
    feedback_fn("Warning: can't remove instance's disks")
6571

    
6572
  logging.info("Removing instance %s out of cluster config", instance.name)
6573

    
6574
  lu.cfg.RemoveInstance(instance.name)
6575

    
6576
  assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
6577
    "Instance lock removal conflict"
6578

    
6579
  # Remove lock for the instance
6580
  lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
6581

    
6582

    
6583
class LUInstanceQuery(NoHooksLU):
6584
  """Logical unit for querying instances.
6585

6586
  """
6587
  # pylint: disable=W0142
6588
  REQ_BGL = False
6589

    
6590
  def CheckArguments(self):
6591
    self.iq = _InstanceQuery(qlang.MakeSimpleFilter("name", self.op.names),
6592
                             self.op.output_fields, self.op.use_locking)
6593

    
6594
  def ExpandNames(self):
6595
    self.iq.ExpandNames(self)
6596

    
6597
  def DeclareLocks(self, level):
6598
    self.iq.DeclareLocks(self, level)
6599

    
6600
  def Exec(self, feedback_fn):
6601
    return self.iq.OldStyleQuery(self)
6602

    
6603

    
6604
class LUInstanceFailover(LogicalUnit):
6605
  """Failover an instance.
6606

6607
  """
6608
  HPATH = "instance-failover"
6609
  HTYPE = constants.HTYPE_INSTANCE
6610
  REQ_BGL = False
6611

    
6612
  def CheckArguments(self):
6613
    """Check the arguments.
6614

6615
    """
6616
    self.iallocator = getattr(self.op, "iallocator", None)
6617
    self.target_node = getattr(self.op, "target_node", None)
6618

    
6619
  def ExpandNames(self):
6620
    self._ExpandAndLockInstance()
6621

    
6622
    if self.op.target_node is not None:
6623
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6624

    
6625
    self.needed_locks[locking.LEVEL_NODE] = []
6626
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6627

    
6628
    ignore_consistency = self.op.ignore_consistency
6629
    shutdown_timeout = self.op.shutdown_timeout
6630
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
6631
                                       cleanup=False,
6632
                                       failover=True,
6633
                                       ignore_consistency=ignore_consistency,
6634
                                       shutdown_timeout=shutdown_timeout)
6635
    self.tasklets = [self._migrater]
6636

    
6637
  def DeclareLocks(self, level):
6638
    if level == locking.LEVEL_NODE:
6639
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
6640
      if instance.disk_template in constants.DTS_EXT_MIRROR:
6641
        if self.op.target_node is None:
6642
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6643
        else:
6644
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
6645
                                                   self.op.target_node]
6646
        del self.recalculate_locks[locking.LEVEL_NODE]
6647
      else:
6648
        self._LockInstancesNodes()
6649

    
6650
  def BuildHooksEnv(self):
6651
    """Build hooks env.
6652

6653
    This runs on master, primary and secondary nodes of the instance.
6654

6655
    """
6656
    instance = self._migrater.instance
6657
    source_node = instance.primary_node
6658
    target_node = self.op.target_node
6659
    env = {
6660
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
6661
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6662
      "OLD_PRIMARY": source_node,
6663
      "NEW_PRIMARY": target_node,
6664
      }
6665

    
6666
    if instance.disk_template in constants.DTS_INT_MIRROR:
6667
      env["OLD_SECONDARY"] = instance.secondary_nodes[0]
6668
      env["NEW_SECONDARY"] = source_node
6669
    else:
6670
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = ""
6671

    
6672
    env.update(_BuildInstanceHookEnvByObject(self, instance))
6673

    
6674
    return env
6675

    
6676
  def BuildHooksNodes(self):
6677
    """Build hooks nodes.
6678

6679
    """
6680
    instance = self._migrater.instance
6681
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
6682
    return (nl, nl + [instance.primary_node])
6683

    
6684

    
6685
class LUInstanceMigrate(LogicalUnit):
6686
  """Migrate an instance.
6687

6688
  This is migration without shutting down, compared to the failover,
6689
  which is done with shutdown.
6690

6691
  """
6692
  HPATH = "instance-migrate"
6693
  HTYPE = constants.HTYPE_INSTANCE
6694
  REQ_BGL = False
6695

    
6696
  def ExpandNames(self):
6697
    self._ExpandAndLockInstance()
6698

    
6699
    if self.op.target_node is not None:
6700
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6701

    
6702
    self.needed_locks[locking.LEVEL_NODE] = []
6703
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6704

    
6705
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
6706
                                       cleanup=self.op.cleanup,
6707
                                       failover=False,
6708
                                       fallback=self.op.allow_failover)
6709
    self.tasklets = [self._migrater]
6710

    
6711
  def DeclareLocks(self, level):
6712
    if level == locking.LEVEL_NODE:
6713
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
6714
      if instance.disk_template in constants.DTS_EXT_MIRROR:
6715
        if self.op.target_node is None:
6716
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6717
        else:
6718
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
6719
                                                   self.op.target_node]
6720
        del self.recalculate_locks[locking.LEVEL_NODE]
6721
      else:
6722
        self._LockInstancesNodes()
6723

    
6724
  def BuildHooksEnv(self):
6725
    """Build hooks env.
6726

6727
    This runs on master, primary and secondary nodes of the instance.
6728

6729
    """
6730
    instance = self._migrater.instance
6731
    source_node = instance.primary_node
6732
    target_node = self.op.target_node
6733
    env = _BuildInstanceHookEnvByObject(self, instance)
6734
    env.update({
6735
      "MIGRATE_LIVE": self._migrater.live,
6736
      "MIGRATE_CLEANUP": self.op.cleanup,
6737
      "OLD_PRIMARY": source_node,
6738
      "NEW_PRIMARY": target_node,
6739
      })
6740

    
6741
    if instance.disk_template in constants.DTS_INT_MIRROR:
6742
      env["OLD_SECONDARY"] = target_node
6743
      env["NEW_SECONDARY"] = source_node
6744
    else:
6745
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
6746

    
6747
    return env
6748

    
6749
  def BuildHooksNodes(self):
6750
    """Build hooks nodes.
6751

6752
    """
6753
    instance = self._migrater.instance
6754
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
6755
    return (nl, nl + [instance.primary_node])
6756

    
6757

    
6758
class LUInstanceMove(LogicalUnit):
6759
  """Move an instance by data-copying.
6760

6761
  """
6762
  HPATH = "instance-move"
6763
  HTYPE = constants.HTYPE_INSTANCE
6764
  REQ_BGL = False
6765

    
6766
  def ExpandNames(self):
6767
    self._ExpandAndLockInstance()
6768
    target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6769
    self.op.target_node = target_node
6770
    self.needed_locks[locking.LEVEL_NODE] = [target_node]
6771
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6772

    
6773
  def DeclareLocks(self, level):
6774
    if level == locking.LEVEL_NODE:
6775
      self._LockInstancesNodes(primary_only=True)
6776

    
6777
  def BuildHooksEnv(self):
6778
    """Build hooks env.
6779

6780
    This runs on master, primary and secondary nodes of the instance.
6781

6782
    """
6783
    env = {
6784
      "TARGET_NODE": self.op.target_node,
6785
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6786
      }
6787
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6788
    return env
6789

    
6790
  def BuildHooksNodes(self):
6791
    """Build hooks nodes.
6792

6793
    """
6794
    nl = [
6795
      self.cfg.GetMasterNode(),
6796
      self.instance.primary_node,
6797
      self.op.target_node,
6798
      ]
6799
    return (nl, nl)
6800

    
6801
  def CheckPrereq(self):
6802
    """Check prerequisites.
6803

6804
    This checks that the instance is in the cluster.
6805

6806
    """
6807
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6808
    assert self.instance is not None, \
6809
      "Cannot retrieve locked instance %s" % self.op.instance_name
6810

    
6811
    node = self.cfg.GetNodeInfo(self.op.target_node)
6812
    assert node is not None, \
6813
      "Cannot retrieve locked node %s" % self.op.target_node
6814

    
6815
    self.target_node = target_node = node.name
6816

    
6817
    if target_node == instance.primary_node:
6818
      raise errors.OpPrereqError("Instance %s is already on the node %s" %
6819
                                 (instance.name, target_node),
6820
                                 errors.ECODE_STATE)
6821

    
6822
    bep = self.cfg.GetClusterInfo().FillBE(instance)
6823

    
6824
    for idx, dsk in enumerate(instance.disks):
6825
      if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
6826
        raise errors.OpPrereqError("Instance disk %d has a complex layout,"
6827
                                   " cannot copy" % idx, errors.ECODE_STATE)
6828

    
6829
    _CheckNodeOnline(self, target_node)
6830
    _CheckNodeNotDrained(self, target_node)
6831
    _CheckNodeVmCapable(self, target_node)
6832

    
6833
    if instance.admin_up:
6834
      # check memory requirements on the secondary node
6835
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
6836
                           instance.name, bep[constants.BE_MEMORY],
6837
                           instance.hypervisor)
6838
    else:
6839
      self.LogInfo("Not checking memory on the secondary node as"
6840
                   " instance will not be started")
6841

    
6842
    # check bridge existance
6843
    _CheckInstanceBridgesExist(self, instance, node=target_node)
6844

    
6845
  def Exec(self, feedback_fn):
6846
    """Move an instance.
6847

6848
    The move is done by shutting it down on its present node, copying
6849
    the data over (slow) and starting it on the new node.
6850

6851
    """
6852
    instance = self.instance
6853

    
6854
    source_node = instance.primary_node
6855
    target_node = self.target_node
6856

    
6857
    self.LogInfo("Shutting down instance %s on source node %s",
6858
                 instance.name, source_node)
6859

    
6860
    result = self.rpc.call_instance_shutdown(source_node, instance,
6861
                                             self.op.shutdown_timeout)
6862
    msg = result.fail_msg
6863
    if msg:
6864
      if self.op.ignore_consistency:
6865
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
6866
                             " Proceeding anyway. Please make sure node"
6867
                             " %s is down. Error details: %s",
6868
                             instance.name, source_node, source_node, msg)
6869
      else:
6870
        raise errors.OpExecError("Could not shutdown instance %s on"
6871
                                 " node %s: %s" %
6872
                                 (instance.name, source_node, msg))
6873

    
6874
    # create the target disks
6875
    try:
6876
      _CreateDisks(self, instance, target_node=target_node)
6877
    except errors.OpExecError:
6878
      self.LogWarning("Device creation failed, reverting...")
6879
      try:
6880
        _RemoveDisks(self, instance, target_node=target_node)
6881
      finally:
6882
        self.cfg.ReleaseDRBDMinors(instance.name)
6883
        raise
6884

    
6885
    cluster_name = self.cfg.GetClusterInfo().cluster_name
6886

    
6887
    errs = []
6888
    # activate, get path, copy the data over
6889
    for idx, disk in enumerate(instance.disks):
6890
      self.LogInfo("Copying data for disk %d", idx)
6891
      result = self.rpc.call_blockdev_assemble(target_node, disk,
6892
                                               instance.name, True, idx)
6893
      if result.fail_msg:
6894
        self.LogWarning("Can't assemble newly created disk %d: %s",
6895
                        idx, result.fail_msg)
6896
        errs.append(result.fail_msg)
6897
        break
6898
      dev_path = result.payload
6899
      result = self.rpc.call_blockdev_export(source_node, disk,
6900
                                             target_node, dev_path,
6901
                                             cluster_name)
6902
      if result.fail_msg:
6903
        self.LogWarning("Can't copy data over for disk %d: %s",
6904
                        idx, result.fail_msg)
6905
        errs.append(result.fail_msg)
6906
        break
6907

    
6908
    if errs:
6909
      self.LogWarning("Some disks failed to copy, aborting")
6910
      try:
6911
        _RemoveDisks(self, instance, target_node=target_node)
6912
      finally:
6913
        self.cfg.ReleaseDRBDMinors(instance.name)
6914
        raise errors.OpExecError("Errors during disk copy: %s" %
6915
                                 (",".join(errs),))
6916

    
6917
    instance.primary_node = target_node
6918
    self.cfg.Update(instance, feedback_fn)
6919

    
6920
    self.LogInfo("Removing the disks on the original node")
6921
    _RemoveDisks(self, instance, target_node=source_node)
6922

    
6923
    # Only start the instance if it's marked as up
6924
    if instance.admin_up:
6925
      self.LogInfo("Starting instance %s on node %s",
6926
                   instance.name, target_node)
6927

    
6928
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
6929
                                           ignore_secondaries=True)
6930
      if not disks_ok:
6931
        _ShutdownInstanceDisks(self, instance)
6932
        raise errors.OpExecError("Can't activate the instance's disks")
6933

    
6934
      result = self.rpc.call_instance_start(target_node, instance,
6935
                                            None, None, False)
6936
      msg = result.fail_msg
6937
      if msg:
6938
        _ShutdownInstanceDisks(self, instance)
6939
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
6940
                                 (instance.name, target_node, msg))
6941

    
6942

    
6943
class LUNodeMigrate(LogicalUnit):
6944
  """Migrate all instances from a node.
6945

6946
  """
6947
  HPATH = "node-migrate"
6948
  HTYPE = constants.HTYPE_NODE
6949
  REQ_BGL = False
6950

    
6951
  def CheckArguments(self):
6952
    pass
6953

    
6954
  def ExpandNames(self):
6955
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6956

    
6957
    self.share_locks = _ShareAll()
6958
    self.needed_locks = {
6959
      locking.LEVEL_NODE: [self.op.node_name],
6960
      }
6961

    
6962
  def BuildHooksEnv(self):
6963
    """Build hooks env.
6964

6965
    This runs on the master, the primary and all the secondaries.
6966

6967
    """
6968
    return {
6969
      "NODE_NAME": self.op.node_name,
6970
      }
6971

    
6972
  def BuildHooksNodes(self):
6973
    """Build hooks nodes.
6974

6975
    """
6976
    nl = [self.cfg.GetMasterNode()]
6977
    return (nl, nl)
6978

    
6979
  def CheckPrereq(self):
6980
    pass
6981

    
6982
  def Exec(self, feedback_fn):
6983
    # Prepare jobs for migration instances
6984
    jobs = [
6985
      [opcodes.OpInstanceMigrate(instance_name=inst.name,
6986
                                 mode=self.op.mode,
6987
                                 live=self.op.live,
6988
                                 iallocator=self.op.iallocator,
6989
                                 target_node=self.op.target_node)]
6990
      for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)
6991
      ]
6992

    
6993
    # TODO: Run iallocator in this opcode and pass correct placement options to
6994
    # OpInstanceMigrate. Since other jobs can modify the cluster between
6995
    # running the iallocator and the actual migration, a good consistency model
6996
    # will have to be found.
6997

    
6998
    assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
6999
            frozenset([self.op.node_name]))
7000

    
7001
    return ResultWithJobs(jobs)
7002

    
7003

    
7004
class TLMigrateInstance(Tasklet):
7005
  """Tasklet class for instance migration.
7006

7007
  @type live: boolean
7008
  @ivar live: whether the migration will be done live or non-live;
7009
      this variable is initalized only after CheckPrereq has run
7010
  @type cleanup: boolean
7011
  @ivar cleanup: Wheater we cleanup from a failed migration
7012
  @type iallocator: string
7013
  @ivar iallocator: The iallocator used to determine target_node
7014
  @type target_node: string
7015
  @ivar target_node: If given, the target_node to reallocate the instance to
7016
  @type failover: boolean
7017
  @ivar failover: Whether operation results in failover or migration
7018
  @type fallback: boolean
7019
  @ivar fallback: Whether fallback to failover is allowed if migration not
7020
                  possible
7021
  @type ignore_consistency: boolean
7022
  @ivar ignore_consistency: Wheter we should ignore consistency between source
7023
                            and target node
7024
  @type shutdown_timeout: int
7025
  @ivar shutdown_timeout: In case of failover timeout of the shutdown
7026

7027
  """
7028

    
7029
  # Constants
7030
  _MIGRATION_POLL_INTERVAL = 0.5
7031

    
7032
  def __init__(self, lu, instance_name, cleanup=False,
7033
               failover=False, fallback=False,
7034
               ignore_consistency=False,
7035
               shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT):
7036
    """Initializes this class.
7037

7038
    """
7039
    Tasklet.__init__(self, lu)
7040

    
7041
    # Parameters
7042
    self.instance_name = instance_name
7043
    self.cleanup = cleanup
7044
    self.live = False # will be overridden later
7045
    self.failover = failover
7046
    self.fallback = fallback
7047
    self.ignore_consistency = ignore_consistency
7048
    self.shutdown_timeout = shutdown_timeout
7049

    
7050
  def CheckPrereq(self):
7051
    """Check prerequisites.
7052

7053
    This checks that the instance is in the cluster.
7054

7055
    """
7056
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
7057
    instance = self.cfg.GetInstanceInfo(instance_name)
7058
    assert instance is not None
7059
    self.instance = instance
7060

    
7061
    if (not self.cleanup and not instance.admin_up and not self.failover and
7062
        self.fallback):
7063
      self.lu.LogInfo("Instance is marked down, fallback allowed, switching"
7064
                      " to failover")
7065
      self.failover = True
7066

    
7067
    if instance.disk_template not in constants.DTS_MIRRORED:
7068
      if self.failover:
7069
        text = "failovers"
7070
      else:
7071
        text = "migrations"
7072
      raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
7073
                                 " %s" % (instance.disk_template, text),
7074
                                 errors.ECODE_STATE)
7075

    
7076
    if instance.disk_template in constants.DTS_EXT_MIRROR:
7077
      _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
7078

    
7079
      if self.lu.op.iallocator:
7080
        self._RunAllocator()
7081
      else:
7082
        # We set set self.target_node as it is required by
7083
        # BuildHooksEnv
7084
        self.target_node = self.lu.op.target_node
7085

    
7086
      # self.target_node is already populated, either directly or by the
7087
      # iallocator run
7088
      target_node = self.target_node
7089
      if self.target_node == instance.primary_node:
7090
        raise errors.OpPrereqError("Cannot migrate instance %s"
7091
                                   " to its primary (%s)" %
7092
                                   (instance.name, instance.primary_node))
7093

    
7094
      if len(self.lu.tasklets) == 1:
7095
        # It is safe to release locks only when we're the only tasklet
7096
        # in the LU
7097
        _ReleaseLocks(self.lu, locking.LEVEL_NODE,
7098
                      keep=[instance.primary_node, self.target_node])
7099

    
7100
    else:
7101
      secondary_nodes = instance.secondary_nodes
7102
      if not secondary_nodes:
7103
        raise errors.ConfigurationError("No secondary node but using"
7104
                                        " %s disk template" %
7105
                                        instance.disk_template)
7106
      target_node = secondary_nodes[0]
7107
      if self.lu.op.iallocator or (self.lu.op.target_node and
7108
                                   self.lu.op.target_node != target_node):
7109
        if self.failover:
7110
          text = "failed over"
7111
        else:
7112
          text = "migrated"
7113
        raise errors.OpPrereqError("Instances with disk template %s cannot"
7114
                                   " be %s to arbitrary nodes"
7115
                                   " (neither an iallocator nor a target"
7116
                                   " node can be passed)" %
7117
                                   (instance.disk_template, text),
7118
                                   errors.ECODE_INVAL)
7119

    
7120
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
7121

    
7122
    # check memory requirements on the secondary node
7123
    if not self.failover or instance.admin_up:
7124
      _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
7125
                           instance.name, i_be[constants.BE_MEMORY],
7126
                           instance.hypervisor)
7127
    else:
7128
      self.lu.LogInfo("Not checking memory on the secondary node as"
7129
                      " instance will not be started")
7130

    
7131
    # check bridge existance
7132
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
7133

    
7134
    if not self.cleanup:
7135
      _CheckNodeNotDrained(self.lu, target_node)
7136
      if not self.failover:
7137
        result = self.rpc.call_instance_migratable(instance.primary_node,
7138
                                                   instance)
7139
        if result.fail_msg and self.fallback:
7140
          self.lu.LogInfo("Can't migrate, instance offline, fallback to"
7141
                          " failover")
7142
          self.failover = True
7143
        else:
7144
          result.Raise("Can't migrate, please use failover",
7145
                       prereq=True, ecode=errors.ECODE_STATE)
7146

    
7147
    assert not (self.failover and self.cleanup)
7148

    
7149
    if not self.failover:
7150
      if self.lu.op.live is not None and self.lu.op.mode is not None:
7151
        raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
7152
                                   " parameters are accepted",
7153
                                   errors.ECODE_INVAL)
7154
      if self.lu.op.live is not None:
7155
        if self.lu.op.live:
7156
          self.lu.op.mode = constants.HT_MIGRATION_LIVE
7157
        else:
7158
          self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
7159
        # reset the 'live' parameter to None so that repeated
7160
        # invocations of CheckPrereq do not raise an exception
7161
        self.lu.op.live = None
7162
      elif self.lu.op.mode is None:
7163
        # read the default value from the hypervisor
7164
        i_hv = self.cfg.GetClusterInfo().FillHV(self.instance,
7165
                                                skip_globals=False)
7166
        self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
7167

    
7168
      self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
7169
    else:
7170
      # Failover is never live
7171
      self.live = False
7172

    
7173
  def _RunAllocator(self):
7174
    """Run the allocator based on input opcode.
7175

7176
    """
7177
    ial = IAllocator(self.cfg, self.rpc,
7178
                     mode=constants.IALLOCATOR_MODE_RELOC,
7179
                     name=self.instance_name,
7180
                     # TODO See why hail breaks with a single node below
7181
                     relocate_from=[self.instance.primary_node,
7182
                                    self.instance.primary_node],
7183
                     )
7184

    
7185
    ial.Run(self.lu.op.iallocator)
7186

    
7187
    if not ial.success:
7188
      raise errors.OpPrereqError("Can't compute nodes using"
7189
                                 " iallocator '%s': %s" %
7190
                                 (self.lu.op.iallocator, ial.info),
7191
                                 errors.ECODE_NORES)
7192
    if len(ial.result) != ial.required_nodes:
7193
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
7194
                                 " of nodes (%s), required %s" %
7195
                                 (self.lu.op.iallocator, len(ial.result),
7196
                                  ial.required_nodes), errors.ECODE_FAULT)
7197
    self.target_node = ial.result[0]
7198
    self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
7199
                 self.instance_name, self.lu.op.iallocator,
7200
                 utils.CommaJoin(ial.result))
7201

    
7202
  def _WaitUntilSync(self):
7203
    """Poll with custom rpc for disk sync.
7204

7205
    This uses our own step-based rpc call.
7206

7207
    """
7208
    self.feedback_fn("* wait until resync is done")
7209
    all_done = False
7210
    while not all_done:
7211
      all_done = True
7212
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
7213
                                            self.nodes_ip,
7214
                                            self.instance.disks)
7215
      min_percent = 100
7216
      for node, nres in result.items():
7217
        nres.Raise("Cannot resync disks on node %s" % node)
7218
        node_done, node_percent = nres.payload
7219
        all_done = all_done and node_done
7220
        if node_percent is not None:
7221
          min_percent = min(min_percent, node_percent)
7222
      if not all_done:
7223
        if min_percent < 100:
7224
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
7225
        time.sleep(2)
7226

    
7227
  def _EnsureSecondary(self, node):
7228
    """Demote a node to secondary.
7229

7230
    """
7231
    self.feedback_fn("* switching node %s to secondary mode" % node)
7232

    
7233
    for dev in self.instance.disks:
7234
      self.cfg.SetDiskID(dev, node)
7235

    
7236
    result = self.rpc.call_blockdev_close(node, self.instance.name,
7237
                                          self.instance.disks)
7238
    result.Raise("Cannot change disk to secondary on node %s" % node)
7239

    
7240
  def _GoStandalone(self):
7241
    """Disconnect from the network.
7242

7243
    """
7244
    self.feedback_fn("* changing into standalone mode")
7245
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
7246
                                               self.instance.disks)
7247
    for node, nres in result.items():
7248
      nres.Raise("Cannot disconnect disks node %s" % node)
7249

    
7250
  def _GoReconnect(self, multimaster):
7251
    """Reconnect to the network.
7252

7253
    """
7254
    if multimaster:
7255
      msg = "dual-master"
7256
    else:
7257
      msg = "single-master"
7258
    self.feedback_fn("* changing disks into %s mode" % msg)
7259
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
7260
                                           self.instance.disks,
7261
                                           self.instance.name, multimaster)
7262
    for node, nres in result.items():
7263
      nres.Raise("Cannot change disks config on node %s" % node)
7264

    
7265
  def _ExecCleanup(self):
7266
    """Try to cleanup after a failed migration.
7267

7268
    The cleanup is done by:
7269
      - check that the instance is running only on one node
7270
        (and update the config if needed)
7271
      - change disks on its secondary node to secondary
7272
      - wait until disks are fully synchronized
7273
      - disconnect from the network
7274
      - change disks into single-master mode
7275
      - wait again until disks are fully synchronized
7276

7277
    """
7278
    instance = self.instance
7279
    target_node = self.target_node
7280
    source_node = self.source_node
7281

    
7282
    # check running on only one node
7283
    self.feedback_fn("* checking where the instance actually runs"
7284
                     " (if this hangs, the hypervisor might be in"
7285
                     " a bad state)")
7286
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
7287
    for node, result in ins_l.items():
7288
      result.Raise("Can't contact node %s" % node)
7289

    
7290
    runningon_source = instance.name in ins_l[source_node].payload
7291
    runningon_target = instance.name in ins_l[target_node].payload
7292

    
7293
    if runningon_source and runningon_target:
7294
      raise errors.OpExecError("Instance seems to be running on two nodes,"
7295
                               " or the hypervisor is confused; you will have"
7296
                               " to ensure manually that it runs only on one"
7297
                               " and restart this operation")
7298

    
7299
    if not (runningon_source or runningon_target):
7300
      raise errors.OpExecError("Instance does not seem to be running at all;"
7301
                               " in this case it's safer to repair by"
7302
                               " running 'gnt-instance stop' to ensure disk"
7303
                               " shutdown, and then restarting it")
7304

    
7305
    if runningon_target:
7306
      # the migration has actually succeeded, we need to update the config
7307
      self.feedback_fn("* instance running on secondary node (%s),"
7308
                       " updating config" % target_node)
7309
      instance.primary_node = target_node
7310
      self.cfg.Update(instance, self.feedback_fn)
7311
      demoted_node = source_node
7312
    else:
7313
      self.feedback_fn("* instance confirmed to be running on its"
7314
                       " primary node (%s)" % source_node)
7315
      demoted_node = target_node
7316

    
7317
    if instance.disk_template in constants.DTS_INT_MIRROR:
7318
      self._EnsureSecondary(demoted_node)
7319
      try:
7320
        self._WaitUntilSync()
7321
      except errors.OpExecError:
7322
        # we ignore here errors, since if the device is standalone, it
7323
        # won't be able to sync
7324
        pass
7325
      self._GoStandalone()
7326
      self._GoReconnect(False)
7327
      self._WaitUntilSync()
7328

    
7329
    self.feedback_fn("* done")
7330

    
7331
  def _RevertDiskStatus(self):
7332
    """Try to revert the disk status after a failed migration.
7333

7334
    """
7335
    target_node = self.target_node
7336
    if self.instance.disk_template in constants.DTS_EXT_MIRROR:
7337
      return
7338

    
7339
    try:
7340
      self._EnsureSecondary(target_node)
7341
      self._GoStandalone()
7342
      self._GoReconnect(False)
7343
      self._WaitUntilSync()
7344
    except errors.OpExecError, err:
7345
      self.lu.LogWarning("Migration failed and I can't reconnect the drives,"
7346
                         " please try to recover the instance manually;"
7347
                         " error '%s'" % str(err))
7348

    
7349
  def _AbortMigration(self):
7350
    """Call the hypervisor code to abort a started migration.
7351

7352
    """
7353
    instance = self.instance
7354
    target_node = self.target_node
7355
    source_node = self.source_node
7356
    migration_info = self.migration_info
7357

    
7358
    abort_result = self.rpc.call_instance_finalize_migration_dst(target_node,
7359
                                                                 instance,
7360
                                                                 migration_info,
7361
                                                                 False)
7362
    abort_msg = abort_result.fail_msg
7363
    if abort_msg:
7364
      logging.error("Aborting migration failed on target node %s: %s",
7365
                    target_node, abort_msg)
7366
      # Don't raise an exception here, as we stil have to try to revert the
7367
      # disk status, even if this step failed.
7368

    
7369
    abort_result = self.rpc.call_instance_finalize_migration_src(source_node,
7370
        instance, False, self.live)
7371
    abort_msg = abort_result.fail_msg
7372
    if abort_msg:
7373
      logging.error("Aborting migration failed on source node %s: %s",
7374
                    source_node, abort_msg)
7375

    
7376
  def _ExecMigration(self):
7377
    """Migrate an instance.
7378

7379
    The migrate is done by:
7380
      - change the disks into dual-master mode
7381
      - wait until disks are fully synchronized again
7382
      - migrate the instance
7383
      - change disks on the new secondary node (the old primary) to secondary
7384
      - wait until disks are fully synchronized
7385
      - change disks into single-master mode
7386

7387
    """
7388
    instance = self.instance
7389
    target_node = self.target_node
7390
    source_node = self.source_node
7391

    
7392
    self.feedback_fn("* checking disk consistency between source and target")
7393
    for dev in instance.disks:
7394
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
7395
        raise errors.OpExecError("Disk %s is degraded or not fully"
7396
                                 " synchronized on target node,"
7397
                                 " aborting migration" % dev.iv_name)
7398

    
7399
    # First get the migration information from the remote node
7400
    result = self.rpc.call_migration_info(source_node, instance)
7401
    msg = result.fail_msg
7402
    if msg:
7403
      log_err = ("Failed fetching source migration information from %s: %s" %
7404
                 (source_node, msg))
7405
      logging.error(log_err)
7406
      raise errors.OpExecError(log_err)
7407

    
7408
    self.migration_info = migration_info = result.payload
7409

    
7410
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
7411
      # Then switch the disks to master/master mode
7412
      self._EnsureSecondary(target_node)
7413
      self._GoStandalone()
7414
      self._GoReconnect(True)
7415
      self._WaitUntilSync()
7416

    
7417
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
7418
    result = self.rpc.call_accept_instance(target_node,
7419
                                           instance,
7420
                                           migration_info,
7421
                                           self.nodes_ip[target_node])
7422

    
7423
    msg = result.fail_msg
7424
    if msg:
7425
      logging.error("Instance pre-migration failed, trying to revert"
7426
                    " disk status: %s", msg)
7427
      self.feedback_fn("Pre-migration failed, aborting")
7428
      self._AbortMigration()
7429
      self._RevertDiskStatus()
7430
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
7431
                               (instance.name, msg))
7432

    
7433
    self.feedback_fn("* migrating instance to %s" % target_node)
7434
    result = self.rpc.call_instance_migrate(source_node, instance,
7435
                                            self.nodes_ip[target_node],
7436
                                            self.live)
7437
    msg = result.fail_msg
7438
    if msg:
7439
      logging.error("Instance migration failed, trying to revert"
7440
                    " disk status: %s", msg)
7441
      self.feedback_fn("Migration failed, aborting")
7442
      self._AbortMigration()
7443
      self._RevertDiskStatus()
7444
      raise errors.OpExecError("Could not migrate instance %s: %s" %
7445
                               (instance.name, msg))
7446

    
7447
    self.feedback_fn("* starting memory transfer")
7448
    while True:
7449
      result = self.rpc.call_instance_get_migration_status(source_node,
7450
                                                           instance)
7451
      msg = result.fail_msg
7452
      ms = result.payload   # MigrationStatus instance
7453
      if msg or (ms.status in constants.HV_MIGRATION_FAILED_STATUSES):
7454
        logging.error("Instance migration failed, trying to revert"
7455
                      " disk status: %s", msg)
7456
        self.feedback_fn("Migration failed, aborting")
7457
        self._AbortMigration()
7458
        self._RevertDiskStatus()
7459
        raise errors.OpExecError("Could not migrate instance %s: %s" %
7460
                                 (instance.name, msg))
7461

    
7462
      if result.payload.status != constants.HV_MIGRATION_ACTIVE:
7463
        self.feedback_fn("* memory transfer complete")
7464
        break
7465

    
7466
      time.sleep(self._MIGRATION_POLL_INTERVAL)
7467

    
7468
    result = self.rpc.call_instance_finalize_migration_src(source_node,
7469
                                                           instance,
7470
                                                           True,
7471
                                                           self.live)
7472
    msg = result.fail_msg
7473
    if msg:
7474
      logging.error("Instance migration succeeded, but finalization failed"
7475
                    " on the source node: %s", msg)
7476
      raise errors.OpExecError("Could not finalize instance migration: %s" %
7477
                               msg)
7478

    
7479
    instance.primary_node = target_node
7480

    
7481
    # distribute new instance config to the other nodes
7482
    self.cfg.Update(instance, self.feedback_fn)
7483

    
7484
    result = self.rpc.call_instance_finalize_migration_dst(target_node,
7485
                                                           instance,
7486
                                                           migration_info,
7487
                                                           True)
7488
    msg = result.fail_msg
7489
    if msg:
7490
      logging.error("Instance migration succeeded, but finalization failed"
7491
                    " on the target node: %s", msg)
7492
      raise errors.OpExecError("Could not finalize instance migration: %s" %
7493
                               msg)
7494

    
7495
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
7496
      self._EnsureSecondary(source_node)
7497
      self._WaitUntilSync()
7498
      self._GoStandalone()
7499
      self._GoReconnect(False)
7500
      self._WaitUntilSync()
7501

    
7502
    self.feedback_fn("* done")
7503

    
7504
  def _ExecFailover(self):
7505
    """Failover an instance.
7506

7507
    The failover is done by shutting it down on its present node and
7508
    starting it on the secondary.
7509

7510
    """
7511
    instance = self.instance
7512
    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
7513

    
7514
    source_node = instance.primary_node
7515
    target_node = self.target_node
7516

    
7517
    if instance.admin_up:
7518
      self.feedback_fn("* checking disk consistency between source and target")
7519
      for dev in instance.disks:
7520
        # for drbd, these are drbd over lvm
7521
        if not _CheckDiskConsistency(self.lu, dev, target_node, False):
7522
          if primary_node.offline:
7523
            self.feedback_fn("Node %s is offline, ignoring degraded disk %s on"
7524
                             " target node %s" %
7525
                             (primary_node.name, dev.iv_name, target_node))
7526
          elif not self.ignore_consistency:
7527
            raise errors.OpExecError("Disk %s is degraded on target node,"
7528
                                     " aborting failover" % dev.iv_name)
7529
    else:
7530
      self.feedback_fn("* not checking disk consistency as instance is not"
7531
                       " running")
7532

    
7533
    self.feedback_fn("* shutting down instance on source node")
7534
    logging.info("Shutting down instance %s on node %s",
7535
                 instance.name, source_node)
7536

    
7537
    result = self.rpc.call_instance_shutdown(source_node, instance,
7538
                                             self.shutdown_timeout)
7539
    msg = result.fail_msg
7540
    if msg:
7541
      if self.ignore_consistency or primary_node.offline:
7542
        self.lu.LogWarning("Could not shutdown instance %s on node %s,"
7543
                           " proceeding anyway; please make sure node"
7544
                           " %s is down; error details: %s",
7545
                           instance.name, source_node, source_node, msg)
7546
      else:
7547
        raise errors.OpExecError("Could not shutdown instance %s on"
7548
                                 " node %s: %s" %
7549
                                 (instance.name, source_node, msg))
7550

    
7551
    self.feedback_fn("* deactivating the instance's disks on source node")
7552
    if not _ShutdownInstanceDisks(self.lu, instance, ignore_primary=True):
7553
      raise errors.OpExecError("Can't shut down the instance's disks")
7554

    
7555
    instance.primary_node = target_node
7556
    # distribute new instance config to the other nodes
7557
    self.cfg.Update(instance, self.feedback_fn)
7558

    
7559
    # Only start the instance if it's marked as up
7560
    if instance.admin_up:
7561
      self.feedback_fn("* activating the instance's disks on target node %s" %
7562
                       target_node)
7563
      logging.info("Starting instance %s on node %s",
7564
                   instance.name, target_node)
7565

    
7566
      disks_ok, _ = _AssembleInstanceDisks(self.lu, instance,
7567
                                           ignore_secondaries=True)
7568
      if not disks_ok:
7569
        _ShutdownInstanceDisks(self.lu, instance)
7570
        raise errors.OpExecError("Can't activate the instance's disks")
7571

    
7572
      self.feedback_fn("* starting the instance on the target node %s" %
7573
                       target_node)
7574
      result = self.rpc.call_instance_start(target_node, instance, None, None,
7575
                                            False)
7576
      msg = result.fail_msg
7577
      if msg:
7578
        _ShutdownInstanceDisks(self.lu, instance)
7579
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
7580
                                 (instance.name, target_node, msg))
7581

    
7582
  def Exec(self, feedback_fn):
7583
    """Perform the migration.
7584

7585
    """
7586
    self.feedback_fn = feedback_fn
7587
    self.source_node = self.instance.primary_node
7588

    
7589
    # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
7590
    if self.instance.disk_template in constants.DTS_INT_MIRROR:
7591
      self.target_node = self.instance.secondary_nodes[0]
7592
      # Otherwise self.target_node has been populated either
7593
      # directly, or through an iallocator.
7594

    
7595
    self.all_nodes = [self.source_node, self.target_node]
7596
    self.nodes_ip = dict((name, node.secondary_ip) for (name, node)
7597
                         in self.cfg.GetMultiNodeInfo(self.all_nodes))
7598

    
7599
    if self.failover:
7600
      feedback_fn("Failover instance %s" % self.instance.name)
7601
      self._ExecFailover()
7602
    else:
7603
      feedback_fn("Migrating instance %s" % self.instance.name)
7604

    
7605
      if self.cleanup:
7606
        return self._ExecCleanup()
7607
      else:
7608
        return self._ExecMigration()
7609

    
7610

    
7611
def _CreateBlockDev(lu, node, instance, device, force_create,
7612
                    info, force_open):
7613
  """Create a tree of block devices on a given node.
7614

7615
  If this device type has to be created on secondaries, create it and
7616
  all its children.
7617

7618
  If not, just recurse to children keeping the same 'force' value.
7619

7620
  @param lu: the lu on whose behalf we execute
7621
  @param node: the node on which to create the device
7622
  @type instance: L{objects.Instance}
7623
  @param instance: the instance which owns the device
7624
  @type device: L{objects.Disk}
7625
  @param device: the device to create
7626
  @type force_create: boolean
7627
  @param force_create: whether to force creation of this device; this
7628
      will be change to True whenever we find a device which has
7629
      CreateOnSecondary() attribute
7630
  @param info: the extra 'metadata' we should attach to the device
7631
      (this will be represented as a LVM tag)
7632
  @type force_open: boolean
7633
  @param force_open: this parameter will be passes to the
7634
      L{backend.BlockdevCreate} function where it specifies
7635
      whether we run on primary or not, and it affects both
7636
      the child assembly and the device own Open() execution
7637

7638
  """
7639
  if device.CreateOnSecondary():
7640
    force_create = True
7641

    
7642
  if device.children:
7643
    for child in device.children:
7644
      _CreateBlockDev(lu, node, instance, child, force_create,
7645
                      info, force_open)
7646

    
7647
  if not force_create:
7648
    return
7649

    
7650
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
7651

    
7652

    
7653
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
7654
  """Create a single block device on a given node.
7655

7656
  This will not recurse over children of the device, so they must be
7657
  created in advance.
7658

7659
  @param lu: the lu on whose behalf we execute
7660
  @param node: the node on which to create the device
7661
  @type instance: L{objects.Instance}
7662
  @param instance: the instance which owns the device
7663
  @type device: L{objects.Disk}
7664
  @param device: the device to create
7665
  @param info: the extra 'metadata' we should attach to the device
7666
      (this will be represented as a LVM tag)
7667
  @type force_open: boolean
7668
  @param force_open: this parameter will be passes to the
7669
      L{backend.BlockdevCreate} function where it specifies
7670
      whether we run on primary or not, and it affects both
7671
      the child assembly and the device own Open() execution
7672

7673
  """
7674
  lu.cfg.SetDiskID(device, node)
7675
  result = lu.rpc.call_blockdev_create(node, device, device.size,
7676
                                       instance.name, force_open, info)
7677
  result.Raise("Can't create block device %s on"
7678
               " node %s for instance %s" % (device, node, instance.name))
7679
  if device.physical_id is None:
7680
    device.physical_id = result.payload
7681

    
7682

    
7683
def _GenerateUniqueNames(lu, exts):
7684
  """Generate a suitable LV name.
7685

7686
  This will generate a logical volume name for the given instance.
7687

7688
  """
7689
  results = []
7690
  for val in exts:
7691
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
7692
    results.append("%s%s" % (new_id, val))
7693
  return results
7694

    
7695

    
7696
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
7697
                         iv_name, p_minor, s_minor):
7698
  """Generate a drbd8 device complete with its children.
7699

7700
  """
7701
  assert len(vgnames) == len(names) == 2
7702
  port = lu.cfg.AllocatePort()
7703
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
7704
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
7705
                          logical_id=(vgnames[0], names[0]))
7706
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
7707
                          logical_id=(vgnames[1], names[1]))
7708
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
7709
                          logical_id=(primary, secondary, port,
7710
                                      p_minor, s_minor,
7711
                                      shared_secret),
7712
                          children=[dev_data, dev_meta],
7713
                          iv_name=iv_name)
7714
  return drbd_dev
7715

    
7716

    
7717
def _GenerateDiskTemplate(lu, template_name,
7718
                          instance_name, primary_node,
7719
                          secondary_nodes, disk_info,
7720
                          file_storage_dir, file_driver,
7721
                          base_index, feedback_fn):
7722
  """Generate the entire disk layout for a given template type.
7723

7724
  """
7725
  #TODO: compute space requirements
7726

    
7727
  vgname = lu.cfg.GetVGName()
7728
  disk_count = len(disk_info)
7729
  disks = []
7730
  if template_name == constants.DT_DISKLESS:
7731
    pass
7732
  elif template_name == constants.DT_PLAIN:
7733
    if len(secondary_nodes) != 0:
7734
      raise errors.ProgrammerError("Wrong template configuration")
7735

    
7736
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
7737
                                      for i in range(disk_count)])
7738
    for idx, disk in enumerate(disk_info):
7739
      disk_index = idx + base_index
7740
      vg = disk.get(constants.IDISK_VG, vgname)
7741
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
7742
      disk_dev = objects.Disk(dev_type=constants.LD_LV,
7743
                              size=disk[constants.IDISK_SIZE],
7744
                              logical_id=(vg, names[idx]),
7745
                              iv_name="disk/%d" % disk_index,
7746
                              mode=disk[constants.IDISK_MODE])
7747
      disks.append(disk_dev)
7748
  elif template_name == constants.DT_DRBD8:
7749
    if len(secondary_nodes) != 1:
7750
      raise errors.ProgrammerError("Wrong template configuration")
7751
    remote_node = secondary_nodes[0]
7752
    minors = lu.cfg.AllocateDRBDMinor(
7753
      [primary_node, remote_node] * len(disk_info), instance_name)
7754

    
7755
    names = []
7756
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
7757
                                               for i in range(disk_count)]):
7758
      names.append(lv_prefix + "_data")
7759
      names.append(lv_prefix + "_meta")
7760
    for idx, disk in enumerate(disk_info):
7761
      disk_index = idx + base_index
7762
      data_vg = disk.get(constants.IDISK_VG, vgname)
7763
      meta_vg = disk.get(constants.IDISK_METAVG, data_vg)
7764
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
7765
                                      disk[constants.IDISK_SIZE],
7766
                                      [data_vg, meta_vg],
7767
                                      names[idx * 2:idx * 2 + 2],
7768
                                      "disk/%d" % disk_index,
7769
                                      minors[idx * 2], minors[idx * 2 + 1])
7770
      disk_dev.mode = disk[constants.IDISK_MODE]
7771
      disks.append(disk_dev)
7772
  elif template_name == constants.DT_FILE:
7773
    if len(secondary_nodes) != 0:
7774
      raise errors.ProgrammerError("Wrong template configuration")
7775

    
7776
    opcodes.RequireFileStorage()
7777

    
7778
    for idx, disk in enumerate(disk_info):
7779
      disk_index = idx + base_index
7780
      disk_dev = objects.Disk(dev_type=constants.LD_FILE,
7781
                              size=disk[constants.IDISK_SIZE],
7782
                              iv_name="disk/%d" % disk_index,
7783
                              logical_id=(file_driver,
7784
                                          "%s/disk%d" % (file_storage_dir,
7785
                                                         disk_index)),
7786
                              mode=disk[constants.IDISK_MODE])
7787
      disks.append(disk_dev)
7788
  elif template_name == constants.DT_SHARED_FILE:
7789
    if len(secondary_nodes) != 0:
7790
      raise errors.ProgrammerError("Wrong template configuration")
7791

    
7792
    opcodes.RequireSharedFileStorage()
7793

    
7794
    for idx, disk in enumerate(disk_info):
7795
      disk_index = idx + base_index
7796
      disk_dev = objects.Disk(dev_type=constants.LD_FILE,
7797
                              size=disk[constants.IDISK_SIZE],
7798
                              iv_name="disk/%d" % disk_index,
7799
                              logical_id=(file_driver,
7800
                                          "%s/disk%d" % (file_storage_dir,
7801
                                                         disk_index)),
7802
                              mode=disk[constants.IDISK_MODE])
7803
      disks.append(disk_dev)
7804
  elif template_name == constants.DT_BLOCK:
7805
    if len(secondary_nodes) != 0:
7806
      raise errors.ProgrammerError("Wrong template configuration")
7807

    
7808
    for idx, disk in enumerate(disk_info):
7809
      disk_index = idx + base_index
7810
      disk_dev = objects.Disk(dev_type=constants.LD_BLOCKDEV,
7811
                              size=disk[constants.IDISK_SIZE],
7812
                              logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
7813
                                          disk[constants.IDISK_ADOPT]),
7814
                              iv_name="disk/%d" % disk_index,
7815
                              mode=disk[constants.IDISK_MODE])
7816
      disks.append(disk_dev)
7817

    
7818
  else:
7819
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
7820
  return disks
7821

    
7822

    
7823
def _GetInstanceInfoText(instance):
7824
  """Compute that text that should be added to the disk's metadata.
7825

7826
  """
7827
  return "originstname+%s" % instance.name
7828

    
7829

    
7830
def _CalcEta(time_taken, written, total_size):
7831
  """Calculates the ETA based on size written and total size.
7832

7833
  @param time_taken: The time taken so far
7834
  @param written: amount written so far
7835
  @param total_size: The total size of data to be written
7836
  @return: The remaining time in seconds
7837

7838
  """
7839
  avg_time = time_taken / float(written)
7840
  return (total_size - written) * avg_time
7841

    
7842

    
7843
def _WipeDisks(lu, instance):
7844
  """Wipes instance disks.
7845

7846
  @type lu: L{LogicalUnit}
7847
  @param lu: the logical unit on whose behalf we execute
7848
  @type instance: L{objects.Instance}
7849
  @param instance: the instance whose disks we should create
7850
  @return: the success of the wipe
7851

7852
  """
7853
  node = instance.primary_node
7854

    
7855
  for device in instance.disks:
7856
    lu.cfg.SetDiskID(device, node)
7857

    
7858
  logging.info("Pause sync of instance %s disks", instance.name)
7859
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
7860

    
7861
  for idx, success in enumerate(result.payload):
7862
    if not success:
7863
      logging.warn("pause-sync of instance %s for disks %d failed",
7864
                   instance.name, idx)
7865

    
7866
  try:
7867
    for idx, device in enumerate(instance.disks):
7868
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
7869
      # MAX_WIPE_CHUNK at max
7870
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
7871
                            constants.MIN_WIPE_CHUNK_PERCENT)
7872
      # we _must_ make this an int, otherwise rounding errors will
7873
      # occur
7874
      wipe_chunk_size = int(wipe_chunk_size)
7875

    
7876
      lu.LogInfo("* Wiping disk %d", idx)
7877
      logging.info("Wiping disk %d for instance %s, node %s using"
7878
                   " chunk size %s", idx, instance.name, node, wipe_chunk_size)
7879

    
7880
      offset = 0
7881
      size = device.size
7882
      last_output = 0
7883
      start_time = time.time()
7884

    
7885
      while offset < size:
7886
        wipe_size = min(wipe_chunk_size, size - offset)
7887
        logging.debug("Wiping disk %d, offset %s, chunk %s",
7888
                      idx, offset, wipe_size)
7889
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
7890
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
7891
                     (idx, offset, wipe_size))
7892
        now = time.time()
7893
        offset += wipe_size
7894
        if now - last_output >= 60:
7895
          eta = _CalcEta(now - start_time, offset, size)
7896
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
7897
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
7898
          last_output = now
7899
  finally:
7900
    logging.info("Resume sync of instance %s disks", instance.name)
7901

    
7902
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
7903

    
7904
    for idx, success in enumerate(result.payload):
7905
      if not success:
7906
        lu.LogWarning("Resume sync of disk %d failed, please have a"
7907
                      " look at the status and troubleshoot the issue", idx)
7908
        logging.warn("resume-sync of instance %s for disks %d failed",
7909
                     instance.name, idx)
7910

    
7911

    
7912
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
7913
  """Create all disks for an instance.
7914

7915
  This abstracts away some work from AddInstance.
7916

7917
  @type lu: L{LogicalUnit}
7918
  @param lu: the logical unit on whose behalf we execute
7919
  @type instance: L{objects.Instance}
7920
  @param instance: the instance whose disks we should create
7921
  @type to_skip: list
7922
  @param to_skip: list of indices to skip
7923
  @type target_node: string
7924
  @param target_node: if passed, overrides the target node for creation
7925
  @rtype: boolean
7926
  @return: the success of the creation
7927

7928
  """
7929
  info = _GetInstanceInfoText(instance)
7930
  if target_node is None:
7931
    pnode = instance.primary_node
7932
    all_nodes = instance.all_nodes
7933
  else:
7934
    pnode = target_node
7935
    all_nodes = [pnode]
7936

    
7937
  if instance.disk_template in constants.DTS_FILEBASED:
7938
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
7939
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
7940

    
7941
    result.Raise("Failed to create directory '%s' on"
7942
                 " node %s" % (file_storage_dir, pnode))
7943

    
7944
  # Note: this needs to be kept in sync with adding of disks in
7945
  # LUInstanceSetParams
7946
  for idx, device in enumerate(instance.disks):
7947
    if to_skip and idx in to_skip:
7948
      continue
7949
    logging.info("Creating volume %s for instance %s",
7950
                 device.iv_name, instance.name)
7951
    #HARDCODE
7952
    for node in all_nodes:
7953
      f_create = node == pnode
7954
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
7955

    
7956

    
7957
def _RemoveDisks(lu, instance, target_node=None):
7958
  """Remove all disks for an instance.
7959

7960
  This abstracts away some work from `AddInstance()` and
7961
  `RemoveInstance()`. Note that in case some of the devices couldn't
7962
  be removed, the removal will continue with the other ones (compare
7963
  with `_CreateDisks()`).
7964

7965
  @type lu: L{LogicalUnit}
7966
  @param lu: the logical unit on whose behalf we execute
7967
  @type instance: L{objects.Instance}
7968
  @param instance: the instance whose disks we should remove
7969
  @type target_node: string
7970
  @param target_node: used to override the node on which to remove the disks
7971
  @rtype: boolean
7972
  @return: the success of the removal
7973

7974
  """
7975
  logging.info("Removing block devices for instance %s", instance.name)
7976

    
7977
  all_result = True
7978
  for device in instance.disks:
7979
    if target_node:
7980
      edata = [(target_node, device)]
7981
    else:
7982
      edata = device.ComputeNodeTree(instance.primary_node)
7983
    for node, disk in edata:
7984
      lu.cfg.SetDiskID(disk, node)
7985
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
7986
      if msg:
7987
        lu.LogWarning("Could not remove block device %s on node %s,"
7988
                      " continuing anyway: %s", device.iv_name, node, msg)
7989
        all_result = False
7990

    
7991
  if instance.disk_template == constants.DT_FILE:
7992
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
7993
    if target_node:
7994
      tgt = target_node
7995
    else:
7996
      tgt = instance.primary_node
7997
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
7998
    if result.fail_msg:
7999
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
8000
                    file_storage_dir, instance.primary_node, result.fail_msg)
8001
      all_result = False
8002

    
8003
  return all_result
8004

    
8005

    
8006
def _ComputeDiskSizePerVG(disk_template, disks):
8007
  """Compute disk size requirements in the volume group
8008

8009
  """
8010
  def _compute(disks, payload):
8011
    """Universal algorithm.
8012

8013
    """
8014
    vgs = {}
8015
    for disk in disks:
8016
      vgs[disk[constants.IDISK_VG]] = \
8017
        vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload
8018

    
8019
    return vgs
8020

    
8021
  # Required free disk space as a function of disk and swap space
8022
  req_size_dict = {
8023
    constants.DT_DISKLESS: {},
8024
    constants.DT_PLAIN: _compute(disks, 0),
8025
    # 128 MB are added for drbd metadata for each disk
8026
    constants.DT_DRBD8: _compute(disks, 128),
8027
    constants.DT_FILE: {},
8028
    constants.DT_SHARED_FILE: {},
8029
  }
8030

    
8031
  if disk_template not in req_size_dict:
8032
    raise errors.ProgrammerError("Disk template '%s' size requirement"
8033
                                 " is unknown" % disk_template)
8034

    
8035
  return req_size_dict[disk_template]
8036

    
8037

    
8038
def _ComputeDiskSize(disk_template, disks):
8039
  """Compute disk size requirements in the volume group
8040

8041
  """
8042
  # Required free disk space as a function of disk and swap space
8043
  req_size_dict = {
8044
    constants.DT_DISKLESS: None,
8045
    constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
8046
    # 128 MB are added for drbd metadata for each disk
8047
    constants.DT_DRBD8: sum(d[constants.IDISK_SIZE] + 128 for d in disks),
8048
    constants.DT_FILE: None,
8049
    constants.DT_SHARED_FILE: 0,
8050
    constants.DT_BLOCK: 0,
8051
  }
8052

    
8053
  if disk_template not in req_size_dict:
8054
    raise errors.ProgrammerError("Disk template '%s' size requirement"
8055
                                 " is unknown" % disk_template)
8056

    
8057
  return req_size_dict[disk_template]
8058

    
8059

    
8060
def _FilterVmNodes(lu, nodenames):
8061
  """Filters out non-vm_capable nodes from a list.
8062

8063
  @type lu: L{LogicalUnit}
8064
  @param lu: the logical unit for which we check
8065
  @type nodenames: list
8066
  @param nodenames: the list of nodes on which we should check
8067
  @rtype: list
8068
  @return: the list of vm-capable nodes
8069

8070
  """
8071
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
8072
  return [name for name in nodenames if name not in vm_nodes]
8073

    
8074

    
8075
def _CheckHVParams(lu, nodenames, hvname, hvparams):
8076
  """Hypervisor parameter validation.
8077

8078
  This function abstract the hypervisor parameter validation to be
8079
  used in both instance create and instance modify.
8080

8081
  @type lu: L{LogicalUnit}
8082
  @param lu: the logical unit for which we check
8083
  @type nodenames: list
8084
  @param nodenames: the list of nodes on which we should check
8085
  @type hvname: string
8086
  @param hvname: the name of the hypervisor we should use
8087
  @type hvparams: dict
8088
  @param hvparams: the parameters which we need to check
8089
  @raise errors.OpPrereqError: if the parameters are not valid
8090

8091
  """
8092
  nodenames = _FilterVmNodes(lu, nodenames)
8093
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
8094
                                                  hvname,
8095
                                                  hvparams)
8096
  for node in nodenames:
8097
    info = hvinfo[node]
8098
    if info.offline:
8099
      continue
8100
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
8101

    
8102

    
8103
def _CheckOSParams(lu, required, nodenames, osname, osparams):
8104
  """OS parameters validation.
8105

8106
  @type lu: L{LogicalUnit}
8107
  @param lu: the logical unit for which we check
8108
  @type required: boolean
8109
  @param required: whether the validation should fail if the OS is not
8110
      found
8111
  @type nodenames: list
8112
  @param nodenames: the list of nodes on which we should check
8113
  @type osname: string
8114
  @param osname: the name of the hypervisor we should use
8115
  @type osparams: dict
8116
  @param osparams: the parameters which we need to check
8117
  @raise errors.OpPrereqError: if the parameters are not valid
8118

8119
  """
8120
  nodenames = _FilterVmNodes(lu, nodenames)
8121
  result = lu.rpc.call_os_validate(required, nodenames, osname,
8122
                                   [constants.OS_VALIDATE_PARAMETERS],
8123
                                   osparams)
8124
  for node, nres in result.items():
8125
    # we don't check for offline cases since this should be run only
8126
    # against the master node and/or an instance's nodes
8127
    nres.Raise("OS Parameters validation failed on node %s" % node)
8128
    if not nres.payload:
8129
      lu.LogInfo("OS %s not found on node %s, validation skipped",
8130
                 osname, node)
8131

    
8132

    
8133
class LUInstanceCreate(LogicalUnit):
8134
  """Create an instance.
8135

8136
  """
8137
  HPATH = "instance-add"
8138
  HTYPE = constants.HTYPE_INSTANCE
8139
  REQ_BGL = False
8140

    
8141
  def CheckArguments(self):
8142
    """Check arguments.
8143

8144
    """
8145
    # do not require name_check to ease forward/backward compatibility
8146
    # for tools
8147
    if self.op.no_install and self.op.start:
8148
      self.LogInfo("No-installation mode selected, disabling startup")
8149
      self.op.start = False
8150
    # validate/normalize the instance name
8151
    self.op.instance_name = \
8152
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
8153

    
8154
    if self.op.ip_check and not self.op.name_check:
8155
      # TODO: make the ip check more flexible and not depend on the name check
8156
      raise errors.OpPrereqError("Cannot do IP address check without a name"
8157
                                 " check", errors.ECODE_INVAL)
8158

    
8159
    # check nics' parameter names
8160
    for nic in self.op.nics:
8161
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
8162

    
8163
    # check disks. parameter names and consistent adopt/no-adopt strategy
8164
    has_adopt = has_no_adopt = False
8165
    for disk in self.op.disks:
8166
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
8167
      if constants.IDISK_ADOPT in disk:
8168
        has_adopt = True
8169
      else:
8170
        has_no_adopt = True
8171
    if has_adopt and has_no_adopt:
8172
      raise errors.OpPrereqError("Either all disks are adopted or none is",
8173
                                 errors.ECODE_INVAL)
8174
    if has_adopt:
8175
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
8176
        raise errors.OpPrereqError("Disk adoption is not supported for the"
8177
                                   " '%s' disk template" %
8178
                                   self.op.disk_template,
8179
                                   errors.ECODE_INVAL)
8180
      if self.op.iallocator is not None:
8181
        raise errors.OpPrereqError("Disk adoption not allowed with an"
8182
                                   " iallocator script", errors.ECODE_INVAL)
8183
      if self.op.mode == constants.INSTANCE_IMPORT:
8184
        raise errors.OpPrereqError("Disk adoption not allowed for"
8185
                                   " instance import", errors.ECODE_INVAL)
8186
    else:
8187
      if self.op.disk_template in constants.DTS_MUST_ADOPT:
8188
        raise errors.OpPrereqError("Disk template %s requires disk adoption,"
8189
                                   " but no 'adopt' parameter given" %
8190
                                   self.op.disk_template,
8191
                                   errors.ECODE_INVAL)
8192

    
8193
    self.adopt_disks = has_adopt
8194

    
8195
    # instance name verification
8196
    if self.op.name_check:
8197
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
8198
      self.op.instance_name = self.hostname1.name
8199
      # used in CheckPrereq for ip ping check
8200
      self.check_ip = self.hostname1.ip
8201
    else:
8202
      self.check_ip = None
8203

    
8204
    # file storage checks
8205
    if (self.op.file_driver and
8206
        not self.op.file_driver in constants.FILE_DRIVER):
8207
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
8208
                                 self.op.file_driver, errors.ECODE_INVAL)
8209

    
8210
    if self.op.disk_template == constants.DT_FILE:
8211
      opcodes.RequireFileStorage()
8212
    elif self.op.disk_template == constants.DT_SHARED_FILE:
8213
      opcodes.RequireSharedFileStorage()
8214

    
8215
    ### Node/iallocator related checks
8216
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
8217

    
8218
    if self.op.pnode is not None:
8219
      if self.op.disk_template in constants.DTS_INT_MIRROR:
8220
        if self.op.snode is None:
8221
          raise errors.OpPrereqError("The networked disk templates need"
8222
                                     " a mirror node", errors.ECODE_INVAL)
8223
      elif self.op.snode:
8224
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
8225
                        " template")
8226
        self.op.snode = None
8227

    
8228
    self._cds = _GetClusterDomainSecret()
8229

    
8230
    if self.op.mode == constants.INSTANCE_IMPORT:
8231
      # On import force_variant must be True, because if we forced it at
8232
      # initial install, our only chance when importing it back is that it
8233
      # works again!
8234
      self.op.force_variant = True
8235

    
8236
      if self.op.no_install:
8237
        self.LogInfo("No-installation mode has no effect during import")
8238

    
8239
    elif self.op.mode == constants.INSTANCE_CREATE:
8240
      if self.op.os_type is None:
8241
        raise errors.OpPrereqError("No guest OS specified",
8242
                                   errors.ECODE_INVAL)
8243
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
8244
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
8245
                                   " installation" % self.op.os_type,
8246
                                   errors.ECODE_STATE)
8247
      if self.op.disk_template is None:
8248
        raise errors.OpPrereqError("No disk template specified",
8249
                                   errors.ECODE_INVAL)
8250

    
8251
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
8252
      # Check handshake to ensure both clusters have the same domain secret
8253
      src_handshake = self.op.source_handshake
8254
      if not src_handshake:
8255
        raise errors.OpPrereqError("Missing source handshake",
8256
                                   errors.ECODE_INVAL)
8257

    
8258
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
8259
                                                           src_handshake)
8260
      if errmsg:
8261
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
8262
                                   errors.ECODE_INVAL)
8263

    
8264
      # Load and check source CA
8265
      self.source_x509_ca_pem = self.op.source_x509_ca
8266
      if not self.source_x509_ca_pem:
8267
        raise errors.OpPrereqError("Missing source X509 CA",
8268
                                   errors.ECODE_INVAL)
8269

    
8270
      try:
8271
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
8272
                                                    self._cds)
8273
      except OpenSSL.crypto.Error, err:
8274
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
8275
                                   (err, ), errors.ECODE_INVAL)
8276

    
8277
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
8278
      if errcode is not None:
8279
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
8280
                                   errors.ECODE_INVAL)
8281

    
8282
      self.source_x509_ca = cert
8283

    
8284
      src_instance_name = self.op.source_instance_name
8285
      if not src_instance_name:
8286
        raise errors.OpPrereqError("Missing source instance name",
8287
                                   errors.ECODE_INVAL)
8288

    
8289
      self.source_instance_name = \
8290
          netutils.GetHostname(name=src_instance_name).name
8291

    
8292
    else:
8293
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
8294
                                 self.op.mode, errors.ECODE_INVAL)
8295

    
8296
  def ExpandNames(self):
8297
    """ExpandNames for CreateInstance.
8298

8299
    Figure out the right locks for instance creation.
8300

8301
    """
8302
    self.needed_locks = {}
8303

    
8304
    instance_name = self.op.instance_name
8305
    # this is just a preventive check, but someone might still add this
8306
    # instance in the meantime, and creation will fail at lock-add time
8307
    if instance_name in self.cfg.GetInstanceList():
8308
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
8309
                                 instance_name, errors.ECODE_EXISTS)
8310

    
8311
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
8312

    
8313
    if self.op.iallocator:
8314
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8315
    else:
8316
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
8317
      nodelist = [self.op.pnode]
8318
      if self.op.snode is not None:
8319
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
8320
        nodelist.append(self.op.snode)
8321
      self.needed_locks[locking.LEVEL_NODE] = nodelist
8322

    
8323
    # in case of import lock the source node too
8324
    if self.op.mode == constants.INSTANCE_IMPORT:
8325
      src_node = self.op.src_node
8326
      src_path = self.op.src_path
8327

    
8328
      if src_path is None:
8329
        self.op.src_path = src_path = self.op.instance_name
8330

    
8331
      if src_node is None:
8332
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8333
        self.op.src_node = None
8334
        if os.path.isabs(src_path):
8335
          raise errors.OpPrereqError("Importing an instance from a path"
8336
                                     " requires a source node option",
8337
                                     errors.ECODE_INVAL)
8338
      else:
8339
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
8340
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
8341
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
8342
        if not os.path.isabs(src_path):
8343
          self.op.src_path = src_path = \
8344
            utils.PathJoin(constants.EXPORT_DIR, src_path)
8345

    
8346
  def _RunAllocator(self):
8347
    """Run the allocator based on input opcode.
8348

8349
    """
8350
    nics = [n.ToDict() for n in self.nics]
8351
    ial = IAllocator(self.cfg, self.rpc,
8352
                     mode=constants.IALLOCATOR_MODE_ALLOC,
8353
                     name=self.op.instance_name,
8354
                     disk_template=self.op.disk_template,
8355
                     tags=self.op.tags,
8356
                     os=self.op.os_type,
8357
                     vcpus=self.be_full[constants.BE_VCPUS],
8358
                     memory=self.be_full[constants.BE_MEMORY],
8359
                     disks=self.disks,
8360
                     nics=nics,
8361
                     hypervisor=self.op.hypervisor,
8362
                     )
8363

    
8364
    ial.Run(self.op.iallocator)
8365

    
8366
    if not ial.success:
8367
      raise errors.OpPrereqError("Can't compute nodes using"
8368
                                 " iallocator '%s': %s" %
8369
                                 (self.op.iallocator, ial.info),
8370
                                 errors.ECODE_NORES)
8371
    if len(ial.result) != ial.required_nodes:
8372
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8373
                                 " of nodes (%s), required %s" %
8374
                                 (self.op.iallocator, len(ial.result),
8375
                                  ial.required_nodes), errors.ECODE_FAULT)
8376
    self.op.pnode = ial.result[0]
8377
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
8378
                 self.op.instance_name, self.op.iallocator,
8379
                 utils.CommaJoin(ial.result))
8380
    if ial.required_nodes == 2:
8381
      self.op.snode = ial.result[1]
8382

    
8383
  def BuildHooksEnv(self):
8384
    """Build hooks env.
8385

8386
    This runs on master, primary and secondary nodes of the instance.
8387

8388
    """
8389
    env = {
8390
      "ADD_MODE": self.op.mode,
8391
      }
8392
    if self.op.mode == constants.INSTANCE_IMPORT:
8393
      env["SRC_NODE"] = self.op.src_node
8394
      env["SRC_PATH"] = self.op.src_path
8395
      env["SRC_IMAGES"] = self.src_images
8396

    
8397
    env.update(_BuildInstanceHookEnv(
8398
      name=self.op.instance_name,
8399
      primary_node=self.op.pnode,
8400
      secondary_nodes=self.secondaries,
8401
      status=self.op.start,
8402
      os_type=self.op.os_type,
8403
      memory=self.be_full[constants.BE_MEMORY],
8404
      vcpus=self.be_full[constants.BE_VCPUS],
8405
      nics=_NICListToTuple(self, self.nics),
8406
      disk_template=self.op.disk_template,
8407
      disks=[(d[constants.IDISK_SIZE], d[constants.IDISK_MODE])
8408
             for d in self.disks],
8409
      bep=self.be_full,
8410
      hvp=self.hv_full,
8411
      hypervisor_name=self.op.hypervisor,
8412
      tags=self.op.tags,
8413
    ))
8414

    
8415
    return env
8416

    
8417
  def BuildHooksNodes(self):
8418
    """Build hooks nodes.
8419

8420
    """
8421
    nl = [self.cfg.GetMasterNode(), self.op.pnode] + self.secondaries
8422
    return nl, nl
8423

    
8424
  def _ReadExportInfo(self):
8425
    """Reads the export information from disk.
8426

8427
    It will override the opcode source node and path with the actual
8428
    information, if these two were not specified before.
8429

8430
    @return: the export information
8431

8432
    """
8433
    assert self.op.mode == constants.INSTANCE_IMPORT
8434

    
8435
    src_node = self.op.src_node
8436
    src_path = self.op.src_path
8437

    
8438
    if src_node is None:
8439
      locked_nodes = self.owned_locks(locking.LEVEL_NODE)
8440
      exp_list = self.rpc.call_export_list(locked_nodes)
8441
      found = False
8442
      for node in exp_list:
8443
        if exp_list[node].fail_msg:
8444
          continue
8445
        if src_path in exp_list[node].payload:
8446
          found = True
8447
          self.op.src_node = src_node = node
8448
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
8449
                                                       src_path)
8450
          break
8451
      if not found:
8452
        raise errors.OpPrereqError("No export found for relative path %s" %
8453
                                    src_path, errors.ECODE_INVAL)
8454

    
8455
    _CheckNodeOnline(self, src_node)
8456
    result = self.rpc.call_export_info(src_node, src_path)
8457
    result.Raise("No export or invalid export found in dir %s" % src_path)
8458

    
8459
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
8460
    if not export_info.has_section(constants.INISECT_EXP):
8461
      raise errors.ProgrammerError("Corrupted export config",
8462
                                   errors.ECODE_ENVIRON)
8463

    
8464
    ei_version = export_info.get(constants.INISECT_EXP, "version")
8465
    if (int(ei_version) != constants.EXPORT_VERSION):
8466
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
8467
                                 (ei_version, constants.EXPORT_VERSION),
8468
                                 errors.ECODE_ENVIRON)
8469
    return export_info
8470

    
8471
  def _ReadExportParams(self, einfo):
8472
    """Use export parameters as defaults.
8473

8474
    In case the opcode doesn't specify (as in override) some instance
8475
    parameters, then try to use them from the export information, if
8476
    that declares them.
8477

8478
    """
8479
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
8480

    
8481
    if self.op.disk_template is None:
8482
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
8483
        self.op.disk_template = einfo.get(constants.INISECT_INS,
8484
                                          "disk_template")
8485
        if self.op.disk_template not in constants.DISK_TEMPLATES:
8486
          raise errors.OpPrereqError("Disk template specified in configuration"
8487
                                     " file is not one of the allowed values:"
8488
                                     " %s" % " ".join(constants.DISK_TEMPLATES))
8489
      else:
8490
        raise errors.OpPrereqError("No disk template specified and the export"
8491
                                   " is missing the disk_template information",
8492
                                   errors.ECODE_INVAL)
8493

    
8494
    if not self.op.disks:
8495
      disks = []
8496
      # TODO: import the disk iv_name too
8497
      for idx in range(constants.MAX_DISKS):
8498
        if einfo.has_option(constants.INISECT_INS, "disk%d_size" % idx):
8499
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
8500
          disks.append({constants.IDISK_SIZE: disk_sz})
8501
      self.op.disks = disks
8502
      if not disks and self.op.disk_template != constants.DT_DISKLESS:
8503
        raise errors.OpPrereqError("No disk info specified and the export"
8504
                                   " is missing the disk information",
8505
                                   errors.ECODE_INVAL)
8506

    
8507
    if not self.op.nics:
8508
      nics = []
8509
      for idx in range(constants.MAX_NICS):
8510
        if einfo.has_option(constants.INISECT_INS, "nic%d_mac" % idx):
8511
          ndict = {}
8512
          for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
8513
            v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
8514
            ndict[name] = v
8515
          nics.append(ndict)
8516
        else:
8517
          break
8518
      self.op.nics = nics
8519

    
8520
    if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
8521
      self.op.tags = einfo.get(constants.INISECT_INS, "tags").split()
8522

    
8523
    if (self.op.hypervisor is None and
8524
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
8525
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
8526

    
8527
    if einfo.has_section(constants.INISECT_HYP):
8528
      # use the export parameters but do not override the ones
8529
      # specified by the user
8530
      for name, value in einfo.items(constants.INISECT_HYP):
8531
        if name not in self.op.hvparams:
8532
          self.op.hvparams[name] = value
8533

    
8534
    if einfo.has_section(constants.INISECT_BEP):
8535
      # use the parameters, without overriding
8536
      for name, value in einfo.items(constants.INISECT_BEP):
8537
        if name not in self.op.beparams:
8538
          self.op.beparams[name] = value
8539
    else:
8540
      # try to read the parameters old style, from the main section
8541
      for name in constants.BES_PARAMETERS:
8542
        if (name not in self.op.beparams and
8543
            einfo.has_option(constants.INISECT_INS, name)):
8544
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
8545

    
8546
    if einfo.has_section(constants.INISECT_OSP):
8547
      # use the parameters, without overriding
8548
      for name, value in einfo.items(constants.INISECT_OSP):
8549
        if name not in self.op.osparams:
8550
          self.op.osparams[name] = value
8551

    
8552
  def _RevertToDefaults(self, cluster):
8553
    """Revert the instance parameters to the default values.
8554

8555
    """
8556
    # hvparams
8557
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
8558
    for name in self.op.hvparams.keys():
8559
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
8560
        del self.op.hvparams[name]
8561
    # beparams
8562
    be_defs = cluster.SimpleFillBE({})
8563
    for name in self.op.beparams.keys():
8564
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
8565
        del self.op.beparams[name]
8566
    # nic params
8567
    nic_defs = cluster.SimpleFillNIC({})
8568
    for nic in self.op.nics:
8569
      for name in constants.NICS_PARAMETERS:
8570
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
8571
          del nic[name]
8572
    # osparams
8573
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
8574
    for name in self.op.osparams.keys():
8575
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
8576
        del self.op.osparams[name]
8577

    
8578
  def _CalculateFileStorageDir(self):
8579
    """Calculate final instance file storage dir.
8580

8581
    """
8582
    # file storage dir calculation/check
8583
    self.instance_file_storage_dir = None
8584
    if self.op.disk_template in constants.DTS_FILEBASED:
8585
      # build the full file storage dir path
8586
      joinargs = []
8587

    
8588
      if self.op.disk_template == constants.DT_SHARED_FILE:
8589
        get_fsd_fn = self.cfg.GetSharedFileStorageDir
8590
      else:
8591
        get_fsd_fn = self.cfg.GetFileStorageDir
8592

    
8593
      cfg_storagedir = get_fsd_fn()
8594
      if not cfg_storagedir:
8595
        raise errors.OpPrereqError("Cluster file storage dir not defined")
8596
      joinargs.append(cfg_storagedir)
8597

    
8598
      if self.op.file_storage_dir is not None:
8599
        joinargs.append(self.op.file_storage_dir)
8600

    
8601
      joinargs.append(self.op.instance_name)
8602

    
8603
      # pylint: disable=W0142
8604
      self.instance_file_storage_dir = utils.PathJoin(*joinargs)
8605

    
8606
  def CheckPrereq(self):
8607
    """Check prerequisites.
8608

8609
    """
8610
    self._CalculateFileStorageDir()
8611

    
8612
    if self.op.mode == constants.INSTANCE_IMPORT:
8613
      export_info = self._ReadExportInfo()
8614
      self._ReadExportParams(export_info)
8615

    
8616
    if (not self.cfg.GetVGName() and
8617
        self.op.disk_template not in constants.DTS_NOT_LVM):
8618
      raise errors.OpPrereqError("Cluster does not support lvm-based"
8619
                                 " instances", errors.ECODE_STATE)
8620

    
8621
    if (self.op.hypervisor is None or
8622
        self.op.hypervisor == constants.VALUE_AUTO):
8623
      self.op.hypervisor = self.cfg.GetHypervisorType()
8624

    
8625
    cluster = self.cfg.GetClusterInfo()
8626
    enabled_hvs = cluster.enabled_hypervisors
8627
    if self.op.hypervisor not in enabled_hvs:
8628
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
8629
                                 " cluster (%s)" % (self.op.hypervisor,
8630
                                  ",".join(enabled_hvs)),
8631
                                 errors.ECODE_STATE)
8632

    
8633
    # Check tag validity
8634
    for tag in self.op.tags:
8635
      objects.TaggableObject.ValidateTag(tag)
8636

    
8637
    # check hypervisor parameter syntax (locally)
8638
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
8639
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
8640
                                      self.op.hvparams)
8641
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
8642
    hv_type.CheckParameterSyntax(filled_hvp)
8643
    self.hv_full = filled_hvp
8644
    # check that we don't specify global parameters on an instance
8645
    _CheckGlobalHvParams(self.op.hvparams)
8646

    
8647
    # fill and remember the beparams dict
8648
    default_beparams = cluster.beparams[constants.PP_DEFAULT]
8649
    for param, value in self.op.beparams.iteritems():
8650
      if value == constants.VALUE_AUTO:
8651
        self.op.beparams[param] = default_beparams[param]
8652
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
8653
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
8654

    
8655
    # build os parameters
8656
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
8657

    
8658
    # now that hvp/bep are in final format, let's reset to defaults,
8659
    # if told to do so
8660
    if self.op.identify_defaults:
8661
      self._RevertToDefaults(cluster)
8662

    
8663
    # NIC buildup
8664
    self.nics = []
8665
    for idx, nic in enumerate(self.op.nics):
8666
      nic_mode_req = nic.get(constants.INIC_MODE, None)
8667
      nic_mode = nic_mode_req
8668
      if nic_mode is None or nic_mode == constants.VALUE_AUTO:
8669
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
8670

    
8671
      # in routed mode, for the first nic, the default ip is 'auto'
8672
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
8673
        default_ip_mode = constants.VALUE_AUTO
8674
      else:
8675
        default_ip_mode = constants.VALUE_NONE
8676

    
8677
      # ip validity checks
8678
      ip = nic.get(constants.INIC_IP, default_ip_mode)
8679
      if ip is None or ip.lower() == constants.VALUE_NONE:
8680
        nic_ip = None
8681
      elif ip.lower() == constants.VALUE_AUTO:
8682
        if not self.op.name_check:
8683
          raise errors.OpPrereqError("IP address set to auto but name checks"
8684
                                     " have been skipped",
8685
                                     errors.ECODE_INVAL)
8686
        nic_ip = self.hostname1.ip
8687
      else:
8688
        if not netutils.IPAddress.IsValid(ip):
8689
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
8690
                                     errors.ECODE_INVAL)
8691
        nic_ip = ip
8692

    
8693
      # TODO: check the ip address for uniqueness
8694
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
8695
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
8696
                                   errors.ECODE_INVAL)
8697

    
8698
      # MAC address verification
8699
      mac = nic.get(constants.INIC_MAC, constants.VALUE_AUTO)
8700
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
8701
        mac = utils.NormalizeAndValidateMac(mac)
8702

    
8703
        try:
8704
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
8705
        except errors.ReservationError:
8706
          raise errors.OpPrereqError("MAC address %s already in use"
8707
                                     " in cluster" % mac,
8708
                                     errors.ECODE_NOTUNIQUE)
8709

    
8710
      #  Build nic parameters
8711
      link = nic.get(constants.INIC_LINK, None)
8712
      if link == constants.VALUE_AUTO:
8713
        link = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_LINK]
8714
      nicparams = {}
8715
      if nic_mode_req:
8716
        nicparams[constants.NIC_MODE] = nic_mode
8717
      if link:
8718
        nicparams[constants.NIC_LINK] = link
8719

    
8720
      check_params = cluster.SimpleFillNIC(nicparams)
8721
      objects.NIC.CheckParameterSyntax(check_params)
8722
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
8723

    
8724
    # disk checks/pre-build
8725
    default_vg = self.cfg.GetVGName()
8726
    self.disks = []
8727
    for disk in self.op.disks:
8728
      mode = disk.get(constants.IDISK_MODE, constants.DISK_RDWR)
8729
      if mode not in constants.DISK_ACCESS_SET:
8730
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
8731
                                   mode, errors.ECODE_INVAL)
8732
      size = disk.get(constants.IDISK_SIZE, None)
8733
      if size is None:
8734
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
8735
      try:
8736
        size = int(size)
8737
      except (TypeError, ValueError):
8738
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
8739
                                   errors.ECODE_INVAL)
8740

    
8741
      data_vg = disk.get(constants.IDISK_VG, default_vg)
8742
      new_disk = {
8743
        constants.IDISK_SIZE: size,
8744
        constants.IDISK_MODE: mode,
8745
        constants.IDISK_VG: data_vg,
8746
        constants.IDISK_METAVG: disk.get(constants.IDISK_METAVG, data_vg),
8747
        }
8748
      if constants.IDISK_ADOPT in disk:
8749
        new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
8750
      self.disks.append(new_disk)
8751

    
8752
    if self.op.mode == constants.INSTANCE_IMPORT:
8753
      disk_images = []
8754
      for idx in range(len(self.disks)):
8755
        option = "disk%d_dump" % idx
8756
        if export_info.has_option(constants.INISECT_INS, option):
8757
          # FIXME: are the old os-es, disk sizes, etc. useful?
8758
          export_name = export_info.get(constants.INISECT_INS, option)
8759
          image = utils.PathJoin(self.op.src_path, export_name)
8760
          disk_images.append(image)
8761
        else:
8762
          disk_images.append(False)
8763

    
8764
      self.src_images = disk_images
8765

    
8766
      old_name = export_info.get(constants.INISECT_INS, "name")
8767
      if self.op.instance_name == old_name:
8768
        for idx, nic in enumerate(self.nics):
8769
          if nic.mac == constants.VALUE_AUTO:
8770
            nic_mac_ini = "nic%d_mac" % idx
8771
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
8772

    
8773
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
8774

    
8775
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
8776
    if self.op.ip_check:
8777
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
8778
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
8779
                                   (self.check_ip, self.op.instance_name),
8780
                                   errors.ECODE_NOTUNIQUE)
8781

    
8782
    #### mac address generation
8783
    # By generating here the mac address both the allocator and the hooks get
8784
    # the real final mac address rather than the 'auto' or 'generate' value.
8785
    # There is a race condition between the generation and the instance object
8786
    # creation, which means that we know the mac is valid now, but we're not
8787
    # sure it will be when we actually add the instance. If things go bad
8788
    # adding the instance will abort because of a duplicate mac, and the
8789
    # creation job will fail.
8790
    for nic in self.nics:
8791
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
8792
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
8793

    
8794
    #### allocator run
8795

    
8796
    if self.op.iallocator is not None:
8797
      self._RunAllocator()
8798

    
8799
    #### node related checks
8800

    
8801
    # check primary node
8802
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
8803
    assert self.pnode is not None, \
8804
      "Cannot retrieve locked node %s" % self.op.pnode
8805
    if pnode.offline:
8806
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
8807
                                 pnode.name, errors.ECODE_STATE)
8808
    if pnode.drained:
8809
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
8810
                                 pnode.name, errors.ECODE_STATE)
8811
    if not pnode.vm_capable:
8812
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
8813
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
8814

    
8815
    self.secondaries = []
8816

    
8817
    # mirror node verification
8818
    if self.op.disk_template in constants.DTS_INT_MIRROR:
8819
      if self.op.snode == pnode.name:
8820
        raise errors.OpPrereqError("The secondary node cannot be the"
8821
                                   " primary node", errors.ECODE_INVAL)
8822
      _CheckNodeOnline(self, self.op.snode)
8823
      _CheckNodeNotDrained(self, self.op.snode)
8824
      _CheckNodeVmCapable(self, self.op.snode)
8825
      self.secondaries.append(self.op.snode)
8826

    
8827
    nodenames = [pnode.name] + self.secondaries
8828

    
8829
    if not self.adopt_disks:
8830
      # Check lv size requirements, if not adopting
8831
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
8832
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
8833

    
8834
    elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
8835
      all_lvs = set(["%s/%s" % (disk[constants.IDISK_VG],
8836
                                disk[constants.IDISK_ADOPT])
8837
                     for disk in self.disks])
8838
      if len(all_lvs) != len(self.disks):
8839
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
8840
                                   errors.ECODE_INVAL)
8841
      for lv_name in all_lvs:
8842
        try:
8843
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
8844
          # to ReserveLV uses the same syntax
8845
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
8846
        except errors.ReservationError:
8847
          raise errors.OpPrereqError("LV named %s used by another instance" %
8848
                                     lv_name, errors.ECODE_NOTUNIQUE)
8849

    
8850
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
8851
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
8852

    
8853
      node_lvs = self.rpc.call_lv_list([pnode.name],
8854
                                       vg_names.payload.keys())[pnode.name]
8855
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
8856
      node_lvs = node_lvs.payload
8857

    
8858
      delta = all_lvs.difference(node_lvs.keys())
8859
      if delta:
8860
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
8861
                                   utils.CommaJoin(delta),
8862
                                   errors.ECODE_INVAL)
8863
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
8864
      if online_lvs:
8865
        raise errors.OpPrereqError("Online logical volumes found, cannot"
8866
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
8867
                                   errors.ECODE_STATE)
8868
      # update the size of disk based on what is found
8869
      for dsk in self.disks:
8870
        dsk[constants.IDISK_SIZE] = \
8871
          int(float(node_lvs["%s/%s" % (dsk[constants.IDISK_VG],
8872
                                        dsk[constants.IDISK_ADOPT])][0]))
8873

    
8874
    elif self.op.disk_template == constants.DT_BLOCK:
8875
      # Normalize and de-duplicate device paths
8876
      all_disks = set([os.path.abspath(disk[constants.IDISK_ADOPT])
8877
                       for disk in self.disks])
8878
      if len(all_disks) != len(self.disks):
8879
        raise errors.OpPrereqError("Duplicate disk names given for adoption",
8880
                                   errors.ECODE_INVAL)
8881
      baddisks = [d for d in all_disks
8882
                  if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
8883
      if baddisks:
8884
        raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
8885
                                   " cannot be adopted" %
8886
                                   (", ".join(baddisks),
8887
                                    constants.ADOPTABLE_BLOCKDEV_ROOT),
8888
                                   errors.ECODE_INVAL)
8889

    
8890
      node_disks = self.rpc.call_bdev_sizes([pnode.name],
8891
                                            list(all_disks))[pnode.name]
8892
      node_disks.Raise("Cannot get block device information from node %s" %
8893
                       pnode.name)
8894
      node_disks = node_disks.payload
8895
      delta = all_disks.difference(node_disks.keys())
8896
      if delta:
8897
        raise errors.OpPrereqError("Missing block device(s): %s" %
8898
                                   utils.CommaJoin(delta),
8899
                                   errors.ECODE_INVAL)
8900
      for dsk in self.disks:
8901
        dsk[constants.IDISK_SIZE] = \
8902
          int(float(node_disks[dsk[constants.IDISK_ADOPT]]))
8903

    
8904
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
8905

    
8906
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
8907
    # check OS parameters (remotely)
8908
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
8909

    
8910
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
8911

    
8912
    # memory check on primary node
8913
    if self.op.start:
8914
      _CheckNodeFreeMemory(self, self.pnode.name,
8915
                           "creating instance %s" % self.op.instance_name,
8916
                           self.be_full[constants.BE_MEMORY],
8917
                           self.op.hypervisor)
8918

    
8919
    self.dry_run_result = list(nodenames)
8920

    
8921
  def Exec(self, feedback_fn):
8922
    """Create and add the instance to the cluster.
8923

8924
    """
8925
    instance = self.op.instance_name
8926
    pnode_name = self.pnode.name
8927

    
8928
    ht_kind = self.op.hypervisor
8929
    if ht_kind in constants.HTS_REQ_PORT:
8930
      network_port = self.cfg.AllocatePort()
8931
    else:
8932
      network_port = None
8933

    
8934
    disks = _GenerateDiskTemplate(self,
8935
                                  self.op.disk_template,
8936
                                  instance, pnode_name,
8937
                                  self.secondaries,
8938
                                  self.disks,
8939
                                  self.instance_file_storage_dir,
8940
                                  self.op.file_driver,
8941
                                  0,
8942
                                  feedback_fn)
8943

    
8944
    iobj = objects.Instance(name=instance, os=self.op.os_type,
8945
                            primary_node=pnode_name,
8946
                            nics=self.nics, disks=disks,
8947
                            disk_template=self.op.disk_template,
8948
                            admin_up=False,
8949
                            network_port=network_port,
8950
                            beparams=self.op.beparams,
8951
                            hvparams=self.op.hvparams,
8952
                            hypervisor=self.op.hypervisor,
8953
                            osparams=self.op.osparams,
8954
                            )
8955

    
8956
    if self.op.tags:
8957
      for tag in self.op.tags:
8958
        iobj.AddTag(tag)
8959

    
8960
    if self.adopt_disks:
8961
      if self.op.disk_template == constants.DT_PLAIN:
8962
        # rename LVs to the newly-generated names; we need to construct
8963
        # 'fake' LV disks with the old data, plus the new unique_id
8964
        tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
8965
        rename_to = []
8966
        for t_dsk, a_dsk in zip(tmp_disks, self.disks):
8967
          rename_to.append(t_dsk.logical_id)
8968
          t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
8969
          self.cfg.SetDiskID(t_dsk, pnode_name)
8970
        result = self.rpc.call_blockdev_rename(pnode_name,
8971
                                               zip(tmp_disks, rename_to))
8972
        result.Raise("Failed to rename adoped LVs")
8973
    else:
8974
      feedback_fn("* creating instance disks...")
8975
      try:
8976
        _CreateDisks(self, iobj)
8977
      except errors.OpExecError:
8978
        self.LogWarning("Device creation failed, reverting...")
8979
        try:
8980
          _RemoveDisks(self, iobj)
8981
        finally:
8982
          self.cfg.ReleaseDRBDMinors(instance)
8983
          raise
8984

    
8985
    feedback_fn("adding instance %s to cluster config" % instance)
8986

    
8987
    self.cfg.AddInstance(iobj, self.proc.GetECId())
8988

    
8989
    # Declare that we don't want to remove the instance lock anymore, as we've
8990
    # added the instance to the config
8991
    del self.remove_locks[locking.LEVEL_INSTANCE]
8992

    
8993
    if self.op.mode == constants.INSTANCE_IMPORT:
8994
      # Release unused nodes
8995
      _ReleaseLocks(self, locking.LEVEL_NODE, keep=[self.op.src_node])
8996
    else:
8997
      # Release all nodes
8998
      _ReleaseLocks(self, locking.LEVEL_NODE)
8999

    
9000
    disk_abort = False
9001
    if not self.adopt_disks and self.cfg.GetClusterInfo().prealloc_wipe_disks:
9002
      feedback_fn("* wiping instance disks...")
9003
      try:
9004
        _WipeDisks(self, iobj)
9005
      except errors.OpExecError, err:
9006
        logging.exception("Wiping disks failed")
9007
        self.LogWarning("Wiping instance disks failed (%s)", err)
9008
        disk_abort = True
9009

    
9010
    if disk_abort:
9011
      # Something is already wrong with the disks, don't do anything else
9012
      pass
9013
    elif self.op.wait_for_sync:
9014
      disk_abort = not _WaitForSync(self, iobj)
9015
    elif iobj.disk_template in constants.DTS_INT_MIRROR:
9016
      # make sure the disks are not degraded (still sync-ing is ok)
9017
      feedback_fn("* checking mirrors status")
9018
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
9019
    else:
9020
      disk_abort = False
9021

    
9022
    if disk_abort:
9023
      _RemoveDisks(self, iobj)
9024
      self.cfg.RemoveInstance(iobj.name)
9025
      # Make sure the instance lock gets removed
9026
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
9027
      raise errors.OpExecError("There are some degraded disks for"
9028
                               " this instance")
9029

    
9030
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
9031
      if self.op.mode == constants.INSTANCE_CREATE:
9032
        if not self.op.no_install:
9033
          pause_sync = (iobj.disk_template in constants.DTS_INT_MIRROR and
9034
                        not self.op.wait_for_sync)
9035
          if pause_sync:
9036
            feedback_fn("* pausing disk sync to install instance OS")
9037
            result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
9038
                                                              iobj.disks, True)
9039
            for idx, success in enumerate(result.payload):
9040
              if not success:
9041
                logging.warn("pause-sync of instance %s for disk %d failed",
9042
                             instance, idx)
9043

    
9044
          feedback_fn("* running the instance OS create scripts...")
9045
          # FIXME: pass debug option from opcode to backend
9046
          os_add_result = \
9047
            self.rpc.call_instance_os_add(pnode_name, iobj, False,
9048
                                          self.op.debug_level)
9049
          if pause_sync:
9050
            feedback_fn("* resuming disk sync")
9051
            result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
9052
                                                              iobj.disks, False)
9053
            for idx, success in enumerate(result.payload):
9054
              if not success:
9055
                logging.warn("resume-sync of instance %s for disk %d failed",
9056
                             instance, idx)
9057

    
9058
          os_add_result.Raise("Could not add os for instance %s"
9059
                              " on node %s" % (instance, pnode_name))
9060

    
9061
      elif self.op.mode == constants.INSTANCE_IMPORT:
9062
        feedback_fn("* running the instance OS import scripts...")
9063

    
9064
        transfers = []
9065

    
9066
        for idx, image in enumerate(self.src_images):
9067
          if not image:
9068
            continue
9069

    
9070
          # FIXME: pass debug option from opcode to backend
9071
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
9072
                                             constants.IEIO_FILE, (image, ),
9073
                                             constants.IEIO_SCRIPT,
9074
                                             (iobj.disks[idx], idx),
9075
                                             None)
9076
          transfers.append(dt)
9077

    
9078
        import_result = \
9079
          masterd.instance.TransferInstanceData(self, feedback_fn,
9080
                                                self.op.src_node, pnode_name,
9081
                                                self.pnode.secondary_ip,
9082
                                                iobj, transfers)
9083
        if not compat.all(import_result):
9084
          self.LogWarning("Some disks for instance %s on node %s were not"
9085
                          " imported successfully" % (instance, pnode_name))
9086

    
9087
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
9088
        feedback_fn("* preparing remote import...")
9089
        # The source cluster will stop the instance before attempting to make a
9090
        # connection. In some cases stopping an instance can take a long time,
9091
        # hence the shutdown timeout is added to the connection timeout.
9092
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
9093
                           self.op.source_shutdown_timeout)
9094
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
9095

    
9096
        assert iobj.primary_node == self.pnode.name
9097
        disk_results = \
9098
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
9099
                                        self.source_x509_ca,
9100
                                        self._cds, timeouts)
9101
        if not compat.all(disk_results):
9102
          # TODO: Should the instance still be started, even if some disks
9103
          # failed to import (valid for local imports, too)?
9104
          self.LogWarning("Some disks for instance %s on node %s were not"
9105
                          " imported successfully" % (instance, pnode_name))
9106

    
9107
        # Run rename script on newly imported instance
9108
        assert iobj.name == instance
9109
        feedback_fn("Running rename script for %s" % instance)
9110
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
9111
                                                   self.source_instance_name,
9112
                                                   self.op.debug_level)
9113
        if result.fail_msg:
9114
          self.LogWarning("Failed to run rename script for %s on node"
9115
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
9116

    
9117
      else:
9118
        # also checked in the prereq part
9119
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
9120
                                     % self.op.mode)
9121

    
9122
    if self.op.start:
9123
      iobj.admin_up = True
9124
      self.cfg.Update(iobj, feedback_fn)
9125
      logging.info("Starting instance %s on node %s", instance, pnode_name)
9126
      feedback_fn("* starting instance...")
9127
      result = self.rpc.call_instance_start(pnode_name, iobj,
9128
                                            None, None, False)
9129
      result.Raise("Could not start instance")
9130

    
9131
    return list(iobj.all_nodes)
9132

    
9133

    
9134
class LUInstanceConsole(NoHooksLU):
9135
  """Connect to an instance's console.
9136

9137
  This is somewhat special in that it returns the command line that
9138
  you need to run on the master node in order to connect to the
9139
  console.
9140

9141
  """
9142
  REQ_BGL = False
9143

    
9144
  def ExpandNames(self):
9145
    self._ExpandAndLockInstance()
9146

    
9147
  def CheckPrereq(self):
9148
    """Check prerequisites.
9149

9150
    This checks that the instance is in the cluster.
9151

9152
    """
9153
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9154
    assert self.instance is not None, \
9155
      "Cannot retrieve locked instance %s" % self.op.instance_name
9156
    _CheckNodeOnline(self, self.instance.primary_node)
9157

    
9158
  def Exec(self, feedback_fn):
9159
    """Connect to the console of an instance
9160

9161
    """
9162
    instance = self.instance
9163
    node = instance.primary_node
9164

    
9165
    node_insts = self.rpc.call_instance_list([node],
9166
                                             [instance.hypervisor])[node]
9167
    node_insts.Raise("Can't get node information from %s" % node)
9168

    
9169
    if instance.name not in node_insts.payload:
9170
      if instance.admin_up:
9171
        state = constants.INSTST_ERRORDOWN
9172
      else:
9173
        state = constants.INSTST_ADMINDOWN
9174
      raise errors.OpExecError("Instance %s is not running (state %s)" %
9175
                               (instance.name, state))
9176

    
9177
    logging.debug("Connecting to console of %s on %s", instance.name, node)
9178

    
9179
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
9180

    
9181

    
9182
def _GetInstanceConsole(cluster, instance):
9183
  """Returns console information for an instance.
9184

9185
  @type cluster: L{objects.Cluster}
9186
  @type instance: L{objects.Instance}
9187
  @rtype: dict
9188

9189
  """
9190
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
9191
  # beparams and hvparams are passed separately, to avoid editing the
9192
  # instance and then saving the defaults in the instance itself.
9193
  hvparams = cluster.FillHV(instance)
9194
  beparams = cluster.FillBE(instance)
9195
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
9196

    
9197
  assert console.instance == instance.name
9198
  assert console.Validate()
9199

    
9200
  return console.ToDict()
9201

    
9202

    
9203
class LUInstanceReplaceDisks(LogicalUnit):
9204
  """Replace the disks of an instance.
9205

9206
  """
9207
  HPATH = "mirrors-replace"
9208
  HTYPE = constants.HTYPE_INSTANCE
9209
  REQ_BGL = False
9210

    
9211
  def CheckArguments(self):
9212
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
9213
                                  self.op.iallocator)
9214

    
9215
  def ExpandNames(self):
9216
    self._ExpandAndLockInstance()
9217

    
9218
    assert locking.LEVEL_NODE not in self.needed_locks
9219
    assert locking.LEVEL_NODEGROUP not in self.needed_locks
9220

    
9221
    assert self.op.iallocator is None or self.op.remote_node is None, \
9222
      "Conflicting options"
9223

    
9224
    if self.op.remote_node is not None:
9225
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9226

    
9227
      # Warning: do not remove the locking of the new secondary here
9228
      # unless DRBD8.AddChildren is changed to work in parallel;
9229
      # currently it doesn't since parallel invocations of
9230
      # FindUnusedMinor will conflict
9231
      self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
9232
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
9233
    else:
9234
      self.needed_locks[locking.LEVEL_NODE] = []
9235
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9236

    
9237
      if self.op.iallocator is not None:
9238
        # iallocator will select a new node in the same group
9239
        self.needed_locks[locking.LEVEL_NODEGROUP] = []
9240

    
9241
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
9242
                                   self.op.iallocator, self.op.remote_node,
9243
                                   self.op.disks, False, self.op.early_release)
9244

    
9245
    self.tasklets = [self.replacer]
9246

    
9247
  def DeclareLocks(self, level):
9248
    if level == locking.LEVEL_NODEGROUP:
9249
      assert self.op.remote_node is None
9250
      assert self.op.iallocator is not None
9251
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
9252

    
9253
      self.share_locks[locking.LEVEL_NODEGROUP] = 1
9254
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
9255
        self.cfg.GetInstanceNodeGroups(self.op.instance_name)
9256

    
9257
    elif level == locking.LEVEL_NODE:
9258
      if self.op.iallocator is not None:
9259
        assert self.op.remote_node is None
9260
        assert not self.needed_locks[locking.LEVEL_NODE]
9261

    
9262
        # Lock member nodes of all locked groups
9263
        self.needed_locks[locking.LEVEL_NODE] = [node_name
9264
          for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
9265
          for node_name in self.cfg.GetNodeGroup(group_uuid).members]
9266
      else:
9267
        self._LockInstancesNodes()
9268

    
9269
  def BuildHooksEnv(self):
9270
    """Build hooks env.
9271

9272
    This runs on the master, the primary and all the secondaries.
9273

9274
    """
9275
    instance = self.replacer.instance
9276
    env = {
9277
      "MODE": self.op.mode,
9278
      "NEW_SECONDARY": self.op.remote_node,
9279
      "OLD_SECONDARY": instance.secondary_nodes[0],
9280
      }
9281
    env.update(_BuildInstanceHookEnvByObject(self, instance))
9282
    return env
9283

    
9284
  def BuildHooksNodes(self):
9285
    """Build hooks nodes.
9286

9287
    """
9288
    instance = self.replacer.instance
9289
    nl = [
9290
      self.cfg.GetMasterNode(),
9291
      instance.primary_node,
9292
      ]
9293
    if self.op.remote_node is not None:
9294
      nl.append(self.op.remote_node)
9295
    return nl, nl
9296

    
9297
  def CheckPrereq(self):
9298
    """Check prerequisites.
9299

9300
    """
9301
    assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
9302
            self.op.iallocator is None)
9303

    
9304
    owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
9305
    if owned_groups:
9306
      _CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
9307

    
9308
    return LogicalUnit.CheckPrereq(self)
9309

    
9310

    
9311
class TLReplaceDisks(Tasklet):
9312
  """Replaces disks for an instance.
9313

9314
  Note: Locking is not within the scope of this class.
9315

9316
  """
9317
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
9318
               disks, delay_iallocator, early_release):
9319
    """Initializes this class.
9320

9321
    """
9322
    Tasklet.__init__(self, lu)
9323

    
9324
    # Parameters
9325
    self.instance_name = instance_name
9326
    self.mode = mode
9327
    self.iallocator_name = iallocator_name
9328
    self.remote_node = remote_node
9329
    self.disks = disks
9330
    self.delay_iallocator = delay_iallocator
9331
    self.early_release = early_release
9332

    
9333
    # Runtime data
9334
    self.instance = None
9335
    self.new_node = None
9336
    self.target_node = None
9337
    self.other_node = None
9338
    self.remote_node_info = None
9339
    self.node_secondary_ip = None
9340

    
9341
  @staticmethod
9342
  def CheckArguments(mode, remote_node, iallocator):
9343
    """Helper function for users of this class.
9344

9345
    """
9346
    # check for valid parameter combination
9347
    if mode == constants.REPLACE_DISK_CHG:
9348
      if remote_node is None and iallocator is None:
9349
        raise errors.OpPrereqError("When changing the secondary either an"
9350
                                   " iallocator script must be used or the"
9351
                                   " new node given", errors.ECODE_INVAL)
9352

    
9353
      if remote_node is not None and iallocator is not None:
9354
        raise errors.OpPrereqError("Give either the iallocator or the new"
9355
                                   " secondary, not both", errors.ECODE_INVAL)
9356

    
9357
    elif remote_node is not None or iallocator is not None:
9358
      # Not replacing the secondary
9359
      raise errors.OpPrereqError("The iallocator and new node options can"
9360
                                 " only be used when changing the"
9361
                                 " secondary node", errors.ECODE_INVAL)
9362

    
9363
  @staticmethod
9364
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
9365
    """Compute a new secondary node using an IAllocator.
9366

9367
    """
9368
    ial = IAllocator(lu.cfg, lu.rpc,
9369
                     mode=constants.IALLOCATOR_MODE_RELOC,
9370
                     name=instance_name,
9371
                     relocate_from=list(relocate_from))
9372

    
9373
    ial.Run(iallocator_name)
9374

    
9375
    if not ial.success:
9376
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
9377
                                 " %s" % (iallocator_name, ial.info),
9378
                                 errors.ECODE_NORES)
9379

    
9380
    if len(ial.result) != ial.required_nodes:
9381
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
9382
                                 " of nodes (%s), required %s" %
9383
                                 (iallocator_name,
9384
                                  len(ial.result), ial.required_nodes),
9385
                                 errors.ECODE_FAULT)
9386

    
9387
    remote_node_name = ial.result[0]
9388

    
9389
    lu.LogInfo("Selected new secondary for instance '%s': %s",
9390
               instance_name, remote_node_name)
9391

    
9392
    return remote_node_name
9393

    
9394
  def _FindFaultyDisks(self, node_name):
9395
    """Wrapper for L{_FindFaultyInstanceDisks}.
9396

9397
    """
9398
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
9399
                                    node_name, True)
9400

    
9401
  def _CheckDisksActivated(self, instance):
9402
    """Checks if the instance disks are activated.
9403

9404
    @param instance: The instance to check disks
9405
    @return: True if they are activated, False otherwise
9406

9407
    """
9408
    nodes = instance.all_nodes
9409

    
9410
    for idx, dev in enumerate(instance.disks):
9411
      for node in nodes:
9412
        self.lu.LogInfo("Checking disk/%d on %s", idx, node)
9413
        self.cfg.SetDiskID(dev, node)
9414

    
9415
        result = self.rpc.call_blockdev_find(node, dev)
9416

    
9417
        if result.offline:
9418
          continue
9419
        elif result.fail_msg or not result.payload:
9420
          return False
9421

    
9422
    return True
9423

    
9424
  def CheckPrereq(self):
9425
    """Check prerequisites.
9426

9427
    This checks that the instance is in the cluster.
9428

9429
    """
9430
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
9431
    assert instance is not None, \
9432
      "Cannot retrieve locked instance %s" % self.instance_name
9433

    
9434
    if instance.disk_template != constants.DT_DRBD8:
9435
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
9436
                                 " instances", errors.ECODE_INVAL)
9437

    
9438
    if len(instance.secondary_nodes) != 1:
9439
      raise errors.OpPrereqError("The instance has a strange layout,"
9440
                                 " expected one secondary but found %d" %
9441
                                 len(instance.secondary_nodes),
9442
                                 errors.ECODE_FAULT)
9443

    
9444
    if not self.delay_iallocator:
9445
      self._CheckPrereq2()
9446

    
9447
  def _CheckPrereq2(self):
9448
    """Check prerequisites, second part.
9449

9450
    This function should always be part of CheckPrereq. It was separated and is
9451
    now called from Exec because during node evacuation iallocator was only
9452
    called with an unmodified cluster model, not taking planned changes into
9453
    account.
9454

9455
    """
9456
    instance = self.instance
9457
    secondary_node = instance.secondary_nodes[0]
9458

    
9459
    if self.iallocator_name is None:
9460
      remote_node = self.remote_node
9461
    else:
9462
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
9463
                                       instance.name, instance.secondary_nodes)
9464

    
9465
    if remote_node is None:
9466
      self.remote_node_info = None
9467
    else:
9468
      assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
9469
             "Remote node '%s' is not locked" % remote_node
9470

    
9471
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
9472
      assert self.remote_node_info is not None, \
9473
        "Cannot retrieve locked node %s" % remote_node
9474

    
9475
    if remote_node == self.instance.primary_node:
9476
      raise errors.OpPrereqError("The specified node is the primary node of"
9477
                                 " the instance", errors.ECODE_INVAL)
9478

    
9479
    if remote_node == secondary_node:
9480
      raise errors.OpPrereqError("The specified node is already the"
9481
                                 " secondary node of the instance",
9482
                                 errors.ECODE_INVAL)
9483

    
9484
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
9485
                                    constants.REPLACE_DISK_CHG):
9486
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
9487
                                 errors.ECODE_INVAL)
9488

    
9489
    if self.mode == constants.REPLACE_DISK_AUTO:
9490
      if not self._CheckDisksActivated(instance):
9491
        raise errors.OpPrereqError("Please run activate-disks on instance %s"
9492
                                   " first" % self.instance_name,
9493
                                   errors.ECODE_STATE)
9494
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
9495
      faulty_secondary = self._FindFaultyDisks(secondary_node)
9496

    
9497
      if faulty_primary and faulty_secondary:
9498
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
9499
                                   " one node and can not be repaired"
9500
                                   " automatically" % self.instance_name,
9501
                                   errors.ECODE_STATE)
9502

    
9503
      if faulty_primary:
9504
        self.disks = faulty_primary
9505
        self.target_node = instance.primary_node
9506
        self.other_node = secondary_node
9507
        check_nodes = [self.target_node, self.other_node]
9508
      elif faulty_secondary:
9509
        self.disks = faulty_secondary
9510
        self.target_node = secondary_node
9511
        self.other_node = instance.primary_node
9512
        check_nodes = [self.target_node, self.other_node]
9513
      else:
9514
        self.disks = []
9515
        check_nodes = []
9516

    
9517
    else:
9518
      # Non-automatic modes
9519
      if self.mode == constants.REPLACE_DISK_PRI:
9520
        self.target_node = instance.primary_node
9521
        self.other_node = secondary_node
9522
        check_nodes = [self.target_node, self.other_node]
9523

    
9524
      elif self.mode == constants.REPLACE_DISK_SEC:
9525
        self.target_node = secondary_node
9526
        self.other_node = instance.primary_node
9527
        check_nodes = [self.target_node, self.other_node]
9528

    
9529
      elif self.mode == constants.REPLACE_DISK_CHG:
9530
        self.new_node = remote_node
9531
        self.other_node = instance.primary_node
9532
        self.target_node = secondary_node
9533
        check_nodes = [self.new_node, self.other_node]
9534

    
9535
        _CheckNodeNotDrained(self.lu, remote_node)
9536
        _CheckNodeVmCapable(self.lu, remote_node)
9537

    
9538
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
9539
        assert old_node_info is not None
9540
        if old_node_info.offline and not self.early_release:
9541
          # doesn't make sense to delay the release
9542
          self.early_release = True
9543
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
9544
                          " early-release mode", secondary_node)
9545

    
9546
      else:
9547
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
9548
                                     self.mode)
9549

    
9550
      # If not specified all disks should be replaced
9551
      if not self.disks:
9552
        self.disks = range(len(self.instance.disks))
9553

    
9554
    for node in check_nodes:
9555
      _CheckNodeOnline(self.lu, node)
9556

    
9557
    touched_nodes = frozenset(node_name for node_name in [self.new_node,
9558
                                                          self.other_node,
9559
                                                          self.target_node]
9560
                              if node_name is not None)
9561

    
9562
    # Release unneeded node locks
9563
    _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
9564

    
9565
    # Release any owned node group
9566
    if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
9567
      _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
9568

    
9569
    # Check whether disks are valid
9570
    for disk_idx in self.disks:
9571
      instance.FindDisk(disk_idx)
9572

    
9573
    # Get secondary node IP addresses
9574
    self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
9575
                                  in self.cfg.GetMultiNodeInfo(touched_nodes))
9576

    
9577
  def Exec(self, feedback_fn):
9578
    """Execute disk replacement.
9579

9580
    This dispatches the disk replacement to the appropriate handler.
9581

9582
    """
9583
    if self.delay_iallocator:
9584
      self._CheckPrereq2()
9585

    
9586
    if __debug__:
9587
      # Verify owned locks before starting operation
9588
      owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
9589
      assert set(owned_nodes) == set(self.node_secondary_ip), \
9590
          ("Incorrect node locks, owning %s, expected %s" %
9591
           (owned_nodes, self.node_secondary_ip.keys()))
9592

    
9593
      owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
9594
      assert list(owned_instances) == [self.instance_name], \
9595
          "Instance '%s' not locked" % self.instance_name
9596

    
9597
      assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
9598
          "Should not own any node group lock at this point"
9599

    
9600
    if not self.disks:
9601
      feedback_fn("No disks need replacement")
9602
      return
9603

    
9604
    feedback_fn("Replacing disk(s) %s for %s" %
9605
                (utils.CommaJoin(self.disks), self.instance.name))
9606

    
9607
    activate_disks = (not self.instance.admin_up)
9608

    
9609
    # Activate the instance disks if we're replacing them on a down instance
9610
    if activate_disks:
9611
      _StartInstanceDisks(self.lu, self.instance, True)
9612

    
9613
    try:
9614
      # Should we replace the secondary node?
9615
      if self.new_node is not None:
9616
        fn = self._ExecDrbd8Secondary
9617
      else:
9618
        fn = self._ExecDrbd8DiskOnly
9619

    
9620
      result = fn(feedback_fn)
9621
    finally:
9622
      # Deactivate the instance disks if we're replacing them on a
9623
      # down instance
9624
      if activate_disks:
9625
        _SafeShutdownInstanceDisks(self.lu, self.instance)
9626

    
9627
    if __debug__:
9628
      # Verify owned locks
9629
      owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
9630
      nodes = frozenset(self.node_secondary_ip)
9631
      assert ((self.early_release and not owned_nodes) or
9632
              (not self.early_release and not (set(owned_nodes) - nodes))), \
9633
        ("Not owning the correct locks, early_release=%s, owned=%r,"
9634
         " nodes=%r" % (self.early_release, owned_nodes, nodes))
9635

    
9636
    return result
9637

    
9638
  def _CheckVolumeGroup(self, nodes):
9639
    self.lu.LogInfo("Checking volume groups")
9640

    
9641
    vgname = self.cfg.GetVGName()
9642

    
9643
    # Make sure volume group exists on all involved nodes
9644
    results = self.rpc.call_vg_list(nodes)
9645
    if not results:
9646
      raise errors.OpExecError("Can't list volume groups on the nodes")
9647

    
9648
    for node in nodes:
9649
      res = results[node]
9650
      res.Raise("Error checking node %s" % node)
9651
      if vgname not in res.payload:
9652
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
9653
                                 (vgname, node))
9654

    
9655
  def _CheckDisksExistence(self, nodes):
9656
    # Check disk existence
9657
    for idx, dev in enumerate(self.instance.disks):
9658
      if idx not in self.disks:
9659
        continue
9660

    
9661
      for node in nodes:
9662
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
9663
        self.cfg.SetDiskID(dev, node)
9664

    
9665
        result = self.rpc.call_blockdev_find(node, dev)
9666

    
9667
        msg = result.fail_msg
9668
        if msg or not result.payload:
9669
          if not msg:
9670
            msg = "disk not found"
9671
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
9672
                                   (idx, node, msg))
9673

    
9674
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
9675
    for idx, dev in enumerate(self.instance.disks):
9676
      if idx not in self.disks:
9677
        continue
9678

    
9679
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
9680
                      (idx, node_name))
9681

    
9682
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
9683
                                   ldisk=ldisk):
9684
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
9685
                                 " replace disks for instance %s" %
9686
                                 (node_name, self.instance.name))
9687

    
9688
  def _CreateNewStorage(self, node_name):
9689
    """Create new storage on the primary or secondary node.
9690

9691
    This is only used for same-node replaces, not for changing the
9692
    secondary node, hence we don't want to modify the existing disk.
9693

9694
    """
9695
    iv_names = {}
9696

    
9697
    for idx, dev in enumerate(self.instance.disks):
9698
      if idx not in self.disks:
9699
        continue
9700

    
9701
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
9702

    
9703
      self.cfg.SetDiskID(dev, node_name)
9704

    
9705
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
9706
      names = _GenerateUniqueNames(self.lu, lv_names)
9707

    
9708
      vg_data = dev.children[0].logical_id[0]
9709
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
9710
                             logical_id=(vg_data, names[0]))
9711
      vg_meta = dev.children[1].logical_id[0]
9712
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
9713
                             logical_id=(vg_meta, names[1]))
9714

    
9715
      new_lvs = [lv_data, lv_meta]
9716
      old_lvs = [child.Copy() for child in dev.children]
9717
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
9718

    
9719
      # we pass force_create=True to force the LVM creation
9720
      for new_lv in new_lvs:
9721
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
9722
                        _GetInstanceInfoText(self.instance), False)
9723

    
9724
    return iv_names
9725

    
9726
  def _CheckDevices(self, node_name, iv_names):
9727
    for name, (dev, _, _) in iv_names.iteritems():
9728
      self.cfg.SetDiskID(dev, node_name)
9729

    
9730
      result = self.rpc.call_blockdev_find(node_name, dev)
9731

    
9732
      msg = result.fail_msg
9733
      if msg or not result.payload:
9734
        if not msg:
9735
          msg = "disk not found"
9736
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
9737
                                 (name, msg))
9738

    
9739
      if result.payload.is_degraded:
9740
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
9741

    
9742
  def _RemoveOldStorage(self, node_name, iv_names):
9743
    for name, (_, old_lvs, _) in iv_names.iteritems():
9744
      self.lu.LogInfo("Remove logical volumes for %s" % name)
9745

    
9746
      for lv in old_lvs:
9747
        self.cfg.SetDiskID(lv, node_name)
9748

    
9749
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
9750
        if msg:
9751
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
9752
                             hint="remove unused LVs manually")
9753

    
9754
  def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
9755
    """Replace a disk on the primary or secondary for DRBD 8.
9756

9757
    The algorithm for replace is quite complicated:
9758

9759
      1. for each disk to be replaced:
9760

9761
        1. create new LVs on the target node with unique names
9762
        1. detach old LVs from the drbd device
9763
        1. rename old LVs to name_replaced.<time_t>
9764
        1. rename new LVs to old LVs
9765
        1. attach the new LVs (with the old names now) to the drbd device
9766

9767
      1. wait for sync across all devices
9768

9769
      1. for each modified disk:
9770

9771
        1. remove old LVs (which have the name name_replaces.<time_t>)
9772

9773
    Failures are not very well handled.
9774

9775
    """
9776
    steps_total = 6
9777

    
9778
    # Step: check device activation
9779
    self.lu.LogStep(1, steps_total, "Check device existence")
9780
    self._CheckDisksExistence([self.other_node, self.target_node])
9781
    self._CheckVolumeGroup([self.target_node, self.other_node])
9782

    
9783
    # Step: check other node consistency
9784
    self.lu.LogStep(2, steps_total, "Check peer consistency")
9785
    self._CheckDisksConsistency(self.other_node,
9786
                                self.other_node == self.instance.primary_node,
9787
                                False)
9788

    
9789
    # Step: create new storage
9790
    self.lu.LogStep(3, steps_total, "Allocate new storage")
9791
    iv_names = self._CreateNewStorage(self.target_node)
9792

    
9793
    # Step: for each lv, detach+rename*2+attach
9794
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
9795
    for dev, old_lvs, new_lvs in iv_names.itervalues():
9796
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
9797

    
9798
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
9799
                                                     old_lvs)
9800
      result.Raise("Can't detach drbd from local storage on node"
9801
                   " %s for device %s" % (self.target_node, dev.iv_name))
9802
      #dev.children = []
9803
      #cfg.Update(instance)
9804

    
9805
      # ok, we created the new LVs, so now we know we have the needed
9806
      # storage; as such, we proceed on the target node to rename
9807
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
9808
      # using the assumption that logical_id == physical_id (which in
9809
      # turn is the unique_id on that node)
9810

    
9811
      # FIXME(iustin): use a better name for the replaced LVs
9812
      temp_suffix = int(time.time())
9813
      ren_fn = lambda d, suff: (d.physical_id[0],
9814
                                d.physical_id[1] + "_replaced-%s" % suff)
9815

    
9816
      # Build the rename list based on what LVs exist on the node
9817
      rename_old_to_new = []
9818
      for to_ren in old_lvs:
9819
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
9820
        if not result.fail_msg and result.payload:
9821
          # device exists
9822
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
9823

    
9824
      self.lu.LogInfo("Renaming the old LVs on the target node")
9825
      result = self.rpc.call_blockdev_rename(self.target_node,
9826
                                             rename_old_to_new)
9827
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
9828

    
9829
      # Now we rename the new LVs to the old LVs
9830
      self.lu.LogInfo("Renaming the new LVs on the target node")
9831
      rename_new_to_old = [(new, old.physical_id)
9832
                           for old, new in zip(old_lvs, new_lvs)]
9833
      result = self.rpc.call_blockdev_rename(self.target_node,
9834
                                             rename_new_to_old)
9835
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
9836

    
9837
      # Intermediate steps of in memory modifications
9838
      for old, new in zip(old_lvs, new_lvs):
9839
        new.logical_id = old.logical_id
9840
        self.cfg.SetDiskID(new, self.target_node)
9841

    
9842
      # We need to modify old_lvs so that removal later removes the
9843
      # right LVs, not the newly added ones; note that old_lvs is a
9844
      # copy here
9845
      for disk in old_lvs:
9846
        disk.logical_id = ren_fn(disk, temp_suffix)
9847
        self.cfg.SetDiskID(disk, self.target_node)
9848

    
9849
      # Now that the new lvs have the old name, we can add them to the device
9850
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
9851
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
9852
                                                  new_lvs)
9853
      msg = result.fail_msg
9854
      if msg:
9855
        for new_lv in new_lvs:
9856
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
9857
                                               new_lv).fail_msg
9858
          if msg2:
9859
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
9860
                               hint=("cleanup manually the unused logical"
9861
                                     "volumes"))
9862
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
9863

    
9864
    cstep = 5
9865
    if self.early_release:
9866
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
9867
      cstep += 1
9868
      self._RemoveOldStorage(self.target_node, iv_names)
9869
      # WARNING: we release both node locks here, do not do other RPCs
9870
      # than WaitForSync to the primary node
9871
      _ReleaseLocks(self.lu, locking.LEVEL_NODE,
9872
                    names=[self.target_node, self.other_node])
9873

    
9874
    # Wait for sync
9875
    # This can fail as the old devices are degraded and _WaitForSync
9876
    # does a combined result over all disks, so we don't check its return value
9877
    self.lu.LogStep(cstep, steps_total, "Sync devices")
9878
    cstep += 1
9879
    _WaitForSync(self.lu, self.instance)
9880

    
9881
    # Check all devices manually
9882
    self._CheckDevices(self.instance.primary_node, iv_names)
9883

    
9884
    # Step: remove old storage
9885
    if not self.early_release:
9886
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
9887
      cstep += 1
9888
      self._RemoveOldStorage(self.target_node, iv_names)
9889

    
9890
  def _ExecDrbd8Secondary(self, feedback_fn):
9891
    """Replace the secondary node for DRBD 8.
9892

9893
    The algorithm for replace is quite complicated:
9894
      - for all disks of the instance:
9895
        - create new LVs on the new node with same names
9896
        - shutdown the drbd device on the old secondary
9897
        - disconnect the drbd network on the primary
9898
        - create the drbd device on the new secondary
9899
        - network attach the drbd on the primary, using an artifice:
9900
          the drbd code for Attach() will connect to the network if it
9901
          finds a device which is connected to the good local disks but
9902
          not network enabled
9903
      - wait for sync across all devices
9904
      - remove all disks from the old secondary
9905

9906
    Failures are not very well handled.
9907

9908
    """
9909
    steps_total = 6
9910

    
9911
    pnode = self.instance.primary_node
9912

    
9913
    # Step: check device activation
9914
    self.lu.LogStep(1, steps_total, "Check device existence")
9915
    self._CheckDisksExistence([self.instance.primary_node])
9916
    self._CheckVolumeGroup([self.instance.primary_node])
9917

    
9918
    # Step: check other node consistency
9919
    self.lu.LogStep(2, steps_total, "Check peer consistency")
9920
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
9921

    
9922
    # Step: create new storage
9923
    self.lu.LogStep(3, steps_total, "Allocate new storage")
9924
    for idx, dev in enumerate(self.instance.disks):
9925
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
9926
                      (self.new_node, idx))
9927
      # we pass force_create=True to force LVM creation
9928
      for new_lv in dev.children:
9929
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
9930
                        _GetInstanceInfoText(self.instance), False)
9931

    
9932
    # Step 4: dbrd minors and drbd setups changes
9933
    # after this, we must manually remove the drbd minors on both the
9934
    # error and the success paths
9935
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
9936
    minors = self.cfg.AllocateDRBDMinor([self.new_node
9937
                                         for dev in self.instance.disks],
9938
                                        self.instance.name)
9939
    logging.debug("Allocated minors %r", minors)
9940

    
9941
    iv_names = {}
9942
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
9943
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
9944
                      (self.new_node, idx))
9945
      # create new devices on new_node; note that we create two IDs:
9946
      # one without port, so the drbd will be activated without
9947
      # networking information on the new node at this stage, and one
9948
      # with network, for the latter activation in step 4
9949
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
9950
      if self.instance.primary_node == o_node1:
9951
        p_minor = o_minor1
9952
      else:
9953
        assert self.instance.primary_node == o_node2, "Three-node instance?"
9954
        p_minor = o_minor2
9955

    
9956
      new_alone_id = (self.instance.primary_node, self.new_node, None,
9957
                      p_minor, new_minor, o_secret)
9958
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
9959
                    p_minor, new_minor, o_secret)
9960

    
9961
      iv_names[idx] = (dev, dev.children, new_net_id)
9962
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
9963
                    new_net_id)
9964
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
9965
                              logical_id=new_alone_id,
9966
                              children=dev.children,
9967
                              size=dev.size)
9968
      try:
9969
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
9970
                              _GetInstanceInfoText(self.instance), False)
9971
      except errors.GenericError:
9972
        self.cfg.ReleaseDRBDMinors(self.instance.name)
9973
        raise
9974

    
9975
    # We have new devices, shutdown the drbd on the old secondary
9976
    for idx, dev in enumerate(self.instance.disks):
9977
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
9978
      self.cfg.SetDiskID(dev, self.target_node)
9979
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
9980
      if msg:
9981
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
9982
                           "node: %s" % (idx, msg),
9983
                           hint=("Please cleanup this device manually as"
9984
                                 " soon as possible"))
9985

    
9986
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
9987
    result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
9988
                                               self.instance.disks)[pnode]
9989

    
9990
    msg = result.fail_msg
9991
    if msg:
9992
      # detaches didn't succeed (unlikely)
9993
      self.cfg.ReleaseDRBDMinors(self.instance.name)
9994
      raise errors.OpExecError("Can't detach the disks from the network on"
9995
                               " old node: %s" % (msg,))
9996

    
9997
    # if we managed to detach at least one, we update all the disks of
9998
    # the instance to point to the new secondary
9999
    self.lu.LogInfo("Updating instance configuration")
10000
    for dev, _, new_logical_id in iv_names.itervalues():
10001
      dev.logical_id = new_logical_id
10002
      self.cfg.SetDiskID(dev, self.instance.primary_node)
10003

    
10004
    self.cfg.Update(self.instance, feedback_fn)
10005

    
10006
    # and now perform the drbd attach
10007
    self.lu.LogInfo("Attaching primary drbds to new secondary"
10008
                    " (standalone => connected)")
10009
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
10010
                                            self.new_node],
10011
                                           self.node_secondary_ip,
10012
                                           self.instance.disks,
10013
                                           self.instance.name,
10014
                                           False)
10015
    for to_node, to_result in result.items():
10016
      msg = to_result.fail_msg
10017
      if msg:
10018
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
10019
                           to_node, msg,
10020
                           hint=("please do a gnt-instance info to see the"
10021
                                 " status of disks"))
10022
    cstep = 5
10023
    if self.early_release:
10024
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
10025
      cstep += 1
10026
      self._RemoveOldStorage(self.target_node, iv_names)
10027
      # WARNING: we release all node locks here, do not do other RPCs
10028
      # than WaitForSync to the primary node
10029
      _ReleaseLocks(self.lu, locking.LEVEL_NODE,
10030
                    names=[self.instance.primary_node,
10031
                           self.target_node,
10032
                           self.new_node])
10033

    
10034
    # Wait for sync
10035
    # This can fail as the old devices are degraded and _WaitForSync
10036
    # does a combined result over all disks, so we don't check its return value
10037
    self.lu.LogStep(cstep, steps_total, "Sync devices")
10038
    cstep += 1
10039
    _WaitForSync(self.lu, self.instance)
10040

    
10041
    # Check all devices manually
10042
    self._CheckDevices(self.instance.primary_node, iv_names)
10043

    
10044
    # Step: remove old storage
10045
    if not self.early_release:
10046
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
10047
      self._RemoveOldStorage(self.target_node, iv_names)
10048

    
10049

    
10050
class LURepairNodeStorage(NoHooksLU):
10051
  """Repairs the volume group on a node.
10052

10053
  """
10054
  REQ_BGL = False
10055

    
10056
  def CheckArguments(self):
10057
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
10058

    
10059
    storage_type = self.op.storage_type
10060

    
10061
    if (constants.SO_FIX_CONSISTENCY not in
10062
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
10063
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
10064
                                 " repaired" % storage_type,
10065
                                 errors.ECODE_INVAL)
10066

    
10067
  def ExpandNames(self):
10068
    self.needed_locks = {
10069
      locking.LEVEL_NODE: [self.op.node_name],
10070
      }
10071

    
10072
  def _CheckFaultyDisks(self, instance, node_name):
10073
    """Ensure faulty disks abort the opcode or at least warn."""
10074
    try:
10075
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
10076
                                  node_name, True):
10077
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
10078
                                   " node '%s'" % (instance.name, node_name),
10079
                                   errors.ECODE_STATE)
10080
    except errors.OpPrereqError, err:
10081
      if self.op.ignore_consistency:
10082
        self.proc.LogWarning(str(err.args[0]))
10083
      else:
10084
        raise
10085

    
10086
  def CheckPrereq(self):
10087
    """Check prerequisites.
10088

10089
    """
10090
    # Check whether any instance on this node has faulty disks
10091
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
10092
      if not inst.admin_up:
10093
        continue
10094
      check_nodes = set(inst.all_nodes)
10095
      check_nodes.discard(self.op.node_name)
10096
      for inst_node_name in check_nodes:
10097
        self._CheckFaultyDisks(inst, inst_node_name)
10098

    
10099
  def Exec(self, feedback_fn):
10100
    feedback_fn("Repairing storage unit '%s' on %s ..." %
10101
                (self.op.name, self.op.node_name))
10102

    
10103
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
10104
    result = self.rpc.call_storage_execute(self.op.node_name,
10105
                                           self.op.storage_type, st_args,
10106
                                           self.op.name,
10107
                                           constants.SO_FIX_CONSISTENCY)
10108
    result.Raise("Failed to repair storage unit '%s' on %s" %
10109
                 (self.op.name, self.op.node_name))
10110

    
10111

    
10112
class LUNodeEvacuate(NoHooksLU):
10113
  """Evacuates instances off a list of nodes.
10114

10115
  """
10116
  REQ_BGL = False
10117

    
10118
  def CheckArguments(self):
10119
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
10120

    
10121
  def ExpandNames(self):
10122
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
10123

    
10124
    if self.op.remote_node is not None:
10125
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
10126
      assert self.op.remote_node
10127

    
10128
      if self.op.remote_node == self.op.node_name:
10129
        raise errors.OpPrereqError("Can not use evacuated node as a new"
10130
                                   " secondary node", errors.ECODE_INVAL)
10131

    
10132
      if self.op.mode != constants.IALLOCATOR_NEVAC_SEC:
10133
        raise errors.OpPrereqError("Without the use of an iallocator only"
10134
                                   " secondary instances can be evacuated",
10135
                                   errors.ECODE_INVAL)
10136

    
10137
    # Declare locks
10138
    self.share_locks = _ShareAll()
10139
    self.needed_locks = {
10140
      locking.LEVEL_INSTANCE: [],
10141
      locking.LEVEL_NODEGROUP: [],
10142
      locking.LEVEL_NODE: [],
10143
      }
10144

    
10145
    if self.op.remote_node is None:
10146
      # Iallocator will choose any node(s) in the same group
10147
      group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
10148
    else:
10149
      group_nodes = frozenset([self.op.remote_node])
10150

    
10151
    # Determine nodes to be locked
10152
    self.lock_nodes = set([self.op.node_name]) | group_nodes
10153

    
10154
  def _DetermineInstances(self):
10155
    """Builds list of instances to operate on.
10156

10157
    """
10158
    assert self.op.mode in constants.IALLOCATOR_NEVAC_MODES
10159

    
10160
    if self.op.mode == constants.IALLOCATOR_NEVAC_PRI:
10161
      # Primary instances only
10162
      inst_fn = _GetNodePrimaryInstances
10163
      assert self.op.remote_node is None, \
10164
        "Evacuating primary instances requires iallocator"
10165
    elif self.op.mode == constants.IALLOCATOR_NEVAC_SEC:
10166
      # Secondary instances only
10167
      inst_fn = _GetNodeSecondaryInstances
10168
    else:
10169
      # All instances
10170
      assert self.op.mode == constants.IALLOCATOR_NEVAC_ALL
10171
      inst_fn = _GetNodeInstances
10172

    
10173
    return inst_fn(self.cfg, self.op.node_name)
10174

    
10175
  def DeclareLocks(self, level):
10176
    if level == locking.LEVEL_INSTANCE:
10177
      # Lock instances optimistically, needs verification once node and group
10178
      # locks have been acquired
10179
      self.needed_locks[locking.LEVEL_INSTANCE] = \
10180
        set(i.name for i in self._DetermineInstances())
10181

    
10182
    elif level == locking.LEVEL_NODEGROUP:
10183
      # Lock node groups optimistically, needs verification once nodes have
10184
      # been acquired
10185
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
10186
        self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
10187

    
10188
    elif level == locking.LEVEL_NODE:
10189
      self.needed_locks[locking.LEVEL_NODE] = self.lock_nodes
10190

    
10191
  def CheckPrereq(self):
10192
    # Verify locks
10193
    owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
10194
    owned_nodes = self.owned_locks(locking.LEVEL_NODE)
10195
    owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
10196

    
10197
    assert owned_nodes == self.lock_nodes
10198

    
10199
    wanted_groups = self.cfg.GetNodeGroupsFromNodes(owned_nodes)
10200
    if owned_groups != wanted_groups:
10201
      raise errors.OpExecError("Node groups changed since locks were acquired,"
10202
                               " current groups are '%s', used to be '%s'" %
10203
                               (utils.CommaJoin(wanted_groups),
10204
                                utils.CommaJoin(owned_groups)))
10205

    
10206
    # Determine affected instances
10207
    self.instances = self._DetermineInstances()
10208
    self.instance_names = [i.name for i in self.instances]
10209

    
10210
    if set(self.instance_names) != owned_instances:
10211
      raise errors.OpExecError("Instances on node '%s' changed since locks"
10212
                               " were acquired, current instances are '%s',"
10213
                               " used to be '%s'" %
10214
                               (self.op.node_name,
10215
                                utils.CommaJoin(self.instance_names),
10216
                                utils.CommaJoin(owned_instances)))
10217

    
10218
    if self.instance_names:
10219
      self.LogInfo("Evacuating instances from node '%s': %s",
10220
                   self.op.node_name,
10221
                   utils.CommaJoin(utils.NiceSort(self.instance_names)))
10222
    else:
10223
      self.LogInfo("No instances to evacuate from node '%s'",
10224
                   self.op.node_name)
10225

    
10226
    if self.op.remote_node is not None:
10227
      for i in self.instances:
10228
        if i.primary_node == self.op.remote_node:
10229
          raise errors.OpPrereqError("Node %s is the primary node of"
10230
                                     " instance %s, cannot use it as"
10231
                                     " secondary" %
10232
                                     (self.op.remote_node, i.name),
10233
                                     errors.ECODE_INVAL)
10234

    
10235
  def Exec(self, feedback_fn):
10236
    assert (self.op.iallocator is not None) ^ (self.op.remote_node is not None)
10237

    
10238
    if not self.instance_names:
10239
      # No instances to evacuate
10240
      jobs = []
10241

    
10242
    elif self.op.iallocator is not None:
10243
      # TODO: Implement relocation to other group
10244
      ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_NODE_EVAC,
10245
                       evac_mode=self.op.mode,
10246
                       instances=list(self.instance_names))
10247

    
10248
      ial.Run(self.op.iallocator)
10249

    
10250
      if not ial.success:
10251
        raise errors.OpPrereqError("Can't compute node evacuation using"
10252
                                   " iallocator '%s': %s" %
10253
                                   (self.op.iallocator, ial.info),
10254
                                   errors.ECODE_NORES)
10255

    
10256
      jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, True)
10257

    
10258
    elif self.op.remote_node is not None:
10259
      assert self.op.mode == constants.IALLOCATOR_NEVAC_SEC
10260
      jobs = [
10261
        [opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
10262
                                        remote_node=self.op.remote_node,
10263
                                        disks=[],
10264
                                        mode=constants.REPLACE_DISK_CHG,
10265
                                        early_release=self.op.early_release)]
10266
        for instance_name in self.instance_names
10267
        ]
10268

    
10269
    else:
10270
      raise errors.ProgrammerError("No iallocator or remote node")
10271

    
10272
    return ResultWithJobs(jobs)
10273

    
10274

    
10275
def _SetOpEarlyRelease(early_release, op):
10276
  """Sets C{early_release} flag on opcodes if available.
10277

10278
  """
10279
  try:
10280
    op.early_release = early_release
10281
  except AttributeError:
10282
    assert not isinstance(op, opcodes.OpInstanceReplaceDisks)
10283

    
10284
  return op
10285

    
10286

    
10287
def _NodeEvacDest(use_nodes, group, nodes):
10288
  """Returns group or nodes depending on caller's choice.
10289

10290
  """
10291
  if use_nodes:
10292
    return utils.CommaJoin(nodes)
10293
  else:
10294
    return group
10295

    
10296

    
10297
def _LoadNodeEvacResult(lu, alloc_result, early_release, use_nodes):
10298
  """Unpacks the result of change-group and node-evacuate iallocator requests.
10299

10300
  Iallocator modes L{constants.IALLOCATOR_MODE_NODE_EVAC} and
10301
  L{constants.IALLOCATOR_MODE_CHG_GROUP}.
10302

10303
  @type lu: L{LogicalUnit}
10304
  @param lu: Logical unit instance
10305
  @type alloc_result: tuple/list
10306
  @param alloc_result: Result from iallocator
10307
  @type early_release: bool
10308
  @param early_release: Whether to release locks early if possible
10309
  @type use_nodes: bool
10310
  @param use_nodes: Whether to display node names instead of groups
10311

10312
  """
10313
  (moved, failed, jobs) = alloc_result
10314

    
10315
  if failed:
10316
    lu.LogWarning("Unable to evacuate instances %s",
10317
                  utils.CommaJoin("%s (%s)" % (name, reason)
10318
                                  for (name, reason) in failed))
10319

    
10320
  if moved:
10321
    lu.LogInfo("Instances to be moved: %s",
10322
               utils.CommaJoin("%s (to %s)" %
10323
                               (name, _NodeEvacDest(use_nodes, group, nodes))
10324
                               for (name, group, nodes) in moved))
10325

    
10326
  return [map(compat.partial(_SetOpEarlyRelease, early_release),
10327
              map(opcodes.OpCode.LoadOpCode, ops))
10328
          for ops in jobs]
10329

    
10330

    
10331
class LUInstanceGrowDisk(LogicalUnit):
10332
  """Grow a disk of an instance.
10333

10334
  """
10335
  HPATH = "disk-grow"
10336
  HTYPE = constants.HTYPE_INSTANCE
10337
  REQ_BGL = False
10338

    
10339
  def ExpandNames(self):
10340
    self._ExpandAndLockInstance()
10341
    self.needed_locks[locking.LEVEL_NODE] = []
10342
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10343

    
10344
  def DeclareLocks(self, level):
10345
    if level == locking.LEVEL_NODE:
10346
      self._LockInstancesNodes()
10347

    
10348
  def BuildHooksEnv(self):
10349
    """Build hooks env.
10350

10351
    This runs on the master, the primary and all the secondaries.
10352

10353
    """
10354
    env = {
10355
      "DISK": self.op.disk,
10356
      "AMOUNT": self.op.amount,
10357
      }
10358
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
10359
    return env
10360

    
10361
  def BuildHooksNodes(self):
10362
    """Build hooks nodes.
10363

10364
    """
10365
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
10366
    return (nl, nl)
10367

    
10368
  def CheckPrereq(self):
10369
    """Check prerequisites.
10370

10371
    This checks that the instance is in the cluster.
10372

10373
    """
10374
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
10375
    assert instance is not None, \
10376
      "Cannot retrieve locked instance %s" % self.op.instance_name
10377
    nodenames = list(instance.all_nodes)
10378
    for node in nodenames:
10379
      _CheckNodeOnline(self, node)
10380

    
10381
    self.instance = instance
10382

    
10383
    if instance.disk_template not in constants.DTS_GROWABLE:
10384
      raise errors.OpPrereqError("Instance's disk layout does not support"
10385
                                 " growing", errors.ECODE_INVAL)
10386

    
10387
    self.disk = instance.FindDisk(self.op.disk)
10388

    
10389
    if instance.disk_template not in (constants.DT_FILE,
10390
                                      constants.DT_SHARED_FILE):
10391
      # TODO: check the free disk space for file, when that feature will be
10392
      # supported
10393
      _CheckNodesFreeDiskPerVG(self, nodenames,
10394
                               self.disk.ComputeGrowth(self.op.amount))
10395

    
10396
  def Exec(self, feedback_fn):
10397
    """Execute disk grow.
10398

10399
    """
10400
    instance = self.instance
10401
    disk = self.disk
10402

    
10403
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
10404
    if not disks_ok:
10405
      raise errors.OpExecError("Cannot activate block device to grow")
10406

    
10407
    # First run all grow ops in dry-run mode
10408
    for node in instance.all_nodes:
10409
      self.cfg.SetDiskID(disk, node)
10410
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, True)
10411
      result.Raise("Grow request failed to node %s" % node)
10412

    
10413
    # We know that (as far as we can test) operations across different
10414
    # nodes will succeed, time to run it for real
10415
    for node in instance.all_nodes:
10416
      self.cfg.SetDiskID(disk, node)
10417
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, False)
10418
      result.Raise("Grow request failed to node %s" % node)
10419

    
10420
      # TODO: Rewrite code to work properly
10421
      # DRBD goes into sync mode for a short amount of time after executing the
10422
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
10423
      # calling "resize" in sync mode fails. Sleeping for a short amount of
10424
      # time is a work-around.
10425
      time.sleep(5)
10426

    
10427
    disk.RecordGrow(self.op.amount)
10428
    self.cfg.Update(instance, feedback_fn)
10429
    if self.op.wait_for_sync:
10430
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
10431
      if disk_abort:
10432
        self.proc.LogWarning("Disk sync-ing has not returned a good"
10433
                             " status; please check the instance")
10434
      if not instance.admin_up:
10435
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
10436
    elif not instance.admin_up:
10437
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
10438
                           " not supposed to be running because no wait for"
10439
                           " sync mode was requested")
10440

    
10441

    
10442
class LUInstanceQueryData(NoHooksLU):
10443
  """Query runtime instance data.
10444

10445
  """
10446
  REQ_BGL = False
10447

    
10448
  def ExpandNames(self):
10449
    self.needed_locks = {}
10450

    
10451
    # Use locking if requested or when non-static information is wanted
10452
    if not (self.op.static or self.op.use_locking):
10453
      self.LogWarning("Non-static data requested, locks need to be acquired")
10454
      self.op.use_locking = True
10455

    
10456
    if self.op.instances or not self.op.use_locking:
10457
      # Expand instance names right here
10458
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
10459
    else:
10460
      # Will use acquired locks
10461
      self.wanted_names = None
10462

    
10463
    if self.op.use_locking:
10464
      self.share_locks = _ShareAll()
10465

    
10466
      if self.wanted_names is None:
10467
        self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
10468
      else:
10469
        self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
10470

    
10471
      self.needed_locks[locking.LEVEL_NODE] = []
10472
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10473

    
10474
  def DeclareLocks(self, level):
10475
    if self.op.use_locking and level == locking.LEVEL_NODE:
10476
      self._LockInstancesNodes()
10477

    
10478
  def CheckPrereq(self):
10479
    """Check prerequisites.
10480

10481
    This only checks the optional instance list against the existing names.
10482

10483
    """
10484
    if self.wanted_names is None:
10485
      assert self.op.use_locking, "Locking was not used"
10486
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
10487

    
10488
    self.wanted_instances = \
10489
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
10490

    
10491
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
10492
    """Returns the status of a block device
10493

10494
    """
10495
    if self.op.static or not node:
10496
      return None
10497

    
10498
    self.cfg.SetDiskID(dev, node)
10499

    
10500
    result = self.rpc.call_blockdev_find(node, dev)
10501
    if result.offline:
10502
      return None
10503

    
10504
    result.Raise("Can't compute disk status for %s" % instance_name)
10505

    
10506
    status = result.payload
10507
    if status is None:
10508
      return None
10509

    
10510
    return (status.dev_path, status.major, status.minor,
10511
            status.sync_percent, status.estimated_time,
10512
            status.is_degraded, status.ldisk_status)
10513

    
10514
  def _ComputeDiskStatus(self, instance, snode, dev):
10515
    """Compute block device status.
10516

10517
    """
10518
    if dev.dev_type in constants.LDS_DRBD:
10519
      # we change the snode then (otherwise we use the one passed in)
10520
      if dev.logical_id[0] == instance.primary_node:
10521
        snode = dev.logical_id[1]
10522
      else:
10523
        snode = dev.logical_id[0]
10524

    
10525
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
10526
                                              instance.name, dev)
10527
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
10528

    
10529
    if dev.children:
10530
      dev_children = map(compat.partial(self._ComputeDiskStatus,
10531
                                        instance, snode),
10532
                         dev.children)
10533
    else:
10534
      dev_children = []
10535

    
10536
    return {
10537
      "iv_name": dev.iv_name,
10538
      "dev_type": dev.dev_type,
10539
      "logical_id": dev.logical_id,
10540
      "physical_id": dev.physical_id,
10541
      "pstatus": dev_pstatus,
10542
      "sstatus": dev_sstatus,
10543
      "children": dev_children,
10544
      "mode": dev.mode,
10545
      "size": dev.size,
10546
      }
10547

    
10548
  def Exec(self, feedback_fn):
10549
    """Gather and return data"""
10550
    result = {}
10551

    
10552
    cluster = self.cfg.GetClusterInfo()
10553

    
10554
    pri_nodes = self.cfg.GetMultiNodeInfo(i.primary_node
10555
                                          for i in self.wanted_instances)
10556
    for instance, (_, pnode) in zip(self.wanted_instances, pri_nodes):
10557
      if self.op.static or pnode.offline:
10558
        remote_state = None
10559
        if pnode.offline:
10560
          self.LogWarning("Primary node %s is marked offline, returning static"
10561
                          " information only for instance %s" %
10562
                          (pnode.name, instance.name))
10563
      else:
10564
        remote_info = self.rpc.call_instance_info(instance.primary_node,
10565
                                                  instance.name,
10566
                                                  instance.hypervisor)
10567
        remote_info.Raise("Error checking node %s" % instance.primary_node)
10568
        remote_info = remote_info.payload
10569
        if remote_info and "state" in remote_info:
10570
          remote_state = "up"
10571
        else:
10572
          remote_state = "down"
10573

    
10574
      if instance.admin_up:
10575
        config_state = "up"
10576
      else:
10577
        config_state = "down"
10578

    
10579
      disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
10580
                  instance.disks)
10581

    
10582
      result[instance.name] = {
10583
        "name": instance.name,
10584
        "config_state": config_state,
10585
        "run_state": remote_state,
10586
        "pnode": instance.primary_node,
10587
        "snodes": instance.secondary_nodes,
10588
        "os": instance.os,
10589
        # this happens to be the same format used for hooks
10590
        "nics": _NICListToTuple(self, instance.nics),
10591
        "disk_template": instance.disk_template,
10592
        "disks": disks,
10593
        "hypervisor": instance.hypervisor,
10594
        "network_port": instance.network_port,
10595
        "hv_instance": instance.hvparams,
10596
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
10597
        "be_instance": instance.beparams,
10598
        "be_actual": cluster.FillBE(instance),
10599
        "os_instance": instance.osparams,
10600
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
10601
        "serial_no": instance.serial_no,
10602
        "mtime": instance.mtime,
10603
        "ctime": instance.ctime,
10604
        "uuid": instance.uuid,
10605
        }
10606

    
10607
    return result
10608

    
10609

    
10610
class LUInstanceSetParams(LogicalUnit):
10611
  """Modifies an instances's parameters.
10612

10613
  """
10614
  HPATH = "instance-modify"
10615
  HTYPE = constants.HTYPE_INSTANCE
10616
  REQ_BGL = False
10617

    
10618
  def CheckArguments(self):
10619
    if not (self.op.nics or self.op.disks or self.op.disk_template or
10620
            self.op.hvparams or self.op.beparams or self.op.os_name):
10621
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
10622

    
10623
    if self.op.hvparams:
10624
      _CheckGlobalHvParams(self.op.hvparams)
10625

    
10626
    # Disk validation
10627
    disk_addremove = 0
10628
    for disk_op, disk_dict in self.op.disks:
10629
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
10630
      if disk_op == constants.DDM_REMOVE:
10631
        disk_addremove += 1
10632
        continue
10633
      elif disk_op == constants.DDM_ADD:
10634
        disk_addremove += 1
10635
      else:
10636
        if not isinstance(disk_op, int):
10637
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
10638
        if not isinstance(disk_dict, dict):
10639
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
10640
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
10641

    
10642
      if disk_op == constants.DDM_ADD:
10643
        mode = disk_dict.setdefault(constants.IDISK_MODE, constants.DISK_RDWR)
10644
        if mode not in constants.DISK_ACCESS_SET:
10645
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
10646
                                     errors.ECODE_INVAL)
10647
        size = disk_dict.get(constants.IDISK_SIZE, None)
10648
        if size is None:
10649
          raise errors.OpPrereqError("Required disk parameter size missing",
10650
                                     errors.ECODE_INVAL)
10651
        try:
10652
          size = int(size)
10653
        except (TypeError, ValueError), err:
10654
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
10655
                                     str(err), errors.ECODE_INVAL)
10656
        disk_dict[constants.IDISK_SIZE] = size
10657
      else:
10658
        # modification of disk
10659
        if constants.IDISK_SIZE in disk_dict:
10660
          raise errors.OpPrereqError("Disk size change not possible, use"
10661
                                     " grow-disk", errors.ECODE_INVAL)
10662

    
10663
    if disk_addremove > 1:
10664
      raise errors.OpPrereqError("Only one disk add or remove operation"
10665
                                 " supported at a time", errors.ECODE_INVAL)
10666

    
10667
    if self.op.disks and self.op.disk_template is not None:
10668
      raise errors.OpPrereqError("Disk template conversion and other disk"
10669
                                 " changes not supported at the same time",
10670
                                 errors.ECODE_INVAL)
10671

    
10672
    if (self.op.disk_template and
10673
        self.op.disk_template in constants.DTS_INT_MIRROR and
10674
        self.op.remote_node is None):
10675
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
10676
                                 " one requires specifying a secondary node",
10677
                                 errors.ECODE_INVAL)
10678

    
10679
    # NIC validation
10680
    nic_addremove = 0
10681
    for nic_op, nic_dict in self.op.nics:
10682
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
10683
      if nic_op == constants.DDM_REMOVE:
10684
        nic_addremove += 1
10685
        continue
10686
      elif nic_op == constants.DDM_ADD:
10687
        nic_addremove += 1
10688
      else:
10689
        if not isinstance(nic_op, int):
10690
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
10691
        if not isinstance(nic_dict, dict):
10692
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
10693
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
10694

    
10695
      # nic_dict should be a dict
10696
      nic_ip = nic_dict.get(constants.INIC_IP, None)
10697
      if nic_ip is not None:
10698
        if nic_ip.lower() == constants.VALUE_NONE:
10699
          nic_dict[constants.INIC_IP] = None
10700
        else:
10701
          if not netutils.IPAddress.IsValid(nic_ip):
10702
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
10703
                                       errors.ECODE_INVAL)
10704

    
10705
      nic_bridge = nic_dict.get("bridge", None)
10706
      nic_link = nic_dict.get(constants.INIC_LINK, None)
10707
      if nic_bridge and nic_link:
10708
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
10709
                                   " at the same time", errors.ECODE_INVAL)
10710
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
10711
        nic_dict["bridge"] = None
10712
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
10713
        nic_dict[constants.INIC_LINK] = None
10714

    
10715
      if nic_op == constants.DDM_ADD:
10716
        nic_mac = nic_dict.get(constants.INIC_MAC, None)
10717
        if nic_mac is None:
10718
          nic_dict[constants.INIC_MAC] = constants.VALUE_AUTO
10719

    
10720
      if constants.INIC_MAC in nic_dict:
10721
        nic_mac = nic_dict[constants.INIC_MAC]
10722
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
10723
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
10724

    
10725
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
10726
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
10727
                                     " modifying an existing nic",
10728
                                     errors.ECODE_INVAL)
10729

    
10730
    if nic_addremove > 1:
10731
      raise errors.OpPrereqError("Only one NIC add or remove operation"
10732
                                 " supported at a time", errors.ECODE_INVAL)
10733

    
10734
  def ExpandNames(self):
10735
    self._ExpandAndLockInstance()
10736
    self.needed_locks[locking.LEVEL_NODE] = []
10737
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10738

    
10739
  def DeclareLocks(self, level):
10740
    if level == locking.LEVEL_NODE:
10741
      self._LockInstancesNodes()
10742
      if self.op.disk_template and self.op.remote_node:
10743
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
10744
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
10745

    
10746
  def BuildHooksEnv(self):
10747
    """Build hooks env.
10748

10749
    This runs on the master, primary and secondaries.
10750

10751
    """
10752
    args = dict()
10753
    if constants.BE_MEMORY in self.be_new:
10754
      args["memory"] = self.be_new[constants.BE_MEMORY]
10755
    if constants.BE_VCPUS in self.be_new:
10756
      args["vcpus"] = self.be_new[constants.BE_VCPUS]
10757
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
10758
    # information at all.
10759
    if self.op.nics:
10760
      args["nics"] = []
10761
      nic_override = dict(self.op.nics)
10762
      for idx, nic in enumerate(self.instance.nics):
10763
        if idx in nic_override:
10764
          this_nic_override = nic_override[idx]
10765
        else:
10766
          this_nic_override = {}
10767
        if constants.INIC_IP in this_nic_override:
10768
          ip = this_nic_override[constants.INIC_IP]
10769
        else:
10770
          ip = nic.ip
10771
        if constants.INIC_MAC in this_nic_override:
10772
          mac = this_nic_override[constants.INIC_MAC]
10773
        else:
10774
          mac = nic.mac
10775
        if idx in self.nic_pnew:
10776
          nicparams = self.nic_pnew[idx]
10777
        else:
10778
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
10779
        mode = nicparams[constants.NIC_MODE]
10780
        link = nicparams[constants.NIC_LINK]
10781
        args["nics"].append((ip, mac, mode, link))
10782
      if constants.DDM_ADD in nic_override:
10783
        ip = nic_override[constants.DDM_ADD].get(constants.INIC_IP, None)
10784
        mac = nic_override[constants.DDM_ADD][constants.INIC_MAC]
10785
        nicparams = self.nic_pnew[constants.DDM_ADD]
10786
        mode = nicparams[constants.NIC_MODE]
10787
        link = nicparams[constants.NIC_LINK]
10788
        args["nics"].append((ip, mac, mode, link))
10789
      elif constants.DDM_REMOVE in nic_override:
10790
        del args["nics"][-1]
10791

    
10792
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
10793
    if self.op.disk_template:
10794
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
10795

    
10796
    return env
10797

    
10798
  def BuildHooksNodes(self):
10799
    """Build hooks nodes.
10800

10801
    """
10802
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
10803
    return (nl, nl)
10804

    
10805
  def CheckPrereq(self):
10806
    """Check prerequisites.
10807

10808
    This only checks the instance list against the existing names.
10809

10810
    """
10811
    # checking the new params on the primary/secondary nodes
10812

    
10813
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
10814
    cluster = self.cluster = self.cfg.GetClusterInfo()
10815
    assert self.instance is not None, \
10816
      "Cannot retrieve locked instance %s" % self.op.instance_name
10817
    pnode = instance.primary_node
10818
    nodelist = list(instance.all_nodes)
10819

    
10820
    # OS change
10821
    if self.op.os_name and not self.op.force:
10822
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
10823
                      self.op.force_variant)
10824
      instance_os = self.op.os_name
10825
    else:
10826
      instance_os = instance.os
10827

    
10828
    if self.op.disk_template:
10829
      if instance.disk_template == self.op.disk_template:
10830
        raise errors.OpPrereqError("Instance already has disk template %s" %
10831
                                   instance.disk_template, errors.ECODE_INVAL)
10832

    
10833
      if (instance.disk_template,
10834
          self.op.disk_template) not in self._DISK_CONVERSIONS:
10835
        raise errors.OpPrereqError("Unsupported disk template conversion from"
10836
                                   " %s to %s" % (instance.disk_template,
10837
                                                  self.op.disk_template),
10838
                                   errors.ECODE_INVAL)
10839
      _CheckInstanceDown(self, instance, "cannot change disk template")
10840
      if self.op.disk_template in constants.DTS_INT_MIRROR:
10841
        if self.op.remote_node == pnode:
10842
          raise errors.OpPrereqError("Given new secondary node %s is the same"
10843
                                     " as the primary node of the instance" %
10844
                                     self.op.remote_node, errors.ECODE_STATE)
10845
        _CheckNodeOnline(self, self.op.remote_node)
10846
        _CheckNodeNotDrained(self, self.op.remote_node)
10847
        # FIXME: here we assume that the old instance type is DT_PLAIN
10848
        assert instance.disk_template == constants.DT_PLAIN
10849
        disks = [{constants.IDISK_SIZE: d.size,
10850
                  constants.IDISK_VG: d.logical_id[0]}
10851
                 for d in instance.disks]
10852
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
10853
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
10854

    
10855
    # hvparams processing
10856
    if self.op.hvparams:
10857
      hv_type = instance.hypervisor
10858
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
10859
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
10860
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
10861

    
10862
      # local check
10863
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
10864
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
10865
      self.hv_proposed = self.hv_new = hv_new # the new actual values
10866
      self.hv_inst = i_hvdict # the new dict (without defaults)
10867
    else:
10868
      self.hv_proposed = cluster.SimpleFillHV(instance.hypervisor, instance.os,
10869
                                              instance.hvparams)
10870
      self.hv_new = self.hv_inst = {}
10871

    
10872
    # beparams processing
10873
    if self.op.beparams:
10874
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
10875
                                   use_none=True)
10876
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
10877
      be_new = cluster.SimpleFillBE(i_bedict)
10878
      self.be_proposed = self.be_new = be_new # the new actual values
10879
      self.be_inst = i_bedict # the new dict (without defaults)
10880
    else:
10881
      self.be_new = self.be_inst = {}
10882
      self.be_proposed = cluster.SimpleFillBE(instance.beparams)
10883
    be_old = cluster.FillBE(instance)
10884

    
10885
    # CPU param validation -- checking every time a paramtere is
10886
    # changed to cover all cases where either CPU mask or vcpus have
10887
    # changed
10888
    if (constants.BE_VCPUS in self.be_proposed and
10889
        constants.HV_CPU_MASK in self.hv_proposed):
10890
      cpu_list = \
10891
        utils.ParseMultiCpuMask(self.hv_proposed[constants.HV_CPU_MASK])
10892
      # Verify mask is consistent with number of vCPUs. Can skip this
10893
      # test if only 1 entry in the CPU mask, which means same mask
10894
      # is applied to all vCPUs.
10895
      if (len(cpu_list) > 1 and
10896
          len(cpu_list) != self.be_proposed[constants.BE_VCPUS]):
10897
        raise errors.OpPrereqError("Number of vCPUs [%d] does not match the"
10898
                                   " CPU mask [%s]" %
10899
                                   (self.be_proposed[constants.BE_VCPUS],
10900
                                    self.hv_proposed[constants.HV_CPU_MASK]),
10901
                                   errors.ECODE_INVAL)
10902

    
10903
      # Only perform this test if a new CPU mask is given
10904
      if constants.HV_CPU_MASK in self.hv_new:
10905
        # Calculate the largest CPU number requested
10906
        max_requested_cpu = max(map(max, cpu_list))
10907
        # Check that all of the instance's nodes have enough physical CPUs to
10908
        # satisfy the requested CPU mask
10909
        _CheckNodesPhysicalCPUs(self, instance.all_nodes,
10910
                                max_requested_cpu + 1, instance.hypervisor)
10911

    
10912
    # osparams processing
10913
    if self.op.osparams:
10914
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
10915
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
10916
      self.os_inst = i_osdict # the new dict (without defaults)
10917
    else:
10918
      self.os_inst = {}
10919

    
10920
    self.warn = []
10921

    
10922
    if (constants.BE_MEMORY in self.op.beparams and not self.op.force and
10923
        be_new[constants.BE_MEMORY] > be_old[constants.BE_MEMORY]):
10924
      mem_check_list = [pnode]
10925
      if be_new[constants.BE_AUTO_BALANCE]:
10926
        # either we changed auto_balance to yes or it was from before
10927
        mem_check_list.extend(instance.secondary_nodes)
10928
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
10929
                                                  instance.hypervisor)
10930
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
10931
                                         instance.hypervisor)
10932
      pninfo = nodeinfo[pnode]
10933
      msg = pninfo.fail_msg
10934
      if msg:
10935
        # Assume the primary node is unreachable and go ahead
10936
        self.warn.append("Can't get info from primary node %s: %s" %
10937
                         (pnode, msg))
10938
      elif not isinstance(pninfo.payload.get("memory_free", None), int):
10939
        self.warn.append("Node data from primary node %s doesn't contain"
10940
                         " free memory information" % pnode)
10941
      elif instance_info.fail_msg:
10942
        self.warn.append("Can't get instance runtime information: %s" %
10943
                        instance_info.fail_msg)
10944
      else:
10945
        if instance_info.payload:
10946
          current_mem = int(instance_info.payload["memory"])
10947
        else:
10948
          # Assume instance not running
10949
          # (there is a slight race condition here, but it's not very probable,
10950
          # and we have no other way to check)
10951
          current_mem = 0
10952
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
10953
                    pninfo.payload["memory_free"])
10954
        if miss_mem > 0:
10955
          raise errors.OpPrereqError("This change will prevent the instance"
10956
                                     " from starting, due to %d MB of memory"
10957
                                     " missing on its primary node" % miss_mem,
10958
                                     errors.ECODE_NORES)
10959

    
10960
      if be_new[constants.BE_AUTO_BALANCE]:
10961
        for node, nres in nodeinfo.items():
10962
          if node not in instance.secondary_nodes:
10963
            continue
10964
          nres.Raise("Can't get info from secondary node %s" % node,
10965
                     prereq=True, ecode=errors.ECODE_STATE)
10966
          if not isinstance(nres.payload.get("memory_free", None), int):
10967
            raise errors.OpPrereqError("Secondary node %s didn't return free"
10968
                                       " memory information" % node,
10969
                                       errors.ECODE_STATE)
10970
          elif be_new[constants.BE_MEMORY] > nres.payload["memory_free"]:
10971
            raise errors.OpPrereqError("This change will prevent the instance"
10972
                                       " from failover to its secondary node"
10973
                                       " %s, due to not enough memory" % node,
10974
                                       errors.ECODE_STATE)
10975

    
10976
    # NIC processing
10977
    self.nic_pnew = {}
10978
    self.nic_pinst = {}
10979
    for nic_op, nic_dict in self.op.nics:
10980
      if nic_op == constants.DDM_REMOVE:
10981
        if not instance.nics:
10982
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
10983
                                     errors.ECODE_INVAL)
10984
        continue
10985
      if nic_op != constants.DDM_ADD:
10986
        # an existing nic
10987
        if not instance.nics:
10988
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
10989
                                     " no NICs" % nic_op,
10990
                                     errors.ECODE_INVAL)
10991
        if nic_op < 0 or nic_op >= len(instance.nics):
10992
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
10993
                                     " are 0 to %d" %
10994
                                     (nic_op, len(instance.nics) - 1),
10995
                                     errors.ECODE_INVAL)
10996
        old_nic_params = instance.nics[nic_op].nicparams
10997
        old_nic_ip = instance.nics[nic_op].ip
10998
      else:
10999
        old_nic_params = {}
11000
        old_nic_ip = None
11001

    
11002
      update_params_dict = dict([(key, nic_dict[key])
11003
                                 for key in constants.NICS_PARAMETERS
11004
                                 if key in nic_dict])
11005

    
11006
      if "bridge" in nic_dict:
11007
        update_params_dict[constants.NIC_LINK] = nic_dict["bridge"]
11008

    
11009
      new_nic_params = _GetUpdatedParams(old_nic_params,
11010
                                         update_params_dict)
11011
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
11012
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
11013
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
11014
      self.nic_pinst[nic_op] = new_nic_params
11015
      self.nic_pnew[nic_op] = new_filled_nic_params
11016
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
11017

    
11018
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
11019
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
11020
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
11021
        if msg:
11022
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
11023
          if self.op.force:
11024
            self.warn.append(msg)
11025
          else:
11026
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
11027
      if new_nic_mode == constants.NIC_MODE_ROUTED:
11028
        if constants.INIC_IP in nic_dict:
11029
          nic_ip = nic_dict[constants.INIC_IP]
11030
        else:
11031
          nic_ip = old_nic_ip
11032
        if nic_ip is None:
11033
          raise errors.OpPrereqError("Cannot set the nic ip to None"
11034
                                     " on a routed nic", errors.ECODE_INVAL)
11035
      if constants.INIC_MAC in nic_dict:
11036
        nic_mac = nic_dict[constants.INIC_MAC]
11037
        if nic_mac is None:
11038
          raise errors.OpPrereqError("Cannot set the nic mac to None",
11039
                                     errors.ECODE_INVAL)
11040
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
11041
          # otherwise generate the mac
11042
          nic_dict[constants.INIC_MAC] = \
11043
            self.cfg.GenerateMAC(self.proc.GetECId())
11044
        else:
11045
          # or validate/reserve the current one
11046
          try:
11047
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
11048
          except errors.ReservationError:
11049
            raise errors.OpPrereqError("MAC address %s already in use"
11050
                                       " in cluster" % nic_mac,
11051
                                       errors.ECODE_NOTUNIQUE)
11052

    
11053
    # DISK processing
11054
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
11055
      raise errors.OpPrereqError("Disk operations not supported for"
11056
                                 " diskless instances",
11057
                                 errors.ECODE_INVAL)
11058
    for disk_op, _ in self.op.disks:
11059
      if disk_op == constants.DDM_REMOVE:
11060
        if len(instance.disks) == 1:
11061
          raise errors.OpPrereqError("Cannot remove the last disk of"
11062
                                     " an instance", errors.ECODE_INVAL)
11063
        _CheckInstanceDown(self, instance, "cannot remove disks")
11064

    
11065
      if (disk_op == constants.DDM_ADD and
11066
          len(instance.disks) >= constants.MAX_DISKS):
11067
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
11068
                                   " add more" % constants.MAX_DISKS,
11069
                                   errors.ECODE_STATE)
11070
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
11071
        # an existing disk
11072
        if disk_op < 0 or disk_op >= len(instance.disks):
11073
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
11074
                                     " are 0 to %d" %
11075
                                     (disk_op, len(instance.disks)),
11076
                                     errors.ECODE_INVAL)
11077

    
11078
    return
11079

    
11080
  def _ConvertPlainToDrbd(self, feedback_fn):
11081
    """Converts an instance from plain to drbd.
11082

11083
    """
11084
    feedback_fn("Converting template to drbd")
11085
    instance = self.instance
11086
    pnode = instance.primary_node
11087
    snode = self.op.remote_node
11088

    
11089
    # create a fake disk info for _GenerateDiskTemplate
11090
    disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
11091
                  constants.IDISK_VG: d.logical_id[0]}
11092
                 for d in instance.disks]
11093
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
11094
                                      instance.name, pnode, [snode],
11095
                                      disk_info, None, None, 0, feedback_fn)
11096
    info = _GetInstanceInfoText(instance)
11097
    feedback_fn("Creating aditional volumes...")
11098
    # first, create the missing data and meta devices
11099
    for disk in new_disks:
11100
      # unfortunately this is... not too nice
11101
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
11102
                            info, True)
11103
      for child in disk.children:
11104
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
11105
    # at this stage, all new LVs have been created, we can rename the
11106
    # old ones
11107
    feedback_fn("Renaming original volumes...")
11108
    rename_list = [(o, n.children[0].logical_id)
11109
                   for (o, n) in zip(instance.disks, new_disks)]
11110
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
11111
    result.Raise("Failed to rename original LVs")
11112

    
11113
    feedback_fn("Initializing DRBD devices...")
11114
    # all child devices are in place, we can now create the DRBD devices
11115
    for disk in new_disks:
11116
      for node in [pnode, snode]:
11117
        f_create = node == pnode
11118
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
11119

    
11120
    # at this point, the instance has been modified
11121
    instance.disk_template = constants.DT_DRBD8
11122
    instance.disks = new_disks
11123
    self.cfg.Update(instance, feedback_fn)
11124

    
11125
    # disks are created, waiting for sync
11126
    disk_abort = not _WaitForSync(self, instance,
11127
                                  oneshot=not self.op.wait_for_sync)
11128
    if disk_abort:
11129
      raise errors.OpExecError("There are some degraded disks for"
11130
                               " this instance, please cleanup manually")
11131

    
11132
  def _ConvertDrbdToPlain(self, feedback_fn):
11133
    """Converts an instance from drbd to plain.
11134

11135
    """
11136
    instance = self.instance
11137
    assert len(instance.secondary_nodes) == 1
11138
    pnode = instance.primary_node
11139
    snode = instance.secondary_nodes[0]
11140
    feedback_fn("Converting template to plain")
11141

    
11142
    old_disks = instance.disks
11143
    new_disks = [d.children[0] for d in old_disks]
11144

    
11145
    # copy over size and mode
11146
    for parent, child in zip(old_disks, new_disks):
11147
      child.size = parent.size
11148
      child.mode = parent.mode
11149

    
11150
    # update instance structure
11151
    instance.disks = new_disks
11152
    instance.disk_template = constants.DT_PLAIN
11153
    self.cfg.Update(instance, feedback_fn)
11154

    
11155
    feedback_fn("Removing volumes on the secondary node...")
11156
    for disk in old_disks:
11157
      self.cfg.SetDiskID(disk, snode)
11158
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
11159
      if msg:
11160
        self.LogWarning("Could not remove block device %s on node %s,"
11161
                        " continuing anyway: %s", disk.iv_name, snode, msg)
11162

    
11163
    feedback_fn("Removing unneeded volumes on the primary node...")
11164
    for idx, disk in enumerate(old_disks):
11165
      meta = disk.children[1]
11166
      self.cfg.SetDiskID(meta, pnode)
11167
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
11168
      if msg:
11169
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
11170
                        " continuing anyway: %s", idx, pnode, msg)
11171

    
11172
  def Exec(self, feedback_fn):
11173
    """Modifies an instance.
11174

11175
    All parameters take effect only at the next restart of the instance.
11176

11177
    """
11178
    # Process here the warnings from CheckPrereq, as we don't have a
11179
    # feedback_fn there.
11180
    for warn in self.warn:
11181
      feedback_fn("WARNING: %s" % warn)
11182

    
11183
    result = []
11184
    instance = self.instance
11185
    # disk changes
11186
    for disk_op, disk_dict in self.op.disks:
11187
      if disk_op == constants.DDM_REMOVE:
11188
        # remove the last disk
11189
        device = instance.disks.pop()
11190
        device_idx = len(instance.disks)
11191
        for node, disk in device.ComputeNodeTree(instance.primary_node):
11192
          self.cfg.SetDiskID(disk, node)
11193
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
11194
          if msg:
11195
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
11196
                            " continuing anyway", device_idx, node, msg)
11197
        result.append(("disk/%d" % device_idx, "remove"))
11198
      elif disk_op == constants.DDM_ADD:
11199
        # add a new disk
11200
        if instance.disk_template in (constants.DT_FILE,
11201
                                        constants.DT_SHARED_FILE):
11202
          file_driver, file_path = instance.disks[0].logical_id
11203
          file_path = os.path.dirname(file_path)
11204
        else:
11205
          file_driver = file_path = None
11206
        disk_idx_base = len(instance.disks)
11207
        new_disk = _GenerateDiskTemplate(self,
11208
                                         instance.disk_template,
11209
                                         instance.name, instance.primary_node,
11210
                                         instance.secondary_nodes,
11211
                                         [disk_dict],
11212
                                         file_path,
11213
                                         file_driver,
11214
                                         disk_idx_base, feedback_fn)[0]
11215
        instance.disks.append(new_disk)
11216
        info = _GetInstanceInfoText(instance)
11217

    
11218
        logging.info("Creating volume %s for instance %s",
11219
                     new_disk.iv_name, instance.name)
11220
        # Note: this needs to be kept in sync with _CreateDisks
11221
        #HARDCODE
11222
        for node in instance.all_nodes:
11223
          f_create = node == instance.primary_node
11224
          try:
11225
            _CreateBlockDev(self, node, instance, new_disk,
11226
                            f_create, info, f_create)
11227
          except errors.OpExecError, err:
11228
            self.LogWarning("Failed to create volume %s (%s) on"
11229
                            " node %s: %s",
11230
                            new_disk.iv_name, new_disk, node, err)
11231
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
11232
                       (new_disk.size, new_disk.mode)))
11233
      else:
11234
        # change a given disk
11235
        instance.disks[disk_op].mode = disk_dict[constants.IDISK_MODE]
11236
        result.append(("disk.mode/%d" % disk_op,
11237
                       disk_dict[constants.IDISK_MODE]))
11238

    
11239
    if self.op.disk_template:
11240
      r_shut = _ShutdownInstanceDisks(self, instance)
11241
      if not r_shut:
11242
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
11243
                                 " proceed with disk template conversion")
11244
      mode = (instance.disk_template, self.op.disk_template)
11245
      try:
11246
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
11247
      except:
11248
        self.cfg.ReleaseDRBDMinors(instance.name)
11249
        raise
11250
      result.append(("disk_template", self.op.disk_template))
11251

    
11252
    # NIC changes
11253
    for nic_op, nic_dict in self.op.nics:
11254
      if nic_op == constants.DDM_REMOVE:
11255
        # remove the last nic
11256
        del instance.nics[-1]
11257
        result.append(("nic.%d" % len(instance.nics), "remove"))
11258
      elif nic_op == constants.DDM_ADD:
11259
        # mac and bridge should be set, by now
11260
        mac = nic_dict[constants.INIC_MAC]
11261
        ip = nic_dict.get(constants.INIC_IP, None)
11262
        nicparams = self.nic_pinst[constants.DDM_ADD]
11263
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
11264
        instance.nics.append(new_nic)
11265
        result.append(("nic.%d" % (len(instance.nics) - 1),
11266
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
11267
                       (new_nic.mac, new_nic.ip,
11268
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
11269
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
11270
                       )))
11271
      else:
11272
        for key in (constants.INIC_MAC, constants.INIC_IP):
11273
          if key in nic_dict:
11274
            setattr(instance.nics[nic_op], key, nic_dict[key])
11275
        if nic_op in self.nic_pinst:
11276
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
11277
        for key, val in nic_dict.iteritems():
11278
          result.append(("nic.%s/%d" % (key, nic_op), val))
11279

    
11280
    # hvparams changes
11281
    if self.op.hvparams:
11282
      instance.hvparams = self.hv_inst
11283
      for key, val in self.op.hvparams.iteritems():
11284
        result.append(("hv/%s" % key, val))
11285

    
11286
    # beparams changes
11287
    if self.op.beparams:
11288
      instance.beparams = self.be_inst
11289
      for key, val in self.op.beparams.iteritems():
11290
        result.append(("be/%s" % key, val))
11291

    
11292
    # OS change
11293
    if self.op.os_name:
11294
      instance.os = self.op.os_name
11295

    
11296
    # osparams changes
11297
    if self.op.osparams:
11298
      instance.osparams = self.os_inst
11299
      for key, val in self.op.osparams.iteritems():
11300
        result.append(("os/%s" % key, val))
11301

    
11302
    self.cfg.Update(instance, feedback_fn)
11303

    
11304
    return result
11305

    
11306
  _DISK_CONVERSIONS = {
11307
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
11308
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
11309
    }
11310

    
11311

    
11312
class LUInstanceChangeGroup(LogicalUnit):
11313
  HPATH = "instance-change-group"
11314
  HTYPE = constants.HTYPE_INSTANCE
11315
  REQ_BGL = False
11316

    
11317
  def ExpandNames(self):
11318
    self.share_locks = _ShareAll()
11319
    self.needed_locks = {
11320
      locking.LEVEL_NODEGROUP: [],
11321
      locking.LEVEL_NODE: [],
11322
      }
11323

    
11324
    self._ExpandAndLockInstance()
11325

    
11326
    if self.op.target_groups:
11327
      self.req_target_uuids = map(self.cfg.LookupNodeGroup,
11328
                                  self.op.target_groups)
11329
    else:
11330
      self.req_target_uuids = None
11331

    
11332
    self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
11333

    
11334
  def DeclareLocks(self, level):
11335
    if level == locking.LEVEL_NODEGROUP:
11336
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
11337

    
11338
      if self.req_target_uuids:
11339
        lock_groups = set(self.req_target_uuids)
11340

    
11341
        # Lock all groups used by instance optimistically; this requires going
11342
        # via the node before it's locked, requiring verification later on
11343
        instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
11344
        lock_groups.update(instance_groups)
11345
      else:
11346
        # No target groups, need to lock all of them
11347
        lock_groups = locking.ALL_SET
11348

    
11349
      self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
11350

    
11351
    elif level == locking.LEVEL_NODE:
11352
      if self.req_target_uuids:
11353
        # Lock all nodes used by instances
11354
        self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
11355
        self._LockInstancesNodes()
11356

    
11357
        # Lock all nodes in all potential target groups
11358
        lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
11359
                       self.cfg.GetInstanceNodeGroups(self.op.instance_name))
11360
        member_nodes = [node_name
11361
                        for group in lock_groups
11362
                        for node_name in self.cfg.GetNodeGroup(group).members]
11363
        self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
11364
      else:
11365
        # Lock all nodes as all groups are potential targets
11366
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11367

    
11368
  def CheckPrereq(self):
11369
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
11370
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
11371
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
11372

    
11373
    assert (self.req_target_uuids is None or
11374
            owned_groups.issuperset(self.req_target_uuids))
11375
    assert owned_instances == set([self.op.instance_name])
11376

    
11377
    # Get instance information
11378
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
11379

    
11380
    # Check if node groups for locked instance are still correct
11381
    assert owned_nodes.issuperset(self.instance.all_nodes), \
11382
      ("Instance %s's nodes changed while we kept the lock" %
11383
       self.op.instance_name)
11384

    
11385
    inst_groups = _CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
11386
                                           owned_groups)
11387

    
11388
    if self.req_target_uuids:
11389
      # User requested specific target groups
11390
      self.target_uuids = self.req_target_uuids
11391
    else:
11392
      # All groups except those used by the instance are potential targets
11393
      self.target_uuids = owned_groups - inst_groups
11394

    
11395
    conflicting_groups = self.target_uuids & inst_groups
11396
    if conflicting_groups:
11397
      raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
11398
                                 " used by the instance '%s'" %
11399
                                 (utils.CommaJoin(conflicting_groups),
11400
                                  self.op.instance_name),
11401
                                 errors.ECODE_INVAL)
11402

    
11403
    if not self.target_uuids:
11404
      raise errors.OpPrereqError("There are no possible target groups",
11405
                                 errors.ECODE_INVAL)
11406

    
11407
  def BuildHooksEnv(self):
11408
    """Build hooks env.
11409

11410
    """
11411
    assert self.target_uuids
11412

    
11413
    env = {
11414
      "TARGET_GROUPS": " ".join(self.target_uuids),
11415
      }
11416

    
11417
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
11418

    
11419
    return env
11420

    
11421
  def BuildHooksNodes(self):
11422
    """Build hooks nodes.
11423

11424
    """
11425
    mn = self.cfg.GetMasterNode()
11426
    return ([mn], [mn])
11427

    
11428
  def Exec(self, feedback_fn):
11429
    instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
11430

    
11431
    assert instances == [self.op.instance_name], "Instance not locked"
11432

    
11433
    ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
11434
                     instances=instances, target_groups=list(self.target_uuids))
11435

    
11436
    ial.Run(self.op.iallocator)
11437

    
11438
    if not ial.success:
11439
      raise errors.OpPrereqError("Can't compute solution for changing group of"
11440
                                 " instance '%s' using iallocator '%s': %s" %
11441
                                 (self.op.instance_name, self.op.iallocator,
11442
                                  ial.info),
11443
                                 errors.ECODE_NORES)
11444

    
11445
    jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
11446

    
11447
    self.LogInfo("Iallocator returned %s job(s) for changing group of"
11448
                 " instance '%s'", len(jobs), self.op.instance_name)
11449

    
11450
    return ResultWithJobs(jobs)
11451

    
11452

    
11453
class LUBackupQuery(NoHooksLU):
11454
  """Query the exports list
11455

11456
  """
11457
  REQ_BGL = False
11458

    
11459
  def ExpandNames(self):
11460
    self.needed_locks = {}
11461
    self.share_locks[locking.LEVEL_NODE] = 1
11462
    if not self.op.nodes:
11463
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11464
    else:
11465
      self.needed_locks[locking.LEVEL_NODE] = \
11466
        _GetWantedNodes(self, self.op.nodes)
11467

    
11468
  def Exec(self, feedback_fn):
11469
    """Compute the list of all the exported system images.
11470

11471
    @rtype: dict
11472
    @return: a dictionary with the structure node->(export-list)
11473
        where export-list is a list of the instances exported on
11474
        that node.
11475

11476
    """
11477
    self.nodes = self.owned_locks(locking.LEVEL_NODE)
11478
    rpcresult = self.rpc.call_export_list(self.nodes)
11479
    result = {}
11480
    for node in rpcresult:
11481
      if rpcresult[node].fail_msg:
11482
        result[node] = False
11483
      else:
11484
        result[node] = rpcresult[node].payload
11485

    
11486
    return result
11487

    
11488

    
11489
class LUBackupPrepare(NoHooksLU):
11490
  """Prepares an instance for an export and returns useful information.
11491

11492
  """
11493
  REQ_BGL = False
11494

    
11495
  def ExpandNames(self):
11496
    self._ExpandAndLockInstance()
11497

    
11498
  def CheckPrereq(self):
11499
    """Check prerequisites.
11500

11501
    """
11502
    instance_name = self.op.instance_name
11503

    
11504
    self.instance = self.cfg.GetInstanceInfo(instance_name)
11505
    assert self.instance is not None, \
11506
          "Cannot retrieve locked instance %s" % self.op.instance_name
11507
    _CheckNodeOnline(self, self.instance.primary_node)
11508

    
11509
    self._cds = _GetClusterDomainSecret()
11510

    
11511
  def Exec(self, feedback_fn):
11512
    """Prepares an instance for an export.
11513

11514
    """
11515
    instance = self.instance
11516

    
11517
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
11518
      salt = utils.GenerateSecret(8)
11519

    
11520
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
11521
      result = self.rpc.call_x509_cert_create(instance.primary_node,
11522
                                              constants.RIE_CERT_VALIDITY)
11523
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
11524

    
11525
      (name, cert_pem) = result.payload
11526

    
11527
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
11528
                                             cert_pem)
11529

    
11530
      return {
11531
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
11532
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
11533
                          salt),
11534
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
11535
        }
11536

    
11537
    return None
11538

    
11539

    
11540
class LUBackupExport(LogicalUnit):
11541
  """Export an instance to an image in the cluster.
11542

11543
  """
11544
  HPATH = "instance-export"
11545
  HTYPE = constants.HTYPE_INSTANCE
11546
  REQ_BGL = False
11547

    
11548
  def CheckArguments(self):
11549
    """Check the arguments.
11550

11551
    """
11552
    self.x509_key_name = self.op.x509_key_name
11553
    self.dest_x509_ca_pem = self.op.destination_x509_ca
11554

    
11555
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
11556
      if not self.x509_key_name:
11557
        raise errors.OpPrereqError("Missing X509 key name for encryption",
11558
                                   errors.ECODE_INVAL)
11559

    
11560
      if not self.dest_x509_ca_pem:
11561
        raise errors.OpPrereqError("Missing destination X509 CA",
11562
                                   errors.ECODE_INVAL)
11563

    
11564
  def ExpandNames(self):
11565
    self._ExpandAndLockInstance()
11566

    
11567
    # Lock all nodes for local exports
11568
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11569
      # FIXME: lock only instance primary and destination node
11570
      #
11571
      # Sad but true, for now we have do lock all nodes, as we don't know where
11572
      # the previous export might be, and in this LU we search for it and
11573
      # remove it from its current node. In the future we could fix this by:
11574
      #  - making a tasklet to search (share-lock all), then create the
11575
      #    new one, then one to remove, after
11576
      #  - removing the removal operation altogether
11577
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11578

    
11579
  def DeclareLocks(self, level):
11580
    """Last minute lock declaration."""
11581
    # All nodes are locked anyway, so nothing to do here.
11582

    
11583
  def BuildHooksEnv(self):
11584
    """Build hooks env.
11585

11586
    This will run on the master, primary node and target node.
11587

11588
    """
11589
    env = {
11590
      "EXPORT_MODE": self.op.mode,
11591
      "EXPORT_NODE": self.op.target_node,
11592
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
11593
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
11594
      # TODO: Generic function for boolean env variables
11595
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
11596
      }
11597

    
11598
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
11599

    
11600
    return env
11601

    
11602
  def BuildHooksNodes(self):
11603
    """Build hooks nodes.
11604

11605
    """
11606
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
11607

    
11608
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11609
      nl.append(self.op.target_node)
11610

    
11611
    return (nl, nl)
11612

    
11613
  def CheckPrereq(self):
11614
    """Check prerequisites.
11615

11616
    This checks that the instance and node names are valid.
11617

11618
    """
11619
    instance_name = self.op.instance_name
11620

    
11621
    self.instance = self.cfg.GetInstanceInfo(instance_name)
11622
    assert self.instance is not None, \
11623
          "Cannot retrieve locked instance %s" % self.op.instance_name
11624
    _CheckNodeOnline(self, self.instance.primary_node)
11625

    
11626
    if (self.op.remove_instance and self.instance.admin_up and
11627
        not self.op.shutdown):
11628
      raise errors.OpPrereqError("Can not remove instance without shutting it"
11629
                                 " down before")
11630

    
11631
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11632
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
11633
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
11634
      assert self.dst_node is not None
11635

    
11636
      _CheckNodeOnline(self, self.dst_node.name)
11637
      _CheckNodeNotDrained(self, self.dst_node.name)
11638

    
11639
      self._cds = None
11640
      self.dest_disk_info = None
11641
      self.dest_x509_ca = None
11642

    
11643
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
11644
      self.dst_node = None
11645

    
11646
      if len(self.op.target_node) != len(self.instance.disks):
11647
        raise errors.OpPrereqError(("Received destination information for %s"
11648
                                    " disks, but instance %s has %s disks") %
11649
                                   (len(self.op.target_node), instance_name,
11650
                                    len(self.instance.disks)),
11651
                                   errors.ECODE_INVAL)
11652

    
11653
      cds = _GetClusterDomainSecret()
11654

    
11655
      # Check X509 key name
11656
      try:
11657
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
11658
      except (TypeError, ValueError), err:
11659
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
11660

    
11661
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
11662
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
11663
                                   errors.ECODE_INVAL)
11664

    
11665
      # Load and verify CA
11666
      try:
11667
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
11668
      except OpenSSL.crypto.Error, err:
11669
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
11670
                                   (err, ), errors.ECODE_INVAL)
11671

    
11672
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
11673
      if errcode is not None:
11674
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
11675
                                   (msg, ), errors.ECODE_INVAL)
11676

    
11677
      self.dest_x509_ca = cert
11678

    
11679
      # Verify target information
11680
      disk_info = []
11681
      for idx, disk_data in enumerate(self.op.target_node):
11682
        try:
11683
          (host, port, magic) = \
11684
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
11685
        except errors.GenericError, err:
11686
          raise errors.OpPrereqError("Target info for disk %s: %s" %
11687
                                     (idx, err), errors.ECODE_INVAL)
11688

    
11689
        disk_info.append((host, port, magic))
11690

    
11691
      assert len(disk_info) == len(self.op.target_node)
11692
      self.dest_disk_info = disk_info
11693

    
11694
    else:
11695
      raise errors.ProgrammerError("Unhandled export mode %r" %
11696
                                   self.op.mode)
11697

    
11698
    # instance disk type verification
11699
    # TODO: Implement export support for file-based disks
11700
    for disk in self.instance.disks:
11701
      if disk.dev_type == constants.LD_FILE:
11702
        raise errors.OpPrereqError("Export not supported for instances with"
11703
                                   " file-based disks", errors.ECODE_INVAL)
11704

    
11705
  def _CleanupExports(self, feedback_fn):
11706
    """Removes exports of current instance from all other nodes.
11707

11708
    If an instance in a cluster with nodes A..D was exported to node C, its
11709
    exports will be removed from the nodes A, B and D.
11710

11711
    """
11712
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
11713

    
11714
    nodelist = self.cfg.GetNodeList()
11715
    nodelist.remove(self.dst_node.name)
11716

    
11717
    # on one-node clusters nodelist will be empty after the removal
11718
    # if we proceed the backup would be removed because OpBackupQuery
11719
    # substitutes an empty list with the full cluster node list.
11720
    iname = self.instance.name
11721
    if nodelist:
11722
      feedback_fn("Removing old exports for instance %s" % iname)
11723
      exportlist = self.rpc.call_export_list(nodelist)
11724
      for node in exportlist:
11725
        if exportlist[node].fail_msg:
11726
          continue
11727
        if iname in exportlist[node].payload:
11728
          msg = self.rpc.call_export_remove(node, iname).fail_msg
11729
          if msg:
11730
            self.LogWarning("Could not remove older export for instance %s"
11731
                            " on node %s: %s", iname, node, msg)
11732

    
11733
  def Exec(self, feedback_fn):
11734
    """Export an instance to an image in the cluster.
11735

11736
    """
11737
    assert self.op.mode in constants.EXPORT_MODES
11738

    
11739
    instance = self.instance
11740
    src_node = instance.primary_node
11741

    
11742
    if self.op.shutdown:
11743
      # shutdown the instance, but not the disks
11744
      feedback_fn("Shutting down instance %s" % instance.name)
11745
      result = self.rpc.call_instance_shutdown(src_node, instance,
11746
                                               self.op.shutdown_timeout)
11747
      # TODO: Maybe ignore failures if ignore_remove_failures is set
11748
      result.Raise("Could not shutdown instance %s on"
11749
                   " node %s" % (instance.name, src_node))
11750

    
11751
    # set the disks ID correctly since call_instance_start needs the
11752
    # correct drbd minor to create the symlinks
11753
    for disk in instance.disks:
11754
      self.cfg.SetDiskID(disk, src_node)
11755

    
11756
    activate_disks = (not instance.admin_up)
11757

    
11758
    if activate_disks:
11759
      # Activate the instance disks if we'exporting a stopped instance
11760
      feedback_fn("Activating disks for %s" % instance.name)
11761
      _StartInstanceDisks(self, instance, None)
11762

    
11763
    try:
11764
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
11765
                                                     instance)
11766

    
11767
      helper.CreateSnapshots()
11768
      try:
11769
        if (self.op.shutdown and instance.admin_up and
11770
            not self.op.remove_instance):
11771
          assert not activate_disks
11772
          feedback_fn("Starting instance %s" % instance.name)
11773
          result = self.rpc.call_instance_start(src_node, instance,
11774
                                                None, None, False)
11775
          msg = result.fail_msg
11776
          if msg:
11777
            feedback_fn("Failed to start instance: %s" % msg)
11778
            _ShutdownInstanceDisks(self, instance)
11779
            raise errors.OpExecError("Could not start instance: %s" % msg)
11780

    
11781
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
11782
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
11783
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
11784
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
11785
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
11786

    
11787
          (key_name, _, _) = self.x509_key_name
11788

    
11789
          dest_ca_pem = \
11790
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
11791
                                            self.dest_x509_ca)
11792

    
11793
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
11794
                                                     key_name, dest_ca_pem,
11795
                                                     timeouts)
11796
      finally:
11797
        helper.Cleanup()
11798

    
11799
      # Check for backwards compatibility
11800
      assert len(dresults) == len(instance.disks)
11801
      assert compat.all(isinstance(i, bool) for i in dresults), \
11802
             "Not all results are boolean: %r" % dresults
11803

    
11804
    finally:
11805
      if activate_disks:
11806
        feedback_fn("Deactivating disks for %s" % instance.name)
11807
        _ShutdownInstanceDisks(self, instance)
11808

    
11809
    if not (compat.all(dresults) and fin_resu):
11810
      failures = []
11811
      if not fin_resu:
11812
        failures.append("export finalization")
11813
      if not compat.all(dresults):
11814
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
11815
                               if not dsk)
11816
        failures.append("disk export: disk(s) %s" % fdsk)
11817

    
11818
      raise errors.OpExecError("Export failed, errors in %s" %
11819
                               utils.CommaJoin(failures))
11820

    
11821
    # At this point, the export was successful, we can cleanup/finish
11822

    
11823
    # Remove instance if requested
11824
    if self.op.remove_instance:
11825
      feedback_fn("Removing instance %s" % instance.name)
11826
      _RemoveInstance(self, feedback_fn, instance,
11827
                      self.op.ignore_remove_failures)
11828

    
11829
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11830
      self._CleanupExports(feedback_fn)
11831

    
11832
    return fin_resu, dresults
11833

    
11834

    
11835
class LUBackupRemove(NoHooksLU):
11836
  """Remove exports related to the named instance.
11837

11838
  """
11839
  REQ_BGL = False
11840

    
11841
  def ExpandNames(self):
11842
    self.needed_locks = {}
11843
    # We need all nodes to be locked in order for RemoveExport to work, but we
11844
    # don't need to lock the instance itself, as nothing will happen to it (and
11845
    # we can remove exports also for a removed instance)
11846
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11847

    
11848
  def Exec(self, feedback_fn):
11849
    """Remove any export.
11850

11851
    """
11852
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
11853
    # If the instance was not found we'll try with the name that was passed in.
11854
    # This will only work if it was an FQDN, though.
11855
    fqdn_warn = False
11856
    if not instance_name:
11857
      fqdn_warn = True
11858
      instance_name = self.op.instance_name
11859

    
11860
    locked_nodes = self.owned_locks(locking.LEVEL_NODE)
11861
    exportlist = self.rpc.call_export_list(locked_nodes)
11862
    found = False
11863
    for node in exportlist:
11864
      msg = exportlist[node].fail_msg
11865
      if msg:
11866
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
11867
        continue
11868
      if instance_name in exportlist[node].payload:
11869
        found = True
11870
        result = self.rpc.call_export_remove(node, instance_name)
11871
        msg = result.fail_msg
11872
        if msg:
11873
          logging.error("Could not remove export for instance %s"
11874
                        " on node %s: %s", instance_name, node, msg)
11875

    
11876
    if fqdn_warn and not found:
11877
      feedback_fn("Export not found. If trying to remove an export belonging"
11878
                  " to a deleted instance please use its Fully Qualified"
11879
                  " Domain Name.")
11880

    
11881

    
11882
class LUGroupAdd(LogicalUnit):
11883
  """Logical unit for creating node groups.
11884

11885
  """
11886
  HPATH = "group-add"
11887
  HTYPE = constants.HTYPE_GROUP
11888
  REQ_BGL = False
11889

    
11890
  def ExpandNames(self):
11891
    # We need the new group's UUID here so that we can create and acquire the
11892
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
11893
    # that it should not check whether the UUID exists in the configuration.
11894
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
11895
    self.needed_locks = {}
11896
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
11897

    
11898
  def CheckPrereq(self):
11899
    """Check prerequisites.
11900

11901
    This checks that the given group name is not an existing node group
11902
    already.
11903

11904
    """
11905
    try:
11906
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11907
    except errors.OpPrereqError:
11908
      pass
11909
    else:
11910
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
11911
                                 " node group (UUID: %s)" %
11912
                                 (self.op.group_name, existing_uuid),
11913
                                 errors.ECODE_EXISTS)
11914

    
11915
    if self.op.ndparams:
11916
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
11917

    
11918
  def BuildHooksEnv(self):
11919
    """Build hooks env.
11920

11921
    """
11922
    return {
11923
      "GROUP_NAME": self.op.group_name,
11924
      }
11925

    
11926
  def BuildHooksNodes(self):
11927
    """Build hooks nodes.
11928

11929
    """
11930
    mn = self.cfg.GetMasterNode()
11931
    return ([mn], [mn])
11932

    
11933
  def Exec(self, feedback_fn):
11934
    """Add the node group to the cluster.
11935

11936
    """
11937
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
11938
                                  uuid=self.group_uuid,
11939
                                  alloc_policy=self.op.alloc_policy,
11940
                                  ndparams=self.op.ndparams)
11941

    
11942
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
11943
    del self.remove_locks[locking.LEVEL_NODEGROUP]
11944

    
11945

    
11946
class LUGroupAssignNodes(NoHooksLU):
11947
  """Logical unit for assigning nodes to groups.
11948

11949
  """
11950
  REQ_BGL = False
11951

    
11952
  def ExpandNames(self):
11953
    # These raise errors.OpPrereqError on their own:
11954
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11955
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
11956

    
11957
    # We want to lock all the affected nodes and groups. We have readily
11958
    # available the list of nodes, and the *destination* group. To gather the
11959
    # list of "source" groups, we need to fetch node information later on.
11960
    self.needed_locks = {
11961
      locking.LEVEL_NODEGROUP: set([self.group_uuid]),
11962
      locking.LEVEL_NODE: self.op.nodes,
11963
      }
11964

    
11965
  def DeclareLocks(self, level):
11966
    if level == locking.LEVEL_NODEGROUP:
11967
      assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
11968

    
11969
      # Try to get all affected nodes' groups without having the group or node
11970
      # lock yet. Needs verification later in the code flow.
11971
      groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
11972

    
11973
      self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
11974

    
11975
  def CheckPrereq(self):
11976
    """Check prerequisites.
11977

11978
    """
11979
    assert self.needed_locks[locking.LEVEL_NODEGROUP]
11980
    assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
11981
            frozenset(self.op.nodes))
11982

    
11983
    expected_locks = (set([self.group_uuid]) |
11984
                      self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
11985
    actual_locks = self.owned_locks(locking.LEVEL_NODEGROUP)
11986
    if actual_locks != expected_locks:
11987
      raise errors.OpExecError("Nodes changed groups since locks were acquired,"
11988
                               " current groups are '%s', used to be '%s'" %
11989
                               (utils.CommaJoin(expected_locks),
11990
                                utils.CommaJoin(actual_locks)))
11991

    
11992
    self.node_data = self.cfg.GetAllNodesInfo()
11993
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
11994
    instance_data = self.cfg.GetAllInstancesInfo()
11995

    
11996
    if self.group is None:
11997
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
11998
                               (self.op.group_name, self.group_uuid))
11999

    
12000
    (new_splits, previous_splits) = \
12001
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
12002
                                             for node in self.op.nodes],
12003
                                            self.node_data, instance_data)
12004

    
12005
    if new_splits:
12006
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
12007

    
12008
      if not self.op.force:
12009
        raise errors.OpExecError("The following instances get split by this"
12010
                                 " change and --force was not given: %s" %
12011
                                 fmt_new_splits)
12012
      else:
12013
        self.LogWarning("This operation will split the following instances: %s",
12014
                        fmt_new_splits)
12015

    
12016
        if previous_splits:
12017
          self.LogWarning("In addition, these already-split instances continue"
12018
                          " to be split across groups: %s",
12019
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
12020

    
12021
  def Exec(self, feedback_fn):
12022
    """Assign nodes to a new group.
12023

12024
    """
12025
    for node in self.op.nodes:
12026
      self.node_data[node].group = self.group_uuid
12027

    
12028
    # FIXME: Depends on side-effects of modifying the result of
12029
    # C{cfg.GetAllNodesInfo}
12030

    
12031
    self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
12032

    
12033
  @staticmethod
12034
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
12035
    """Check for split instances after a node assignment.
12036

12037
    This method considers a series of node assignments as an atomic operation,
12038
    and returns information about split instances after applying the set of
12039
    changes.
12040

12041
    In particular, it returns information about newly split instances, and
12042
    instances that were already split, and remain so after the change.
12043

12044
    Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
12045
    considered.
12046

12047
    @type changes: list of (node_name, new_group_uuid) pairs.
12048
    @param changes: list of node assignments to consider.
12049
    @param node_data: a dict with data for all nodes
12050
    @param instance_data: a dict with all instances to consider
12051
    @rtype: a two-tuple
12052
    @return: a list of instances that were previously okay and result split as a
12053
      consequence of this change, and a list of instances that were previously
12054
      split and this change does not fix.
12055

12056
    """
12057
    changed_nodes = dict((node, group) for node, group in changes
12058
                         if node_data[node].group != group)
12059

    
12060
    all_split_instances = set()
12061
    previously_split_instances = set()
12062

    
12063
    def InstanceNodes(instance):
12064
      return [instance.primary_node] + list(instance.secondary_nodes)
12065

    
12066
    for inst in instance_data.values():
12067
      if inst.disk_template not in constants.DTS_INT_MIRROR:
12068
        continue
12069

    
12070
      instance_nodes = InstanceNodes(inst)
12071

    
12072
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
12073
        previously_split_instances.add(inst.name)
12074

    
12075
      if len(set(changed_nodes.get(node, node_data[node].group)
12076
                 for node in instance_nodes)) > 1:
12077
        all_split_instances.add(inst.name)
12078

    
12079
    return (list(all_split_instances - previously_split_instances),
12080
            list(previously_split_instances & all_split_instances))
12081

    
12082

    
12083
class _GroupQuery(_QueryBase):
12084
  FIELDS = query.GROUP_FIELDS
12085

    
12086
  def ExpandNames(self, lu):
12087
    lu.needed_locks = {}
12088

    
12089
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
12090
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
12091

    
12092
    if not self.names:
12093
      self.wanted = [name_to_uuid[name]
12094
                     for name in utils.NiceSort(name_to_uuid.keys())]
12095
    else:
12096
      # Accept names to be either names or UUIDs.
12097
      missing = []
12098
      self.wanted = []
12099
      all_uuid = frozenset(self._all_groups.keys())
12100

    
12101
      for name in self.names:
12102
        if name in all_uuid:
12103
          self.wanted.append(name)
12104
        elif name in name_to_uuid:
12105
          self.wanted.append(name_to_uuid[name])
12106
        else:
12107
          missing.append(name)
12108

    
12109
      if missing:
12110
        raise errors.OpPrereqError("Some groups do not exist: %s" %
12111
                                   utils.CommaJoin(missing),
12112
                                   errors.ECODE_NOENT)
12113

    
12114
  def DeclareLocks(self, lu, level):
12115
    pass
12116

    
12117
  def _GetQueryData(self, lu):
12118
    """Computes the list of node groups and their attributes.
12119

12120
    """
12121
    do_nodes = query.GQ_NODE in self.requested_data
12122
    do_instances = query.GQ_INST in self.requested_data
12123

    
12124
    group_to_nodes = None
12125
    group_to_instances = None
12126

    
12127
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
12128
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
12129
    # latter GetAllInstancesInfo() is not enough, for we have to go through
12130
    # instance->node. Hence, we will need to process nodes even if we only need
12131
    # instance information.
12132
    if do_nodes or do_instances:
12133
      all_nodes = lu.cfg.GetAllNodesInfo()
12134
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
12135
      node_to_group = {}
12136

    
12137
      for node in all_nodes.values():
12138
        if node.group in group_to_nodes:
12139
          group_to_nodes[node.group].append(node.name)
12140
          node_to_group[node.name] = node.group
12141

    
12142
      if do_instances:
12143
        all_instances = lu.cfg.GetAllInstancesInfo()
12144
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
12145

    
12146
        for instance in all_instances.values():
12147
          node = instance.primary_node
12148
          if node in node_to_group:
12149
            group_to_instances[node_to_group[node]].append(instance.name)
12150

    
12151
        if not do_nodes:
12152
          # Do not pass on node information if it was not requested.
12153
          group_to_nodes = None
12154

    
12155
    return query.GroupQueryData([self._all_groups[uuid]
12156
                                 for uuid in self.wanted],
12157
                                group_to_nodes, group_to_instances)
12158

    
12159

    
12160
class LUGroupQuery(NoHooksLU):
12161
  """Logical unit for querying node groups.
12162

12163
  """
12164
  REQ_BGL = False
12165

    
12166
  def CheckArguments(self):
12167
    self.gq = _GroupQuery(qlang.MakeSimpleFilter("name", self.op.names),
12168
                          self.op.output_fields, False)
12169

    
12170
  def ExpandNames(self):
12171
    self.gq.ExpandNames(self)
12172

    
12173
  def DeclareLocks(self, level):
12174
    self.gq.DeclareLocks(self, level)
12175

    
12176
  def Exec(self, feedback_fn):
12177
    return self.gq.OldStyleQuery(self)
12178

    
12179

    
12180
class LUGroupSetParams(LogicalUnit):
12181
  """Modifies the parameters of a node group.
12182

12183
  """
12184
  HPATH = "group-modify"
12185
  HTYPE = constants.HTYPE_GROUP
12186
  REQ_BGL = False
12187

    
12188
  def CheckArguments(self):
12189
    all_changes = [
12190
      self.op.ndparams,
12191
      self.op.alloc_policy,
12192
      ]
12193

    
12194
    if all_changes.count(None) == len(all_changes):
12195
      raise errors.OpPrereqError("Please pass at least one modification",
12196
                                 errors.ECODE_INVAL)
12197

    
12198
  def ExpandNames(self):
12199
    # This raises errors.OpPrereqError on its own:
12200
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12201

    
12202
    self.needed_locks = {
12203
      locking.LEVEL_NODEGROUP: [self.group_uuid],
12204
      }
12205

    
12206
  def CheckPrereq(self):
12207
    """Check prerequisites.
12208

12209
    """
12210
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
12211

    
12212
    if self.group is None:
12213
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
12214
                               (self.op.group_name, self.group_uuid))
12215

    
12216
    if self.op.ndparams:
12217
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
12218
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
12219
      self.new_ndparams = new_ndparams
12220

    
12221
  def BuildHooksEnv(self):
12222
    """Build hooks env.
12223

12224
    """
12225
    return {
12226
      "GROUP_NAME": self.op.group_name,
12227
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
12228
      }
12229

    
12230
  def BuildHooksNodes(self):
12231
    """Build hooks nodes.
12232

12233
    """
12234
    mn = self.cfg.GetMasterNode()
12235
    return ([mn], [mn])
12236

    
12237
  def Exec(self, feedback_fn):
12238
    """Modifies the node group.
12239

12240
    """
12241
    result = []
12242

    
12243
    if self.op.ndparams:
12244
      self.group.ndparams = self.new_ndparams
12245
      result.append(("ndparams", str(self.group.ndparams)))
12246

    
12247
    if self.op.alloc_policy:
12248
      self.group.alloc_policy = self.op.alloc_policy
12249

    
12250
    self.cfg.Update(self.group, feedback_fn)
12251
    return result
12252

    
12253

    
12254
class LUGroupRemove(LogicalUnit):
12255
  HPATH = "group-remove"
12256
  HTYPE = constants.HTYPE_GROUP
12257
  REQ_BGL = False
12258

    
12259
  def ExpandNames(self):
12260
    # This will raises errors.OpPrereqError on its own:
12261
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12262
    self.needed_locks = {
12263
      locking.LEVEL_NODEGROUP: [self.group_uuid],
12264
      }
12265

    
12266
  def CheckPrereq(self):
12267
    """Check prerequisites.
12268

12269
    This checks that the given group name exists as a node group, that is
12270
    empty (i.e., contains no nodes), and that is not the last group of the
12271
    cluster.
12272

12273
    """
12274
    # Verify that the group is empty.
12275
    group_nodes = [node.name
12276
                   for node in self.cfg.GetAllNodesInfo().values()
12277
                   if node.group == self.group_uuid]
12278

    
12279
    if group_nodes:
12280
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
12281
                                 " nodes: %s" %
12282
                                 (self.op.group_name,
12283
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
12284
                                 errors.ECODE_STATE)
12285

    
12286
    # Verify the cluster would not be left group-less.
12287
    if len(self.cfg.GetNodeGroupList()) == 1:
12288
      raise errors.OpPrereqError("Group '%s' is the only group,"
12289
                                 " cannot be removed" %
12290
                                 self.op.group_name,
12291
                                 errors.ECODE_STATE)
12292

    
12293
  def BuildHooksEnv(self):
12294
    """Build hooks env.
12295

12296
    """
12297
    return {
12298
      "GROUP_NAME": self.op.group_name,
12299
      }
12300

    
12301
  def BuildHooksNodes(self):
12302
    """Build hooks nodes.
12303

12304
    """
12305
    mn = self.cfg.GetMasterNode()
12306
    return ([mn], [mn])
12307

    
12308
  def Exec(self, feedback_fn):
12309
    """Remove the node group.
12310

12311
    """
12312
    try:
12313
      self.cfg.RemoveNodeGroup(self.group_uuid)
12314
    except errors.ConfigurationError:
12315
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
12316
                               (self.op.group_name, self.group_uuid))
12317

    
12318
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
12319

    
12320

    
12321
class LUGroupRename(LogicalUnit):
12322
  HPATH = "group-rename"
12323
  HTYPE = constants.HTYPE_GROUP
12324
  REQ_BGL = False
12325

    
12326
  def ExpandNames(self):
12327
    # This raises errors.OpPrereqError on its own:
12328
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12329

    
12330
    self.needed_locks = {
12331
      locking.LEVEL_NODEGROUP: [self.group_uuid],
12332
      }
12333

    
12334
  def CheckPrereq(self):
12335
    """Check prerequisites.
12336

12337
    Ensures requested new name is not yet used.
12338

12339
    """
12340
    try:
12341
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
12342
    except errors.OpPrereqError:
12343
      pass
12344
    else:
12345
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
12346
                                 " node group (UUID: %s)" %
12347
                                 (self.op.new_name, new_name_uuid),
12348
                                 errors.ECODE_EXISTS)
12349

    
12350
  def BuildHooksEnv(self):
12351
    """Build hooks env.
12352

12353
    """
12354
    return {
12355
      "OLD_NAME": self.op.group_name,
12356
      "NEW_NAME": self.op.new_name,
12357
      }
12358

    
12359
  def BuildHooksNodes(self):
12360
    """Build hooks nodes.
12361

12362
    """
12363
    mn = self.cfg.GetMasterNode()
12364

    
12365
    all_nodes = self.cfg.GetAllNodesInfo()
12366
    all_nodes.pop(mn, None)
12367

    
12368
    run_nodes = [mn]
12369
    run_nodes.extend(node.name for node in all_nodes.values()
12370
                     if node.group == self.group_uuid)
12371

    
12372
    return (run_nodes, run_nodes)
12373

    
12374
  def Exec(self, feedback_fn):
12375
    """Rename the node group.
12376

12377
    """
12378
    group = self.cfg.GetNodeGroup(self.group_uuid)
12379

    
12380
    if group is None:
12381
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
12382
                               (self.op.group_name, self.group_uuid))
12383

    
12384
    group.name = self.op.new_name
12385
    self.cfg.Update(group, feedback_fn)
12386

    
12387
    return self.op.new_name
12388

    
12389

    
12390
class LUGroupEvacuate(LogicalUnit):
12391
  HPATH = "group-evacuate"
12392
  HTYPE = constants.HTYPE_GROUP
12393
  REQ_BGL = False
12394

    
12395
  def ExpandNames(self):
12396
    # This raises errors.OpPrereqError on its own:
12397
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12398

    
12399
    if self.op.target_groups:
12400
      self.req_target_uuids = map(self.cfg.LookupNodeGroup,
12401
                                  self.op.target_groups)
12402
    else:
12403
      self.req_target_uuids = []
12404

    
12405
    if self.group_uuid in self.req_target_uuids:
12406
      raise errors.OpPrereqError("Group to be evacuated (%s) can not be used"
12407
                                 " as a target group (targets are %s)" %
12408
                                 (self.group_uuid,
12409
                                  utils.CommaJoin(self.req_target_uuids)),
12410
                                 errors.ECODE_INVAL)
12411

    
12412
    self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
12413

    
12414
    self.share_locks = _ShareAll()
12415
    self.needed_locks = {
12416
      locking.LEVEL_INSTANCE: [],
12417
      locking.LEVEL_NODEGROUP: [],
12418
      locking.LEVEL_NODE: [],
12419
      }
12420

    
12421
  def DeclareLocks(self, level):
12422
    if level == locking.LEVEL_INSTANCE:
12423
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
12424

    
12425
      # Lock instances optimistically, needs verification once node and group
12426
      # locks have been acquired
12427
      self.needed_locks[locking.LEVEL_INSTANCE] = \
12428
        self.cfg.GetNodeGroupInstances(self.group_uuid)
12429

    
12430
    elif level == locking.LEVEL_NODEGROUP:
12431
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
12432

    
12433
      if self.req_target_uuids:
12434
        lock_groups = set([self.group_uuid] + self.req_target_uuids)
12435

    
12436
        # Lock all groups used by instances optimistically; this requires going
12437
        # via the node before it's locked, requiring verification later on
12438
        lock_groups.update(group_uuid
12439
                           for instance_name in
12440
                             self.owned_locks(locking.LEVEL_INSTANCE)
12441
                           for group_uuid in
12442
                             self.cfg.GetInstanceNodeGroups(instance_name))
12443
      else:
12444
        # No target groups, need to lock all of them
12445
        lock_groups = locking.ALL_SET
12446

    
12447
      self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
12448

    
12449
    elif level == locking.LEVEL_NODE:
12450
      # This will only lock the nodes in the group to be evacuated which
12451
      # contain actual instances
12452
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
12453
      self._LockInstancesNodes()
12454

    
12455
      # Lock all nodes in group to be evacuated and target groups
12456
      owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
12457
      assert self.group_uuid in owned_groups
12458
      member_nodes = [node_name
12459
                      for group in owned_groups
12460
                      for node_name in self.cfg.GetNodeGroup(group).members]
12461
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
12462

    
12463
  def CheckPrereq(self):
12464
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
12465
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
12466
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
12467

    
12468
    assert owned_groups.issuperset(self.req_target_uuids)
12469
    assert self.group_uuid in owned_groups
12470

    
12471
    # Check if locked instances are still correct
12472
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
12473

    
12474
    # Get instance information
12475
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
12476

    
12477
    # Check if node groups for locked instances are still correct
12478
    for instance_name in owned_instances:
12479
      inst = self.instances[instance_name]
12480
      assert owned_nodes.issuperset(inst.all_nodes), \
12481
        "Instance %s's nodes changed while we kept the lock" % instance_name
12482

    
12483
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
12484
                                             owned_groups)
12485

    
12486
      assert self.group_uuid in inst_groups, \
12487
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
12488

    
12489
    if self.req_target_uuids:
12490
      # User requested specific target groups
12491
      self.target_uuids = self.req_target_uuids
12492
    else:
12493
      # All groups except the one to be evacuated are potential targets
12494
      self.target_uuids = [group_uuid for group_uuid in owned_groups
12495
                           if group_uuid != self.group_uuid]
12496

    
12497
      if not self.target_uuids:
12498
        raise errors.OpPrereqError("There are no possible target groups",
12499
                                   errors.ECODE_INVAL)
12500

    
12501
  def BuildHooksEnv(self):
12502
    """Build hooks env.
12503

12504
    """
12505
    return {
12506
      "GROUP_NAME": self.op.group_name,
12507
      "TARGET_GROUPS": " ".join(self.target_uuids),
12508
      }
12509

    
12510
  def BuildHooksNodes(self):
12511
    """Build hooks nodes.
12512

12513
    """
12514
    mn = self.cfg.GetMasterNode()
12515

    
12516
    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
12517

    
12518
    run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members
12519

    
12520
    return (run_nodes, run_nodes)
12521

    
12522
  def Exec(self, feedback_fn):
12523
    instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
12524

    
12525
    assert self.group_uuid not in self.target_uuids
12526

    
12527
    ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
12528
                     instances=instances, target_groups=self.target_uuids)
12529

    
12530
    ial.Run(self.op.iallocator)
12531

    
12532
    if not ial.success:
12533
      raise errors.OpPrereqError("Can't compute group evacuation using"
12534
                                 " iallocator '%s': %s" %
12535
                                 (self.op.iallocator, ial.info),
12536
                                 errors.ECODE_NORES)
12537

    
12538
    jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
12539

    
12540
    self.LogInfo("Iallocator returned %s job(s) for evacuating node group %s",
12541
                 len(jobs), self.op.group_name)
12542

    
12543
    return ResultWithJobs(jobs)
12544

    
12545

    
12546
class TagsLU(NoHooksLU): # pylint: disable=W0223
12547
  """Generic tags LU.
12548

12549
  This is an abstract class which is the parent of all the other tags LUs.
12550

12551
  """
12552
  def ExpandNames(self):
12553
    self.group_uuid = None
12554
    self.needed_locks = {}
12555
    if self.op.kind == constants.TAG_NODE:
12556
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
12557
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
12558
    elif self.op.kind == constants.TAG_INSTANCE:
12559
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
12560
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
12561
    elif self.op.kind == constants.TAG_NODEGROUP:
12562
      self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)
12563

    
12564
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
12565
    # not possible to acquire the BGL based on opcode parameters)
12566

    
12567
  def CheckPrereq(self):
12568
    """Check prerequisites.
12569

12570
    """
12571
    if self.op.kind == constants.TAG_CLUSTER:
12572
      self.target = self.cfg.GetClusterInfo()
12573
    elif self.op.kind == constants.TAG_NODE:
12574
      self.target = self.cfg.GetNodeInfo(self.op.name)
12575
    elif self.op.kind == constants.TAG_INSTANCE:
12576
      self.target = self.cfg.GetInstanceInfo(self.op.name)
12577
    elif self.op.kind == constants.TAG_NODEGROUP:
12578
      self.target = self.cfg.GetNodeGroup(self.group_uuid)
12579
    else:
12580
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
12581
                                 str(self.op.kind), errors.ECODE_INVAL)
12582

    
12583

    
12584
class LUTagsGet(TagsLU):
12585
  """Returns the tags of a given object.
12586

12587
  """
12588
  REQ_BGL = False
12589

    
12590
  def ExpandNames(self):
12591
    TagsLU.ExpandNames(self)
12592

    
12593
    # Share locks as this is only a read operation
12594
    self.share_locks = _ShareAll()
12595

    
12596
  def Exec(self, feedback_fn):
12597
    """Returns the tag list.
12598

12599
    """
12600
    return list(self.target.GetTags())
12601

    
12602

    
12603
class LUTagsSearch(NoHooksLU):
12604
  """Searches the tags for a given pattern.
12605

12606
  """
12607
  REQ_BGL = False
12608

    
12609
  def ExpandNames(self):
12610
    self.needed_locks = {}
12611

    
12612
  def CheckPrereq(self):
12613
    """Check prerequisites.
12614

12615
    This checks the pattern passed for validity by compiling it.
12616

12617
    """
12618
    try:
12619
      self.re = re.compile(self.op.pattern)
12620
    except re.error, err:
12621
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
12622
                                 (self.op.pattern, err), errors.ECODE_INVAL)
12623

    
12624
  def Exec(self, feedback_fn):
12625
    """Returns the tag list.
12626

12627
    """
12628
    cfg = self.cfg
12629
    tgts = [("/cluster", cfg.GetClusterInfo())]
12630
    ilist = cfg.GetAllInstancesInfo().values()
12631
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
12632
    nlist = cfg.GetAllNodesInfo().values()
12633
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
12634
    tgts.extend(("/nodegroup/%s" % n.name, n)
12635
                for n in cfg.GetAllNodeGroupsInfo().values())
12636
    results = []
12637
    for path, target in tgts:
12638
      for tag in target.GetTags():
12639
        if self.re.search(tag):
12640
          results.append((path, tag))
12641
    return results
12642

    
12643

    
12644
class LUTagsSet(TagsLU):
12645
  """Sets a tag on a given object.
12646

12647
  """
12648
  REQ_BGL = False
12649

    
12650
  def CheckPrereq(self):
12651
    """Check prerequisites.
12652

12653
    This checks the type and length of the tag name and value.
12654

12655
    """
12656
    TagsLU.CheckPrereq(self)
12657
    for tag in self.op.tags:
12658
      objects.TaggableObject.ValidateTag(tag)
12659

    
12660
  def Exec(self, feedback_fn):
12661
    """Sets the tag.
12662

12663
    """
12664
    try:
12665
      for tag in self.op.tags:
12666
        self.target.AddTag(tag)
12667
    except errors.TagError, err:
12668
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
12669
    self.cfg.Update(self.target, feedback_fn)
12670

    
12671

    
12672
class LUTagsDel(TagsLU):
12673
  """Delete a list of tags from a given object.
12674

12675
  """
12676
  REQ_BGL = False
12677

    
12678
  def CheckPrereq(self):
12679
    """Check prerequisites.
12680

12681
    This checks that we have the given tag.
12682

12683
    """
12684
    TagsLU.CheckPrereq(self)
12685
    for tag in self.op.tags:
12686
      objects.TaggableObject.ValidateTag(tag)
12687
    del_tags = frozenset(self.op.tags)
12688
    cur_tags = self.target.GetTags()
12689

    
12690
    diff_tags = del_tags - cur_tags
12691
    if diff_tags:
12692
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
12693
      raise errors.OpPrereqError("Tag(s) %s not found" %
12694
                                 (utils.CommaJoin(diff_names), ),
12695
                                 errors.ECODE_NOENT)
12696

    
12697
  def Exec(self, feedback_fn):
12698
    """Remove the tag from the object.
12699

12700
    """
12701
    for tag in self.op.tags:
12702
      self.target.RemoveTag(tag)
12703
    self.cfg.Update(self.target, feedback_fn)
12704

    
12705

    
12706
class LUTestDelay(NoHooksLU):
12707
  """Sleep for a specified amount of time.
12708

12709
  This LU sleeps on the master and/or nodes for a specified amount of
12710
  time.
12711

12712
  """
12713
  REQ_BGL = False
12714

    
12715
  def ExpandNames(self):
12716
    """Expand names and set required locks.
12717

12718
    This expands the node list, if any.
12719

12720
    """
12721
    self.needed_locks = {}
12722
    if self.op.on_nodes:
12723
      # _GetWantedNodes can be used here, but is not always appropriate to use
12724
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
12725
      # more information.
12726
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
12727
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
12728

    
12729
  def _TestDelay(self):
12730
    """Do the actual sleep.
12731

12732
    """
12733
    if self.op.on_master:
12734
      if not utils.TestDelay(self.op.duration):
12735
        raise errors.OpExecError("Error during master delay test")
12736
    if self.op.on_nodes:
12737
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
12738
      for node, node_result in result.items():
12739
        node_result.Raise("Failure during rpc call to node %s" % node)
12740

    
12741
  def Exec(self, feedback_fn):
12742
    """Execute the test delay opcode, with the wanted repetitions.
12743

12744
    """
12745
    if self.op.repeat == 0:
12746
      self._TestDelay()
12747
    else:
12748
      top_value = self.op.repeat - 1
12749
      for i in range(self.op.repeat):
12750
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
12751
        self._TestDelay()
12752

    
12753

    
12754
class LUTestJqueue(NoHooksLU):
12755
  """Utility LU to test some aspects of the job queue.
12756

12757
  """
12758
  REQ_BGL = False
12759

    
12760
  # Must be lower than default timeout for WaitForJobChange to see whether it
12761
  # notices changed jobs
12762
  _CLIENT_CONNECT_TIMEOUT = 20.0
12763
  _CLIENT_CONFIRM_TIMEOUT = 60.0
12764

    
12765
  @classmethod
12766
  def _NotifyUsingSocket(cls, cb, errcls):
12767
    """Opens a Unix socket and waits for another program to connect.
12768

12769
    @type cb: callable
12770
    @param cb: Callback to send socket name to client
12771
    @type errcls: class
12772
    @param errcls: Exception class to use for errors
12773

12774
    """
12775
    # Using a temporary directory as there's no easy way to create temporary
12776
    # sockets without writing a custom loop around tempfile.mktemp and
12777
    # socket.bind
12778
    tmpdir = tempfile.mkdtemp()
12779
    try:
12780
      tmpsock = utils.PathJoin(tmpdir, "sock")
12781

    
12782
      logging.debug("Creating temporary socket at %s", tmpsock)
12783
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
12784
      try:
12785
        sock.bind(tmpsock)
12786
        sock.listen(1)
12787

    
12788
        # Send details to client
12789
        cb(tmpsock)
12790

    
12791
        # Wait for client to connect before continuing
12792
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
12793
        try:
12794
          (conn, _) = sock.accept()
12795
        except socket.error, err:
12796
          raise errcls("Client didn't connect in time (%s)" % err)
12797
      finally:
12798
        sock.close()
12799
    finally:
12800
      # Remove as soon as client is connected
12801
      shutil.rmtree(tmpdir)
12802

    
12803
    # Wait for client to close
12804
    try:
12805
      try:
12806
        # pylint: disable=E1101
12807
        # Instance of '_socketobject' has no ... member
12808
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
12809
        conn.recv(1)
12810
      except socket.error, err:
12811
        raise errcls("Client failed to confirm notification (%s)" % err)
12812
    finally:
12813
      conn.close()
12814

    
12815
  def _SendNotification(self, test, arg, sockname):
12816
    """Sends a notification to the client.
12817

12818
    @type test: string
12819
    @param test: Test name
12820
    @param arg: Test argument (depends on test)
12821
    @type sockname: string
12822
    @param sockname: Socket path
12823

12824
    """
12825
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
12826

    
12827
  def _Notify(self, prereq, test, arg):
12828
    """Notifies the client of a test.
12829

12830
    @type prereq: bool
12831
    @param prereq: Whether this is a prereq-phase test
12832
    @type test: string
12833
    @param test: Test name
12834
    @param arg: Test argument (depends on test)
12835

12836
    """
12837
    if prereq:
12838
      errcls = errors.OpPrereqError
12839
    else:
12840
      errcls = errors.OpExecError
12841

    
12842
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
12843
                                                  test, arg),
12844
                                   errcls)
12845

    
12846
  def CheckArguments(self):
12847
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
12848
    self.expandnames_calls = 0
12849

    
12850
  def ExpandNames(self):
12851
    checkargs_calls = getattr(self, "checkargs_calls", 0)
12852
    if checkargs_calls < 1:
12853
      raise errors.ProgrammerError("CheckArguments was not called")
12854

    
12855
    self.expandnames_calls += 1
12856

    
12857
    if self.op.notify_waitlock:
12858
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
12859

    
12860
    self.LogInfo("Expanding names")
12861

    
12862
    # Get lock on master node (just to get a lock, not for a particular reason)
12863
    self.needed_locks = {
12864
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
12865
      }
12866

    
12867
  def Exec(self, feedback_fn):
12868
    if self.expandnames_calls < 1:
12869
      raise errors.ProgrammerError("ExpandNames was not called")
12870

    
12871
    if self.op.notify_exec:
12872
      self._Notify(False, constants.JQT_EXEC, None)
12873

    
12874
    self.LogInfo("Executing")
12875

    
12876
    if self.op.log_messages:
12877
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
12878
      for idx, msg in enumerate(self.op.log_messages):
12879
        self.LogInfo("Sending log message %s", idx + 1)
12880
        feedback_fn(constants.JQT_MSGPREFIX + msg)
12881
        # Report how many test messages have been sent
12882
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
12883

    
12884
    if self.op.fail:
12885
      raise errors.OpExecError("Opcode failure was requested")
12886

    
12887
    return True
12888

    
12889

    
12890
class IAllocator(object):
12891
  """IAllocator framework.
12892

12893
  An IAllocator instance has three sets of attributes:
12894
    - cfg that is needed to query the cluster
12895
    - input data (all members of the _KEYS class attribute are required)
12896
    - four buffer attributes (in|out_data|text), that represent the
12897
      input (to the external script) in text and data structure format,
12898
      and the output from it, again in two formats
12899
    - the result variables from the script (success, info, nodes) for
12900
      easy usage
12901

12902
  """
12903
  # pylint: disable=R0902
12904
  # lots of instance attributes
12905

    
12906
  def __init__(self, cfg, rpc, mode, **kwargs):
12907
    self.cfg = cfg
12908
    self.rpc = rpc
12909
    # init buffer variables
12910
    self.in_text = self.out_text = self.in_data = self.out_data = None
12911
    # init all input fields so that pylint is happy
12912
    self.mode = mode
12913
    self.memory = self.disks = self.disk_template = None
12914
    self.os = self.tags = self.nics = self.vcpus = None
12915
    self.hypervisor = None
12916
    self.relocate_from = None
12917
    self.name = None
12918
    self.instances = None
12919
    self.evac_mode = None
12920
    self.target_groups = []
12921
    # computed fields
12922
    self.required_nodes = None
12923
    # init result fields
12924
    self.success = self.info = self.result = None
12925

    
12926
    try:
12927
      (fn, keydata, self._result_check) = self._MODE_DATA[self.mode]
12928
    except KeyError:
12929
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
12930
                                   " IAllocator" % self.mode)
12931

    
12932
    keyset = [n for (n, _) in keydata]
12933

    
12934
    for key in kwargs:
12935
      if key not in keyset:
12936
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
12937
                                     " IAllocator" % key)
12938
      setattr(self, key, kwargs[key])
12939

    
12940
    for key in keyset:
12941
      if key not in kwargs:
12942
        raise errors.ProgrammerError("Missing input parameter '%s' to"
12943
                                     " IAllocator" % key)
12944
    self._BuildInputData(compat.partial(fn, self), keydata)
12945

    
12946
  def _ComputeClusterData(self):
12947
    """Compute the generic allocator input data.
12948

12949
    This is the data that is independent of the actual operation.
12950

12951
    """
12952
    cfg = self.cfg
12953
    cluster_info = cfg.GetClusterInfo()
12954
    # cluster data
12955
    data = {
12956
      "version": constants.IALLOCATOR_VERSION,
12957
      "cluster_name": cfg.GetClusterName(),
12958
      "cluster_tags": list(cluster_info.GetTags()),
12959
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
12960
      # we don't have job IDs
12961
      }
12962
    ninfo = cfg.GetAllNodesInfo()
12963
    iinfo = cfg.GetAllInstancesInfo().values()
12964
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
12965

    
12966
    # node data
12967
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
12968

    
12969
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
12970
      hypervisor_name = self.hypervisor
12971
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
12972
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
12973
    else:
12974
      hypervisor_name = cluster_info.enabled_hypervisors[0]
12975

    
12976
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
12977
                                        hypervisor_name)
12978
    node_iinfo = \
12979
      self.rpc.call_all_instances_info(node_list,
12980
                                       cluster_info.enabled_hypervisors)
12981

    
12982
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
12983

    
12984
    config_ndata = self._ComputeBasicNodeData(ninfo)
12985
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
12986
                                                 i_list, config_ndata)
12987
    assert len(data["nodes"]) == len(ninfo), \
12988
        "Incomplete node data computed"
12989

    
12990
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
12991

    
12992
    self.in_data = data
12993

    
12994
  @staticmethod
12995
  def _ComputeNodeGroupData(cfg):
12996
    """Compute node groups data.
12997

12998
    """
12999
    ng = dict((guuid, {
13000
      "name": gdata.name,
13001
      "alloc_policy": gdata.alloc_policy,
13002
      })
13003
      for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
13004

    
13005
    return ng
13006

    
13007
  @staticmethod
13008
  def _ComputeBasicNodeData(node_cfg):
13009
    """Compute global node data.
13010

13011
    @rtype: dict
13012
    @returns: a dict of name: (node dict, node config)
13013

13014
    """
13015
    # fill in static (config-based) values
13016
    node_results = dict((ninfo.name, {
13017
      "tags": list(ninfo.GetTags()),
13018
      "primary_ip": ninfo.primary_ip,
13019
      "secondary_ip": ninfo.secondary_ip,
13020
      "offline": ninfo.offline,
13021
      "drained": ninfo.drained,
13022
      "master_candidate": ninfo.master_candidate,
13023
      "group": ninfo.group,
13024
      "master_capable": ninfo.master_capable,
13025
      "vm_capable": ninfo.vm_capable,
13026
      })
13027
      for ninfo in node_cfg.values())
13028

    
13029
    return node_results
13030

    
13031
  @staticmethod
13032
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
13033
                              node_results):
13034
    """Compute global node data.
13035

13036
    @param node_results: the basic node structures as filled from the config
13037

13038
    """
13039
    # make a copy of the current dict
13040
    node_results = dict(node_results)
13041
    for nname, nresult in node_data.items():
13042
      assert nname in node_results, "Missing basic data for node %s" % nname
13043
      ninfo = node_cfg[nname]
13044

    
13045
      if not (ninfo.offline or ninfo.drained):
13046
        nresult.Raise("Can't get data for node %s" % nname)
13047
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
13048
                                nname)
13049
        remote_info = nresult.payload
13050

    
13051
        for attr in ["memory_total", "memory_free", "memory_dom0",
13052
                     "vg_size", "vg_free", "cpu_total"]:
13053
          if attr not in remote_info:
13054
            raise errors.OpExecError("Node '%s' didn't return attribute"
13055
                                     " '%s'" % (nname, attr))
13056
          if not isinstance(remote_info[attr], int):
13057
            raise errors.OpExecError("Node '%s' returned invalid value"
13058
                                     " for '%s': %s" %
13059
                                     (nname, attr, remote_info[attr]))
13060
        # compute memory used by primary instances
13061
        i_p_mem = i_p_up_mem = 0
13062
        for iinfo, beinfo in i_list:
13063
          if iinfo.primary_node == nname:
13064
            i_p_mem += beinfo[constants.BE_MEMORY]
13065
            if iinfo.name not in node_iinfo[nname].payload:
13066
              i_used_mem = 0
13067
            else:
13068
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
13069
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
13070
            remote_info["memory_free"] -= max(0, i_mem_diff)
13071

    
13072
            if iinfo.admin_up:
13073
              i_p_up_mem += beinfo[constants.BE_MEMORY]
13074

    
13075
        # compute memory used by instances
13076
        pnr_dyn = {
13077
          "total_memory": remote_info["memory_total"],
13078
          "reserved_memory": remote_info["memory_dom0"],
13079
          "free_memory": remote_info["memory_free"],
13080
          "total_disk": remote_info["vg_size"],
13081
          "free_disk": remote_info["vg_free"],
13082
          "total_cpus": remote_info["cpu_total"],
13083
          "i_pri_memory": i_p_mem,
13084
          "i_pri_up_memory": i_p_up_mem,
13085
          }
13086
        pnr_dyn.update(node_results[nname])
13087
        node_results[nname] = pnr_dyn
13088

    
13089
    return node_results
13090

    
13091
  @staticmethod
13092
  def _ComputeInstanceData(cluster_info, i_list):
13093
    """Compute global instance data.
13094

13095
    """
13096
    instance_data = {}
13097
    for iinfo, beinfo in i_list:
13098
      nic_data = []
13099
      for nic in iinfo.nics:
13100
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
13101
        nic_dict = {
13102
          "mac": nic.mac,
13103
          "ip": nic.ip,
13104
          "mode": filled_params[constants.NIC_MODE],
13105
          "link": filled_params[constants.NIC_LINK],
13106
          }
13107
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
13108
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
13109
        nic_data.append(nic_dict)
13110
      pir = {
13111
        "tags": list(iinfo.GetTags()),
13112
        "admin_up": iinfo.admin_up,
13113
        "vcpus": beinfo[constants.BE_VCPUS],
13114
        "memory": beinfo[constants.BE_MEMORY],
13115
        "os": iinfo.os,
13116
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
13117
        "nics": nic_data,
13118
        "disks": [{constants.IDISK_SIZE: dsk.size,
13119
                   constants.IDISK_MODE: dsk.mode}
13120
                  for dsk in iinfo.disks],
13121
        "disk_template": iinfo.disk_template,
13122
        "hypervisor": iinfo.hypervisor,
13123
        }
13124
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
13125
                                                 pir["disks"])
13126
      instance_data[iinfo.name] = pir
13127

    
13128
    return instance_data
13129

    
13130
  def _AddNewInstance(self):
13131
    """Add new instance data to allocator structure.
13132

13133
    This in combination with _AllocatorGetClusterData will create the
13134
    correct structure needed as input for the allocator.
13135

13136
    The checks for the completeness of the opcode must have already been
13137
    done.
13138

13139
    """
13140
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
13141

    
13142
    if self.disk_template in constants.DTS_INT_MIRROR:
13143
      self.required_nodes = 2
13144
    else:
13145
      self.required_nodes = 1
13146

    
13147
    request = {
13148
      "name": self.name,
13149
      "disk_template": self.disk_template,
13150
      "tags": self.tags,
13151
      "os": self.os,
13152
      "vcpus": self.vcpus,
13153
      "memory": self.memory,
13154
      "disks": self.disks,
13155
      "disk_space_total": disk_space,
13156
      "nics": self.nics,
13157
      "required_nodes": self.required_nodes,
13158
      "hypervisor": self.hypervisor,
13159
      }
13160

    
13161
    return request
13162

    
13163
  def _AddRelocateInstance(self):
13164
    """Add relocate instance data to allocator structure.
13165

13166
    This in combination with _IAllocatorGetClusterData will create the
13167
    correct structure needed as input for the allocator.
13168

13169
    The checks for the completeness of the opcode must have already been
13170
    done.
13171

13172
    """
13173
    instance = self.cfg.GetInstanceInfo(self.name)
13174
    if instance is None:
13175
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
13176
                                   " IAllocator" % self.name)
13177

    
13178
    if instance.disk_template not in constants.DTS_MIRRORED:
13179
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
13180
                                 errors.ECODE_INVAL)
13181

    
13182
    if instance.disk_template in constants.DTS_INT_MIRROR and \
13183
        len(instance.secondary_nodes) != 1:
13184
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
13185
                                 errors.ECODE_STATE)
13186

    
13187
    self.required_nodes = 1
13188
    disk_sizes = [{constants.IDISK_SIZE: disk.size} for disk in instance.disks]
13189
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
13190

    
13191
    request = {
13192
      "name": self.name,
13193
      "disk_space_total": disk_space,
13194
      "required_nodes": self.required_nodes,
13195
      "relocate_from": self.relocate_from,
13196
      }
13197
    return request
13198

    
13199
  def _AddNodeEvacuate(self):
13200
    """Get data for node-evacuate requests.
13201

13202
    """
13203
    return {
13204
      "instances": self.instances,
13205
      "evac_mode": self.evac_mode,
13206
      }
13207

    
13208
  def _AddChangeGroup(self):
13209
    """Get data for node-evacuate requests.
13210

13211
    """
13212
    return {
13213
      "instances": self.instances,
13214
      "target_groups": self.target_groups,
13215
      }
13216

    
13217
  def _BuildInputData(self, fn, keydata):
13218
    """Build input data structures.
13219

13220
    """
13221
    self._ComputeClusterData()
13222

    
13223
    request = fn()
13224
    request["type"] = self.mode
13225
    for keyname, keytype in keydata:
13226
      if keyname not in request:
13227
        raise errors.ProgrammerError("Request parameter %s is missing" %
13228
                                     keyname)
13229
      val = request[keyname]
13230
      if not keytype(val):
13231
        raise errors.ProgrammerError("Request parameter %s doesn't pass"
13232
                                     " validation, value %s, expected"
13233
                                     " type %s" % (keyname, val, keytype))
13234
    self.in_data["request"] = request
13235

    
13236
    self.in_text = serializer.Dump(self.in_data)
13237

    
13238
  _STRING_LIST = ht.TListOf(ht.TString)
13239
  _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
13240
     # pylint: disable=E1101
13241
     # Class '...' has no 'OP_ID' member
13242
     "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
13243
                          opcodes.OpInstanceMigrate.OP_ID,
13244
                          opcodes.OpInstanceReplaceDisks.OP_ID])
13245
     })))
13246

    
13247
  _NEVAC_MOVED = \
13248
    ht.TListOf(ht.TAnd(ht.TIsLength(3),
13249
                       ht.TItems([ht.TNonEmptyString,
13250
                                  ht.TNonEmptyString,
13251
                                  ht.TListOf(ht.TNonEmptyString),
13252
                                 ])))
13253
  _NEVAC_FAILED = \
13254
    ht.TListOf(ht.TAnd(ht.TIsLength(2),
13255
                       ht.TItems([ht.TNonEmptyString,
13256
                                  ht.TMaybeString,
13257
                                 ])))
13258
  _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
13259
                          ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
13260

    
13261
  _MODE_DATA = {
13262
    constants.IALLOCATOR_MODE_ALLOC:
13263
      (_AddNewInstance,
13264
       [
13265
        ("name", ht.TString),
13266
        ("memory", ht.TInt),
13267
        ("disks", ht.TListOf(ht.TDict)),
13268
        ("disk_template", ht.TString),
13269
        ("os", ht.TString),
13270
        ("tags", _STRING_LIST),
13271
        ("nics", ht.TListOf(ht.TDict)),
13272
        ("vcpus", ht.TInt),
13273
        ("hypervisor", ht.TString),
13274
        ], ht.TList),
13275
    constants.IALLOCATOR_MODE_RELOC:
13276
      (_AddRelocateInstance,
13277
       [("name", ht.TString), ("relocate_from", _STRING_LIST)],
13278
       ht.TList),
13279
     constants.IALLOCATOR_MODE_NODE_EVAC:
13280
      (_AddNodeEvacuate, [
13281
        ("instances", _STRING_LIST),
13282
        ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
13283
        ], _NEVAC_RESULT),
13284
     constants.IALLOCATOR_MODE_CHG_GROUP:
13285
      (_AddChangeGroup, [
13286
        ("instances", _STRING_LIST),
13287
        ("target_groups", _STRING_LIST),
13288
        ], _NEVAC_RESULT),
13289
    }
13290

    
13291
  def Run(self, name, validate=True, call_fn=None):
13292
    """Run an instance allocator and return the results.
13293

13294
    """
13295
    if call_fn is None:
13296
      call_fn = self.rpc.call_iallocator_runner
13297

    
13298
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
13299
    result.Raise("Failure while running the iallocator script")
13300

    
13301
    self.out_text = result.payload
13302
    if validate:
13303
      self._ValidateResult()
13304

    
13305
  def _ValidateResult(self):
13306
    """Process the allocator results.
13307

13308
    This will process and if successful save the result in
13309
    self.out_data and the other parameters.
13310

13311
    """
13312
    try:
13313
      rdict = serializer.Load(self.out_text)
13314
    except Exception, err:
13315
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
13316

    
13317
    if not isinstance(rdict, dict):
13318
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
13319

    
13320
    # TODO: remove backwards compatiblity in later versions
13321
    if "nodes" in rdict and "result" not in rdict:
13322
      rdict["result"] = rdict["nodes"]
13323
      del rdict["nodes"]
13324

    
13325
    for key in "success", "info", "result":
13326
      if key not in rdict:
13327
        raise errors.OpExecError("Can't parse iallocator results:"
13328
                                 " missing key '%s'" % key)
13329
      setattr(self, key, rdict[key])
13330

    
13331
    if not self._result_check(self.result):
13332
      raise errors.OpExecError("Iallocator returned invalid result,"
13333
                               " expected %s, got %s" %
13334
                               (self._result_check, self.result),
13335
                               errors.ECODE_INVAL)
13336

    
13337
    if self.mode == constants.IALLOCATOR_MODE_RELOC:
13338
      assert self.relocate_from is not None
13339
      assert self.required_nodes == 1
13340

    
13341
      node2group = dict((name, ndata["group"])
13342
                        for (name, ndata) in self.in_data["nodes"].items())
13343

    
13344
      fn = compat.partial(self._NodesToGroups, node2group,
13345
                          self.in_data["nodegroups"])
13346

    
13347
      instance = self.cfg.GetInstanceInfo(self.name)
13348
      request_groups = fn(self.relocate_from + [instance.primary_node])
13349
      result_groups = fn(rdict["result"] + [instance.primary_node])
13350

    
13351
      if self.success and not set(result_groups).issubset(request_groups):
13352
        raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
13353
                                 " differ from original groups (%s)" %
13354
                                 (utils.CommaJoin(result_groups),
13355
                                  utils.CommaJoin(request_groups)))
13356

    
13357
    elif self.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
13358
      assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
13359

    
13360
    self.out_data = rdict
13361

    
13362
  @staticmethod
13363
  def _NodesToGroups(node2group, groups, nodes):
13364
    """Returns a list of unique group names for a list of nodes.
13365

13366
    @type node2group: dict
13367
    @param node2group: Map from node name to group UUID
13368
    @type groups: dict
13369
    @param groups: Group information
13370
    @type nodes: list
13371
    @param nodes: Node names
13372

13373
    """
13374
    result = set()
13375

    
13376
    for node in nodes:
13377
      try:
13378
        group_uuid = node2group[node]
13379
      except KeyError:
13380
        # Ignore unknown node
13381
        pass
13382
      else:
13383
        try:
13384
          group = groups[group_uuid]
13385
        except KeyError:
13386
          # Can't find group, let's use UUID
13387
          group_name = group_uuid
13388
        else:
13389
          group_name = group["name"]
13390

    
13391
        result.add(group_name)
13392

    
13393
    return sorted(result)
13394

    
13395

    
13396
class LUTestAllocator(NoHooksLU):
13397
  """Run allocator tests.
13398

13399
  This LU runs the allocator tests
13400

13401
  """
13402
  def CheckPrereq(self):
13403
    """Check prerequisites.
13404

13405
    This checks the opcode parameters depending on the director and mode test.
13406

13407
    """
13408
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
13409
      for attr in ["memory", "disks", "disk_template",
13410
                   "os", "tags", "nics", "vcpus"]:
13411
        if not hasattr(self.op, attr):
13412
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
13413
                                     attr, errors.ECODE_INVAL)
13414
      iname = self.cfg.ExpandInstanceName(self.op.name)
13415
      if iname is not None:
13416
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
13417
                                   iname, errors.ECODE_EXISTS)
13418
      if not isinstance(self.op.nics, list):
13419
        raise errors.OpPrereqError("Invalid parameter 'nics'",
13420
                                   errors.ECODE_INVAL)
13421
      if not isinstance(self.op.disks, list):
13422
        raise errors.OpPrereqError("Invalid parameter 'disks'",
13423
                                   errors.ECODE_INVAL)
13424
      for row in self.op.disks:
13425
        if (not isinstance(row, dict) or
13426
            constants.IDISK_SIZE not in row or
13427
            not isinstance(row[constants.IDISK_SIZE], int) or
13428
            constants.IDISK_MODE not in row or
13429
            row[constants.IDISK_MODE] not in constants.DISK_ACCESS_SET):
13430
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
13431
                                     " parameter", errors.ECODE_INVAL)
13432
      if self.op.hypervisor is None:
13433
        self.op.hypervisor = self.cfg.GetHypervisorType()
13434
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
13435
      fname = _ExpandInstanceName(self.cfg, self.op.name)
13436
      self.op.name = fname
13437
      self.relocate_from = \
13438
          list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
13439
    elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
13440
                          constants.IALLOCATOR_MODE_NODE_EVAC):
13441
      if not self.op.instances:
13442
        raise errors.OpPrereqError("Missing instances", errors.ECODE_INVAL)
13443
      self.op.instances = _GetWantedInstances(self, self.op.instances)
13444
    else:
13445
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
13446
                                 self.op.mode, errors.ECODE_INVAL)
13447

    
13448
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
13449
      if self.op.allocator is None:
13450
        raise errors.OpPrereqError("Missing allocator name",
13451
                                   errors.ECODE_INVAL)
13452
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
13453
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
13454
                                 self.op.direction, errors.ECODE_INVAL)
13455

    
13456
  def Exec(self, feedback_fn):
13457
    """Run the allocator test.
13458

13459
    """
13460
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
13461
      ial = IAllocator(self.cfg, self.rpc,
13462
                       mode=self.op.mode,
13463
                       name=self.op.name,
13464
                       memory=self.op.memory,
13465
                       disks=self.op.disks,
13466
                       disk_template=self.op.disk_template,
13467
                       os=self.op.os,
13468
                       tags=self.op.tags,
13469
                       nics=self.op.nics,
13470
                       vcpus=self.op.vcpus,
13471
                       hypervisor=self.op.hypervisor,
13472
                       )
13473
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
13474
      ial = IAllocator(self.cfg, self.rpc,
13475
                       mode=self.op.mode,
13476
                       name=self.op.name,
13477
                       relocate_from=list(self.relocate_from),
13478
                       )
13479
    elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
13480
      ial = IAllocator(self.cfg, self.rpc,
13481
                       mode=self.op.mode,
13482
                       instances=self.op.instances,
13483
                       target_groups=self.op.target_groups)
13484
    elif self.op.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
13485
      ial = IAllocator(self.cfg, self.rpc,
13486
                       mode=self.op.mode,
13487
                       instances=self.op.instances,
13488
                       evac_mode=self.op.evac_mode)
13489
    else:
13490
      raise errors.ProgrammerError("Uncatched mode %s in"
13491
                                   " LUTestAllocator.Exec", self.op.mode)
13492

    
13493
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
13494
      result = ial.in_text
13495
    else:
13496
      ial.Run(self.op.allocator, validate=False)
13497
      result = ial.out_text
13498
    return result
13499

    
13500

    
13501
#: Query type implementations
13502
_QUERY_IMPL = {
13503
  constants.QR_INSTANCE: _InstanceQuery,
13504
  constants.QR_NODE: _NodeQuery,
13505
  constants.QR_GROUP: _GroupQuery,
13506
  constants.QR_OS: _OsQuery,
13507
  }
13508

    
13509
assert set(_QUERY_IMPL.keys()) == constants.QR_VIA_OP
13510

    
13511

    
13512
def _GetQueryImplementation(name):
13513
  """Returns the implemtnation for a query type.
13514

13515
  @param name: Query type, must be one of L{constants.QR_VIA_OP}
13516

13517
  """
13518
  try:
13519
    return _QUERY_IMPL[name]
13520
  except KeyError:
13521
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
13522
                               errors.ECODE_INVAL)