Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 54b010ca

History | View | Annotate | Download (475.5 kB)

1
#
2
#
3

    
4
# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012 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 to 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 = False
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
    self.needed_locks = dict.fromkeys(locking.LEVELS, locking.ALL_SET)
1576
    self.share_locks = _ShareAll()
1577

    
1578
  def CheckPrereq(self):
1579
    """Check prerequisites.
1580

1581
    """
1582
    # Retrieve all information
1583
    self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
1584
    self.all_node_info = self.cfg.GetAllNodesInfo()
1585
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1586

    
1587
  def Exec(self, feedback_fn):
1588
    """Verify integrity of cluster, performing various test on nodes.
1589

1590
    """
1591
    self.bad = False
1592
    self._feedback_fn = feedback_fn
1593

    
1594
    feedback_fn("* Verifying cluster config")
1595

    
1596
    for msg in self.cfg.VerifyConfig():
1597
      self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
1598

    
1599
    feedback_fn("* Verifying cluster certificate files")
1600

    
1601
    for cert_filename in constants.ALL_CERT_FILES:
1602
      (errcode, msg) = _VerifyCertificate(cert_filename)
1603
      self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
1604

    
1605
    feedback_fn("* Verifying hypervisor parameters")
1606

    
1607
    self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
1608
                                                self.all_inst_info.values()))
1609

    
1610
    feedback_fn("* Verifying all nodes belong to an existing group")
1611

    
1612
    # We do this verification here because, should this bogus circumstance
1613
    # occur, it would never be caught by VerifyGroup, which only acts on
1614
    # nodes/instances reachable from existing node groups.
1615

    
1616
    dangling_nodes = set(node.name for node in self.all_node_info.values()
1617
                         if node.group not in self.all_group_info)
1618

    
1619
    dangling_instances = {}
1620
    no_node_instances = []
1621

    
1622
    for inst in self.all_inst_info.values():
1623
      if inst.primary_node in dangling_nodes:
1624
        dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
1625
      elif inst.primary_node not in self.all_node_info:
1626
        no_node_instances.append(inst.name)
1627

    
1628
    pretty_dangling = [
1629
        "%s (%s)" %
1630
        (node.name,
1631
         utils.CommaJoin(dangling_instances.get(node.name,
1632
                                                ["no instances"])))
1633
        for node in dangling_nodes]
1634

    
1635
    self._ErrorIf(bool(dangling_nodes), self.ECLUSTERDANGLINGNODES, None,
1636
                  "the following nodes (and their instances) belong to a non"
1637
                  " existing group: %s", utils.CommaJoin(pretty_dangling))
1638

    
1639
    self._ErrorIf(bool(no_node_instances), self.ECLUSTERDANGLINGINST, None,
1640
                  "the following instances have a non-existing primary-node:"
1641
                  " %s", utils.CommaJoin(no_node_instances))
1642

    
1643
    return not self.bad
1644

    
1645

    
1646
class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
1647
  """Verifies the status of a node group.
1648

1649
  """
1650
  HPATH = "cluster-verify"
1651
  HTYPE = constants.HTYPE_CLUSTER
1652
  REQ_BGL = False
1653

    
1654
  _HOOKS_INDENT_RE = re.compile("^", re.M)
1655

    
1656
  class NodeImage(object):
1657
    """A class representing the logical and physical status of a node.
1658

1659
    @type name: string
1660
    @ivar name: the node name to which this object refers
1661
    @ivar volumes: a structure as returned from
1662
        L{ganeti.backend.GetVolumeList} (runtime)
1663
    @ivar instances: a list of running instances (runtime)
1664
    @ivar pinst: list of configured primary instances (config)
1665
    @ivar sinst: list of configured secondary instances (config)
1666
    @ivar sbp: dictionary of {primary-node: list of instances} for all
1667
        instances for which this node is secondary (config)
1668
    @ivar mfree: free memory, as reported by hypervisor (runtime)
1669
    @ivar dfree: free disk, as reported by the node (runtime)
1670
    @ivar offline: the offline status (config)
1671
    @type rpc_fail: boolean
1672
    @ivar rpc_fail: whether the RPC verify call was successfull (overall,
1673
        not whether the individual keys were correct) (runtime)
1674
    @type lvm_fail: boolean
1675
    @ivar lvm_fail: whether the RPC call didn't return valid LVM data
1676
    @type hyp_fail: boolean
1677
    @ivar hyp_fail: whether the RPC call didn't return the instance list
1678
    @type ghost: boolean
1679
    @ivar ghost: whether this is a known node or not (config)
1680
    @type os_fail: boolean
1681
    @ivar os_fail: whether the RPC call didn't return valid OS data
1682
    @type oslist: list
1683
    @ivar oslist: list of OSes as diagnosed by DiagnoseOS
1684
    @type vm_capable: boolean
1685
    @ivar vm_capable: whether the node can host instances
1686

1687
    """
1688
    def __init__(self, offline=False, name=None, vm_capable=True):
1689
      self.name = name
1690
      self.volumes = {}
1691
      self.instances = []
1692
      self.pinst = []
1693
      self.sinst = []
1694
      self.sbp = {}
1695
      self.mfree = 0
1696
      self.dfree = 0
1697
      self.offline = offline
1698
      self.vm_capable = vm_capable
1699
      self.rpc_fail = False
1700
      self.lvm_fail = False
1701
      self.hyp_fail = False
1702
      self.ghost = False
1703
      self.os_fail = False
1704
      self.oslist = {}
1705

    
1706
  def ExpandNames(self):
1707
    # This raises errors.OpPrereqError on its own:
1708
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
1709

    
1710
    # Get instances in node group; this is unsafe and needs verification later
1711
    inst_names = \
1712
      self.cfg.GetNodeGroupInstances(self.group_uuid, primary_only=True)
1713

    
1714
    self.needed_locks = {
1715
      locking.LEVEL_INSTANCE: inst_names,
1716
      locking.LEVEL_NODEGROUP: [self.group_uuid],
1717
      locking.LEVEL_NODE: [],
1718
      }
1719

    
1720
    self.share_locks = _ShareAll()
1721

    
1722
  def DeclareLocks(self, level):
1723
    if level == locking.LEVEL_NODE:
1724
      # Get members of node group; this is unsafe and needs verification later
1725
      nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
1726

    
1727
      all_inst_info = self.cfg.GetAllInstancesInfo()
1728

    
1729
      # In Exec(), we warn about mirrored instances that have primary and
1730
      # secondary living in separate node groups. To fully verify that
1731
      # volumes for these instances are healthy, we will need to do an
1732
      # extra call to their secondaries. We ensure here those nodes will
1733
      # be locked.
1734
      for inst in self.owned_locks(locking.LEVEL_INSTANCE):
1735
        # Important: access only the instances whose lock is owned
1736
        if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
1737
          nodes.update(all_inst_info[inst].secondary_nodes)
1738

    
1739
      self.needed_locks[locking.LEVEL_NODE] = nodes
1740

    
1741
  def CheckPrereq(self):
1742
    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
1743
    self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
1744

    
1745
    group_nodes = set(self.group_info.members)
1746
    group_instances = \
1747
      self.cfg.GetNodeGroupInstances(self.group_uuid, primary_only=True)
1748

    
1749
    unlocked_nodes = \
1750
        group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1751

    
1752
    unlocked_instances = \
1753
        group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
1754

    
1755
    if unlocked_nodes:
1756
      raise errors.OpPrereqError("Missing lock for nodes: %s" %
1757
                                 utils.CommaJoin(unlocked_nodes),
1758
                                 errors.ECODE_STATE)
1759

    
1760
    if unlocked_instances:
1761
      raise errors.OpPrereqError("Missing lock for instances: %s" %
1762
                                 utils.CommaJoin(unlocked_instances),
1763
                                 errors.ECODE_STATE)
1764

    
1765
    self.all_node_info = self.cfg.GetAllNodesInfo()
1766
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1767

    
1768
    self.my_node_names = utils.NiceSort(group_nodes)
1769
    self.my_inst_names = utils.NiceSort(group_instances)
1770

    
1771
    self.my_node_info = dict((name, self.all_node_info[name])
1772
                             for name in self.my_node_names)
1773

    
1774
    self.my_inst_info = dict((name, self.all_inst_info[name])
1775
                             for name in self.my_inst_names)
1776

    
1777
    # We detect here the nodes that will need the extra RPC calls for verifying
1778
    # split LV volumes; they should be locked.
1779
    extra_lv_nodes = set()
1780

    
1781
    for inst in self.my_inst_info.values():
1782
      if inst.disk_template in constants.DTS_INT_MIRROR:
1783
        for nname in inst.all_nodes:
1784
          if self.all_node_info[nname].group != self.group_uuid:
1785
            extra_lv_nodes.add(nname)
1786

    
1787
    unlocked_lv_nodes = \
1788
        extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1789

    
1790
    if unlocked_lv_nodes:
1791
      raise errors.OpPrereqError("Missing node locks for LV check: %s" %
1792
                                 utils.CommaJoin(unlocked_lv_nodes),
1793
                                 errors.ECODE_STATE)
1794
    self.extra_lv_nodes = list(extra_lv_nodes)
1795

    
1796
  def _VerifyNode(self, ninfo, nresult):
1797
    """Perform some basic validation on data returned from a node.
1798

1799
      - check the result data structure is well formed and has all the
1800
        mandatory fields
1801
      - check ganeti version
1802

1803
    @type ninfo: L{objects.Node}
1804
    @param ninfo: the node to check
1805
    @param nresult: the results from the node
1806
    @rtype: boolean
1807
    @return: whether overall this call was successful (and we can expect
1808
         reasonable values in the respose)
1809

1810
    """
1811
    node = ninfo.name
1812
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1813

    
1814
    # main result, nresult should be a non-empty dict
1815
    test = not nresult or not isinstance(nresult, dict)
1816
    _ErrorIf(test, self.ENODERPC, node,
1817
                  "unable to verify node: no data returned")
1818
    if test:
1819
      return False
1820

    
1821
    # compares ganeti version
1822
    local_version = constants.PROTOCOL_VERSION
1823
    remote_version = nresult.get("version", None)
1824
    test = not (remote_version and
1825
                isinstance(remote_version, (list, tuple)) and
1826
                len(remote_version) == 2)
1827
    _ErrorIf(test, self.ENODERPC, node,
1828
             "connection to node returned invalid data")
1829
    if test:
1830
      return False
1831

    
1832
    test = local_version != remote_version[0]
1833
    _ErrorIf(test, self.ENODEVERSION, node,
1834
             "incompatible protocol versions: master %s,"
1835
             " node %s", local_version, remote_version[0])
1836
    if test:
1837
      return False
1838

    
1839
    # node seems compatible, we can actually try to look into its results
1840

    
1841
    # full package version
1842
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1843
                  self.ENODEVERSION, node,
1844
                  "software version mismatch: master %s, node %s",
1845
                  constants.RELEASE_VERSION, remote_version[1],
1846
                  code=self.ETYPE_WARNING)
1847

    
1848
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1849
    if ninfo.vm_capable and isinstance(hyp_result, dict):
1850
      for hv_name, hv_result in hyp_result.iteritems():
1851
        test = hv_result is not None
1852
        _ErrorIf(test, self.ENODEHV, node,
1853
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1854

    
1855
    hvp_result = nresult.get(constants.NV_HVPARAMS, None)
1856
    if ninfo.vm_capable and isinstance(hvp_result, list):
1857
      for item, hv_name, hv_result in hvp_result:
1858
        _ErrorIf(True, self.ENODEHV, node,
1859
                 "hypervisor %s parameter verify failure (source %s): %s",
1860
                 hv_name, item, hv_result)
1861

    
1862
    test = nresult.get(constants.NV_NODESETUP,
1863
                       ["Missing NODESETUP results"])
1864
    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
1865
             "; ".join(test))
1866

    
1867
    return True
1868

    
1869
  def _VerifyNodeTime(self, ninfo, nresult,
1870
                      nvinfo_starttime, nvinfo_endtime):
1871
    """Check the node time.
1872

1873
    @type ninfo: L{objects.Node}
1874
    @param ninfo: the node to check
1875
    @param nresult: the remote results for the node
1876
    @param nvinfo_starttime: the start time of the RPC call
1877
    @param nvinfo_endtime: the end time of the RPC call
1878

1879
    """
1880
    node = ninfo.name
1881
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1882

    
1883
    ntime = nresult.get(constants.NV_TIME, None)
1884
    try:
1885
      ntime_merged = utils.MergeTime(ntime)
1886
    except (ValueError, TypeError):
1887
      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
1888
      return
1889

    
1890
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1891
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1892
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1893
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1894
    else:
1895
      ntime_diff = None
1896

    
1897
    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
1898
             "Node time diverges by at least %s from master node time",
1899
             ntime_diff)
1900

    
1901
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1902
    """Check the node LVM results.
1903

1904
    @type ninfo: L{objects.Node}
1905
    @param ninfo: the node to check
1906
    @param nresult: the remote results for the node
1907
    @param vg_name: the configured VG name
1908

1909
    """
1910
    if vg_name is None:
1911
      return
1912

    
1913
    node = ninfo.name
1914
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1915

    
1916
    # checks vg existence and size > 20G
1917
    vglist = nresult.get(constants.NV_VGLIST, None)
1918
    test = not vglist
1919
    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
1920
    if not test:
1921
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
1922
                                            constants.MIN_VG_SIZE)
1923
      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
1924

    
1925
    # check pv names
1926
    pvlist = nresult.get(constants.NV_PVLIST, None)
1927
    test = pvlist is None
1928
    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
1929
    if not test:
1930
      # check that ':' is not present in PV names, since it's a
1931
      # special character for lvcreate (denotes the range of PEs to
1932
      # use on the PV)
1933
      for _, pvname, owner_vg in pvlist:
1934
        test = ":" in pvname
1935
        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
1936
                 " '%s' of VG '%s'", pvname, owner_vg)
1937

    
1938
  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
1939
    """Check the node bridges.
1940

1941
    @type ninfo: L{objects.Node}
1942
    @param ninfo: the node to check
1943
    @param nresult: the remote results for the node
1944
    @param bridges: the expected list of bridges
1945

1946
    """
1947
    if not bridges:
1948
      return
1949

    
1950
    node = ninfo.name
1951
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1952

    
1953
    missing = nresult.get(constants.NV_BRIDGES, None)
1954
    test = not isinstance(missing, list)
1955
    _ErrorIf(test, self.ENODENET, node,
1956
             "did not return valid bridge information")
1957
    if not test:
1958
      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
1959
               utils.CommaJoin(sorted(missing)))
1960

    
1961
  def _VerifyNodeNetwork(self, ninfo, nresult):
1962
    """Check the node network connectivity results.
1963

1964
    @type ninfo: L{objects.Node}
1965
    @param ninfo: the node to check
1966
    @param nresult: the remote results for the node
1967

1968
    """
1969
    node = ninfo.name
1970
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1971

    
1972
    test = constants.NV_NODELIST not in nresult
1973
    _ErrorIf(test, self.ENODESSH, node,
1974
             "node hasn't returned node ssh connectivity data")
1975
    if not test:
1976
      if nresult[constants.NV_NODELIST]:
1977
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1978
          _ErrorIf(True, self.ENODESSH, node,
1979
                   "ssh communication with node '%s': %s", a_node, a_msg)
1980

    
1981
    test = constants.NV_NODENETTEST not in nresult
1982
    _ErrorIf(test, self.ENODENET, node,
1983
             "node hasn't returned node tcp connectivity data")
1984
    if not test:
1985
      if nresult[constants.NV_NODENETTEST]:
1986
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1987
        for anode in nlist:
1988
          _ErrorIf(True, self.ENODENET, node,
1989
                   "tcp communication with node '%s': %s",
1990
                   anode, nresult[constants.NV_NODENETTEST][anode])
1991

    
1992
    test = constants.NV_MASTERIP not in nresult
1993
    _ErrorIf(test, self.ENODENET, node,
1994
             "node hasn't returned node master IP reachability data")
1995
    if not test:
1996
      if not nresult[constants.NV_MASTERIP]:
1997
        if node == self.master_node:
1998
          msg = "the master node cannot reach the master IP (not configured?)"
1999
        else:
2000
          msg = "cannot reach the master IP"
2001
        _ErrorIf(True, self.ENODENET, node, msg)
2002

    
2003
  def _VerifyInstance(self, instance, instanceconfig, node_image,
2004
                      diskstatus):
2005
    """Verify an instance.
2006

2007
    This function checks to see if the required block devices are
2008
    available on the instance's node.
2009

2010
    """
2011
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2012
    node_current = instanceconfig.primary_node
2013

    
2014
    node_vol_should = {}
2015
    instanceconfig.MapLVsByNode(node_vol_should)
2016

    
2017
    for node in node_vol_should:
2018
      n_img = node_image[node]
2019
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2020
        # ignore missing volumes on offline or broken nodes
2021
        continue
2022
      for volume in node_vol_should[node]:
2023
        test = volume not in n_img.volumes
2024
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
2025
                 "volume %s missing on node %s", volume, node)
2026

    
2027
    if instanceconfig.admin_up:
2028
      pri_img = node_image[node_current]
2029
      test = instance not in pri_img.instances and not pri_img.offline
2030
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
2031
               "instance not running on its primary node %s",
2032
               node_current)
2033

    
2034
    diskdata = [(nname, success, status, idx)
2035
                for (nname, disks) in diskstatus.items()
2036
                for idx, (success, status) in enumerate(disks)]
2037

    
2038
    for nname, success, bdev_status, idx in diskdata:
2039
      # the 'ghost node' construction in Exec() ensures that we have a
2040
      # node here
2041
      snode = node_image[nname]
2042
      bad_snode = snode.ghost or snode.offline
2043
      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
2044
               self.EINSTANCEFAULTYDISK, instance,
2045
               "couldn't retrieve status for disk/%s on %s: %s",
2046
               idx, nname, bdev_status)
2047
      _ErrorIf((instanceconfig.admin_up and success and
2048
                bdev_status.ldisk_status == constants.LDS_FAULTY),
2049
               self.EINSTANCEFAULTYDISK, instance,
2050
               "disk/%s on %s is faulty", idx, nname)
2051

    
2052
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
2053
    """Verify if there are any unknown volumes in the cluster.
2054

2055
    The .os, .swap and backup volumes are ignored. All other volumes are
2056
    reported as unknown.
2057

2058
    @type reserved: L{ganeti.utils.FieldSet}
2059
    @param reserved: a FieldSet of reserved volume names
2060

2061
    """
2062
    for node, n_img in node_image.items():
2063
      if (n_img.offline or n_img.rpc_fail or n_img.lvm_fail or
2064
          self.all_node_info[node].group != self.group_uuid):
2065
        # skip non-healthy nodes
2066
        continue
2067
      for volume in n_img.volumes:
2068
        test = ((node not in node_vol_should or
2069
                volume not in node_vol_should[node]) and
2070
                not reserved.Matches(volume))
2071
        self._ErrorIf(test, self.ENODEORPHANLV, node,
2072
                      "volume %s is unknown", volume)
2073

    
2074
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
2075
    """Verify N+1 Memory Resilience.
2076

2077
    Check that if one single node dies we can still start all the
2078
    instances it was primary for.
2079

2080
    """
2081
    cluster_info = self.cfg.GetClusterInfo()
2082
    for node, n_img in node_image.items():
2083
      # This code checks that every node which is now listed as
2084
      # secondary has enough memory to host all instances it is
2085
      # supposed to should a single other node in the cluster fail.
2086
      # FIXME: not ready for failover to an arbitrary node
2087
      # FIXME: does not support file-backed instances
2088
      # WARNING: we currently take into account down instances as well
2089
      # as up ones, considering that even if they're down someone
2090
      # might want to start them even in the event of a node failure.
2091
      if n_img.offline or self.all_node_info[node].group != self.group_uuid:
2092
        # we're skipping nodes marked offline and nodes in other groups from
2093
        # the N+1 warning, since most likely we don't have good memory
2094
        # infromation from them; we already list instances living on such
2095
        # nodes, and that's enough warning
2096
        continue
2097
      for prinode, instances in n_img.sbp.items():
2098
        needed_mem = 0
2099
        for instance in instances:
2100
          bep = cluster_info.FillBE(instance_cfg[instance])
2101
          if bep[constants.BE_AUTO_BALANCE]:
2102
            needed_mem += bep[constants.BE_MEMORY]
2103
        test = n_img.mfree < needed_mem
2104
        self._ErrorIf(test, self.ENODEN1, node,
2105
                      "not enough memory to accomodate instance failovers"
2106
                      " should node %s fail (%dMiB needed, %dMiB available)",
2107
                      prinode, needed_mem, n_img.mfree)
2108

    
2109
  @classmethod
2110
  def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
2111
                   (files_all, files_all_opt, files_mc, files_vm)):
2112
    """Verifies file checksums collected from all nodes.
2113

2114
    @param errorif: Callback for reporting errors
2115
    @param nodeinfo: List of L{objects.Node} objects
2116
    @param master_node: Name of master node
2117
    @param all_nvinfo: RPC results
2118

2119
    """
2120
    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
2121
            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
2122
           "Found file listed in more than one file list"
2123

    
2124
    # Define functions determining which nodes to consider for a file
2125
    files2nodefn = [
2126
      (files_all, None),
2127
      (files_all_opt, None),
2128
      (files_mc, lambda node: (node.master_candidate or
2129
                               node.name == master_node)),
2130
      (files_vm, lambda node: node.vm_capable),
2131
      ]
2132

    
2133
    # Build mapping from filename to list of nodes which should have the file
2134
    nodefiles = {}
2135
    for (files, fn) in files2nodefn:
2136
      if fn is None:
2137
        filenodes = nodeinfo
2138
      else:
2139
        filenodes = filter(fn, nodeinfo)
2140
      nodefiles.update((filename,
2141
                        frozenset(map(operator.attrgetter("name"), filenodes)))
2142
                       for filename in files)
2143

    
2144
    assert set(nodefiles) == (files_all | files_all_opt | files_mc | files_vm)
2145

    
2146
    fileinfo = dict((filename, {}) for filename in nodefiles)
2147
    ignore_nodes = set()
2148

    
2149
    for node in nodeinfo:
2150
      if node.offline:
2151
        ignore_nodes.add(node.name)
2152
        continue
2153

    
2154
      nresult = all_nvinfo[node.name]
2155

    
2156
      if nresult.fail_msg or not nresult.payload:
2157
        node_files = None
2158
      else:
2159
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
2160

    
2161
      test = not (node_files and isinstance(node_files, dict))
2162
      errorif(test, cls.ENODEFILECHECK, node.name,
2163
              "Node did not return file checksum data")
2164
      if test:
2165
        ignore_nodes.add(node.name)
2166
        continue
2167

    
2168
      # Build per-checksum mapping from filename to nodes having it
2169
      for (filename, checksum) in node_files.items():
2170
        assert filename in nodefiles
2171
        fileinfo[filename].setdefault(checksum, set()).add(node.name)
2172

    
2173
    for (filename, checksums) in fileinfo.items():
2174
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2175

    
2176
      # Nodes having the file
2177
      with_file = frozenset(node_name
2178
                            for nodes in fileinfo[filename].values()
2179
                            for node_name in nodes) - ignore_nodes
2180

    
2181
      expected_nodes = nodefiles[filename] - ignore_nodes
2182

    
2183
      # Nodes missing file
2184
      missing_file = expected_nodes - with_file
2185

    
2186
      if filename in files_all_opt:
2187
        # All or no nodes
2188
        errorif(missing_file and missing_file != expected_nodes,
2189
                cls.ECLUSTERFILECHECK, None,
2190
                "File %s is optional, but it must exist on all or no"
2191
                " nodes (not found on %s)",
2192
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2193
      else:
2194
        # Non-optional files
2195
        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
2196
                "File %s is missing from node(s) %s", filename,
2197
                utils.CommaJoin(utils.NiceSort(missing_file)))
2198

    
2199
        # Warn if a node has a file it shouldn't
2200
        unexpected = with_file - expected_nodes
2201
        errorif(unexpected,
2202
                cls.ECLUSTERFILECHECK, None,
2203
                "File %s should not exist on node(s) %s",
2204
                filename, utils.CommaJoin(utils.NiceSort(unexpected)))
2205

    
2206
      # See if there are multiple versions of the file
2207
      test = len(checksums) > 1
2208
      if test:
2209
        variants = ["variant %s on %s" %
2210
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2211
                    for (idx, (checksum, nodes)) in
2212
                      enumerate(sorted(checksums.items()))]
2213
      else:
2214
        variants = []
2215

    
2216
      errorif(test, cls.ECLUSTERFILECHECK, None,
2217
              "File %s found with %s different checksums (%s)",
2218
              filename, len(checksums), "; ".join(variants))
2219

    
2220
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2221
                      drbd_map):
2222
    """Verifies and the node DRBD status.
2223

2224
    @type ninfo: L{objects.Node}
2225
    @param ninfo: the node to check
2226
    @param nresult: the remote results for the node
2227
    @param instanceinfo: the dict of instances
2228
    @param drbd_helper: the configured DRBD usermode helper
2229
    @param drbd_map: the DRBD map as returned by
2230
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2231

2232
    """
2233
    node = ninfo.name
2234
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2235

    
2236
    if drbd_helper:
2237
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2238
      test = (helper_result == None)
2239
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
2240
               "no drbd usermode helper returned")
2241
      if helper_result:
2242
        status, payload = helper_result
2243
        test = not status
2244
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2245
                 "drbd usermode helper check unsuccessful: %s", payload)
2246
        test = status and (payload != drbd_helper)
2247
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2248
                 "wrong drbd usermode helper: %s", payload)
2249

    
2250
    # compute the DRBD minors
2251
    node_drbd = {}
2252
    for minor, instance in drbd_map[node].items():
2253
      test = instance not in instanceinfo
2254
      _ErrorIf(test, self.ECLUSTERCFG, None,
2255
               "ghost instance '%s' in temporary DRBD map", instance)
2256
        # ghost instance should not be running, but otherwise we
2257
        # don't give double warnings (both ghost instance and
2258
        # unallocated minor in use)
2259
      if test:
2260
        node_drbd[minor] = (instance, False)
2261
      else:
2262
        instance = instanceinfo[instance]
2263
        node_drbd[minor] = (instance.name, instance.admin_up)
2264

    
2265
    # and now check them
2266
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2267
    test = not isinstance(used_minors, (tuple, list))
2268
    _ErrorIf(test, self.ENODEDRBD, node,
2269
             "cannot parse drbd status file: %s", str(used_minors))
2270
    if test:
2271
      # we cannot check drbd status
2272
      return
2273

    
2274
    for minor, (iname, must_exist) in node_drbd.items():
2275
      test = minor not in used_minors and must_exist
2276
      _ErrorIf(test, self.ENODEDRBD, node,
2277
               "drbd minor %d of instance %s is not active", minor, iname)
2278
    for minor in used_minors:
2279
      test = minor not in node_drbd
2280
      _ErrorIf(test, self.ENODEDRBD, node,
2281
               "unallocated drbd minor %d is in use", minor)
2282

    
2283
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2284
    """Builds the node OS structures.
2285

2286
    @type ninfo: L{objects.Node}
2287
    @param ninfo: the node to check
2288
    @param nresult: the remote results for the node
2289
    @param nimg: the node image object
2290

2291
    """
2292
    node = ninfo.name
2293
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2294

    
2295
    remote_os = nresult.get(constants.NV_OSLIST, None)
2296
    test = (not isinstance(remote_os, list) or
2297
            not compat.all(isinstance(v, list) and len(v) == 7
2298
                           for v in remote_os))
2299

    
2300
    _ErrorIf(test, self.ENODEOS, node,
2301
             "node hasn't returned valid OS data")
2302

    
2303
    nimg.os_fail = test
2304

    
2305
    if test:
2306
      return
2307

    
2308
    os_dict = {}
2309

    
2310
    for (name, os_path, status, diagnose,
2311
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2312

    
2313
      if name not in os_dict:
2314
        os_dict[name] = []
2315

    
2316
      # parameters is a list of lists instead of list of tuples due to
2317
      # JSON lacking a real tuple type, fix it:
2318
      parameters = [tuple(v) for v in parameters]
2319
      os_dict[name].append((os_path, status, diagnose,
2320
                            set(variants), set(parameters), set(api_ver)))
2321

    
2322
    nimg.oslist = os_dict
2323

    
2324
  def _VerifyNodeOS(self, ninfo, nimg, base):
2325
    """Verifies the node OS list.
2326

2327
    @type ninfo: L{objects.Node}
2328
    @param ninfo: the node to check
2329
    @param nimg: the node image object
2330
    @param base: the 'template' node we match against (e.g. from the master)
2331

2332
    """
2333
    node = ninfo.name
2334
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2335

    
2336
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2337

    
2338
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2339
    for os_name, os_data in nimg.oslist.items():
2340
      assert os_data, "Empty OS status for OS %s?!" % os_name
2341
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2342
      _ErrorIf(not f_status, self.ENODEOS, node,
2343
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2344
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
2345
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2346
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2347
      # comparisons with the 'base' image
2348
      test = os_name not in base.oslist
2349
      _ErrorIf(test, self.ENODEOS, node,
2350
               "Extra OS %s not present on reference node (%s)",
2351
               os_name, base.name)
2352
      if test:
2353
        continue
2354
      assert base.oslist[os_name], "Base node has empty OS status?"
2355
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2356
      if not b_status:
2357
        # base OS is invalid, skipping
2358
        continue
2359
      for kind, a, b in [("API version", f_api, b_api),
2360
                         ("variants list", f_var, b_var),
2361
                         ("parameters", beautify_params(f_param),
2362
                          beautify_params(b_param))]:
2363
        _ErrorIf(a != b, self.ENODEOS, node,
2364
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2365
                 kind, os_name, base.name,
2366
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2367

    
2368
    # check any missing OSes
2369
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2370
    _ErrorIf(missing, self.ENODEOS, node,
2371
             "OSes present on reference node %s but missing on this node: %s",
2372
             base.name, utils.CommaJoin(missing))
2373

    
2374
  def _VerifyOob(self, ninfo, nresult):
2375
    """Verifies out of band functionality of a node.
2376

2377
    @type ninfo: L{objects.Node}
2378
    @param ninfo: the node to check
2379
    @param nresult: the remote results for the node
2380

2381
    """
2382
    node = ninfo.name
2383
    # We just have to verify the paths on master and/or master candidates
2384
    # as the oob helper is invoked on the master
2385
    if ((ninfo.master_candidate or ninfo.master_capable) and
2386
        constants.NV_OOB_PATHS in nresult):
2387
      for path_result in nresult[constants.NV_OOB_PATHS]:
2388
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
2389

    
2390
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2391
    """Verifies and updates the node volume data.
2392

2393
    This function will update a L{NodeImage}'s internal structures
2394
    with data from the remote call.
2395

2396
    @type ninfo: L{objects.Node}
2397
    @param ninfo: the node to check
2398
    @param nresult: the remote results for the node
2399
    @param nimg: the node image object
2400
    @param vg_name: the configured VG name
2401

2402
    """
2403
    node = ninfo.name
2404
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2405

    
2406
    nimg.lvm_fail = True
2407
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2408
    if vg_name is None:
2409
      pass
2410
    elif isinstance(lvdata, basestring):
2411
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
2412
               utils.SafeEncode(lvdata))
2413
    elif not isinstance(lvdata, dict):
2414
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
2415
    else:
2416
      nimg.volumes = lvdata
2417
      nimg.lvm_fail = False
2418

    
2419
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2420
    """Verifies and updates the node instance list.
2421

2422
    If the listing was successful, then updates this node's instance
2423
    list. Otherwise, it marks the RPC call as failed for the instance
2424
    list key.
2425

2426
    @type ninfo: L{objects.Node}
2427
    @param ninfo: the node to check
2428
    @param nresult: the remote results for the node
2429
    @param nimg: the node image object
2430

2431
    """
2432
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2433
    test = not isinstance(idata, list)
2434
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
2435
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
2436
    if test:
2437
      nimg.hyp_fail = True
2438
    else:
2439
      nimg.instances = idata
2440

    
2441
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2442
    """Verifies and computes a node information map
2443

2444
    @type ninfo: L{objects.Node}
2445
    @param ninfo: the node to check
2446
    @param nresult: the remote results for the node
2447
    @param nimg: the node image object
2448
    @param vg_name: the configured VG name
2449

2450
    """
2451
    node = ninfo.name
2452
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2453

    
2454
    # try to read free memory (from the hypervisor)
2455
    hv_info = nresult.get(constants.NV_HVINFO, None)
2456
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2457
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
2458
    if not test:
2459
      try:
2460
        nimg.mfree = int(hv_info["memory_free"])
2461
      except (ValueError, TypeError):
2462
        _ErrorIf(True, self.ENODERPC, node,
2463
                 "node returned invalid nodeinfo, check hypervisor")
2464

    
2465
    # FIXME: devise a free space model for file based instances as well
2466
    if vg_name is not None:
2467
      test = (constants.NV_VGLIST not in nresult or
2468
              vg_name not in nresult[constants.NV_VGLIST])
2469
      _ErrorIf(test, self.ENODELVM, node,
2470
               "node didn't return data for the volume group '%s'"
2471
               " - it is either missing or broken", vg_name)
2472
      if not test:
2473
        try:
2474
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2475
        except (ValueError, TypeError):
2476
          _ErrorIf(True, self.ENODERPC, node,
2477
                   "node returned invalid LVM info, check LVM status")
2478

    
2479
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2480
    """Gets per-disk status information for all instances.
2481

2482
    @type nodelist: list of strings
2483
    @param nodelist: Node names
2484
    @type node_image: dict of (name, L{objects.Node})
2485
    @param node_image: Node objects
2486
    @type instanceinfo: dict of (name, L{objects.Instance})
2487
    @param instanceinfo: Instance objects
2488
    @rtype: {instance: {node: [(succes, payload)]}}
2489
    @return: a dictionary of per-instance dictionaries with nodes as
2490
        keys and disk information as values; the disk information is a
2491
        list of tuples (success, payload)
2492

2493
    """
2494
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2495

    
2496
    node_disks = {}
2497
    node_disks_devonly = {}
2498
    diskless_instances = set()
2499
    diskless = constants.DT_DISKLESS
2500

    
2501
    for nname in nodelist:
2502
      node_instances = list(itertools.chain(node_image[nname].pinst,
2503
                                            node_image[nname].sinst))
2504
      diskless_instances.update(inst for inst in node_instances
2505
                                if instanceinfo[inst].disk_template == diskless)
2506
      disks = [(inst, disk)
2507
               for inst in node_instances
2508
               for disk in instanceinfo[inst].disks]
2509

    
2510
      if not disks:
2511
        # No need to collect data
2512
        continue
2513

    
2514
      node_disks[nname] = disks
2515

    
2516
      # Creating copies as SetDiskID below will modify the objects and that can
2517
      # lead to incorrect data returned from nodes
2518
      devonly = [dev.Copy() for (_, dev) in disks]
2519

    
2520
      for dev in devonly:
2521
        self.cfg.SetDiskID(dev, nname)
2522

    
2523
      node_disks_devonly[nname] = devonly
2524

    
2525
    assert len(node_disks) == len(node_disks_devonly)
2526

    
2527
    # Collect data from all nodes with disks
2528
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2529
                                                          node_disks_devonly)
2530

    
2531
    assert len(result) == len(node_disks)
2532

    
2533
    instdisk = {}
2534

    
2535
    for (nname, nres) in result.items():
2536
      disks = node_disks[nname]
2537

    
2538
      if nres.offline:
2539
        # No data from this node
2540
        data = len(disks) * [(False, "node offline")]
2541
      else:
2542
        msg = nres.fail_msg
2543
        _ErrorIf(msg, self.ENODERPC, nname,
2544
                 "while getting disk information: %s", msg)
2545
        if msg:
2546
          # No data from this node
2547
          data = len(disks) * [(False, msg)]
2548
        else:
2549
          data = []
2550
          for idx, i in enumerate(nres.payload):
2551
            if isinstance(i, (tuple, list)) and len(i) == 2:
2552
              data.append(i)
2553
            else:
2554
              logging.warning("Invalid result from node %s, entry %d: %s",
2555
                              nname, idx, i)
2556
              data.append((False, "Invalid result from the remote node"))
2557

    
2558
      for ((inst, _), status) in zip(disks, data):
2559
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2560

    
2561
    # Add empty entries for diskless instances.
2562
    for inst in diskless_instances:
2563
      assert inst not in instdisk
2564
      instdisk[inst] = {}
2565

    
2566
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2567
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2568
                      compat.all(isinstance(s, (tuple, list)) and
2569
                                 len(s) == 2 for s in statuses)
2570
                      for inst, nnames in instdisk.items()
2571
                      for nname, statuses in nnames.items())
2572
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2573

    
2574
    return instdisk
2575

    
2576
  @staticmethod
2577
  def _SshNodeSelector(group_uuid, all_nodes):
2578
    """Create endless iterators for all potential SSH check hosts.
2579

2580
    """
2581
    nodes = [node for node in all_nodes
2582
             if (node.group != group_uuid and
2583
                 not node.offline)]
2584
    keyfunc = operator.attrgetter("group")
2585

    
2586
    return map(itertools.cycle,
2587
               [sorted(map(operator.attrgetter("name"), names))
2588
                for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
2589
                                                  keyfunc)])
2590

    
2591
  @classmethod
2592
  def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
2593
    """Choose which nodes should talk to which other nodes.
2594

2595
    We will make nodes contact all nodes in their group, and one node from
2596
    every other group.
2597

2598
    @warning: This algorithm has a known issue if one node group is much
2599
      smaller than others (e.g. just one node). In such a case all other
2600
      nodes will talk to the single node.
2601

2602
    """
2603
    online_nodes = sorted(node.name for node in group_nodes if not node.offline)
2604
    sel = cls._SshNodeSelector(group_uuid, all_nodes)
2605

    
2606
    return (online_nodes,
2607
            dict((name, sorted([i.next() for i in sel]))
2608
                 for name in online_nodes))
2609

    
2610
  def BuildHooksEnv(self):
2611
    """Build hooks env.
2612

2613
    Cluster-Verify hooks just ran in the post phase and their failure makes
2614
    the output be logged in the verify output and the verification to fail.
2615

2616
    """
2617
    env = {
2618
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2619
      }
2620

    
2621
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2622
               for node in self.my_node_info.values())
2623

    
2624
    return env
2625

    
2626
  def BuildHooksNodes(self):
2627
    """Build hooks nodes.
2628

2629
    """
2630
    return ([], self.my_node_names)
2631

    
2632
  def Exec(self, feedback_fn):
2633
    """Verify integrity of the node group, performing various test on nodes.
2634

2635
    """
2636
    # This method has too many local variables. pylint: disable=R0914
2637
    feedback_fn("* Verifying group '%s'" % self.group_info.name)
2638

    
2639
    if not self.my_node_names:
2640
      # empty node group
2641
      feedback_fn("* Empty node group, skipping verification")
2642
      return True
2643

    
2644
    self.bad = False
2645
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2646
    verbose = self.op.verbose
2647
    self._feedback_fn = feedback_fn
2648

    
2649
    vg_name = self.cfg.GetVGName()
2650
    drbd_helper = self.cfg.GetDRBDHelper()
2651
    cluster = self.cfg.GetClusterInfo()
2652
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2653
    hypervisors = cluster.enabled_hypervisors
2654
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2655

    
2656
    i_non_redundant = [] # Non redundant instances
2657
    i_non_a_balanced = [] # Non auto-balanced instances
2658
    n_offline = 0 # Count of offline nodes
2659
    n_drained = 0 # Count of nodes being drained
2660
    node_vol_should = {}
2661

    
2662
    # FIXME: verify OS list
2663

    
2664
    # File verification
2665
    filemap = _ComputeAncillaryFiles(cluster, False)
2666

    
2667
    # do local checksums
2668
    master_node = self.master_node = self.cfg.GetMasterNode()
2669
    master_ip = self.cfg.GetMasterIP()
2670

    
2671
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2672

    
2673
    node_verify_param = {
2674
      constants.NV_FILELIST:
2675
        utils.UniqueSequence(filename
2676
                             for files in filemap
2677
                             for filename in files),
2678
      constants.NV_NODELIST:
2679
        self._SelectSshCheckNodes(node_data_list, self.group_uuid,
2680
                                  self.all_node_info.values()),
2681
      constants.NV_HYPERVISOR: hypervisors,
2682
      constants.NV_HVPARAMS:
2683
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2684
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2685
                                 for node in node_data_list
2686
                                 if not node.offline],
2687
      constants.NV_INSTANCELIST: hypervisors,
2688
      constants.NV_VERSION: None,
2689
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2690
      constants.NV_NODESETUP: None,
2691
      constants.NV_TIME: None,
2692
      constants.NV_MASTERIP: (master_node, master_ip),
2693
      constants.NV_OSLIST: None,
2694
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2695
      }
2696

    
2697
    if vg_name is not None:
2698
      node_verify_param[constants.NV_VGLIST] = None
2699
      node_verify_param[constants.NV_LVLIST] = vg_name
2700
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2701
      node_verify_param[constants.NV_DRBDLIST] = None
2702

    
2703
    if drbd_helper:
2704
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2705

    
2706
    # bridge checks
2707
    # FIXME: this needs to be changed per node-group, not cluster-wide
2708
    bridges = set()
2709
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2710
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2711
      bridges.add(default_nicpp[constants.NIC_LINK])
2712
    for instance in self.my_inst_info.values():
2713
      for nic in instance.nics:
2714
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2715
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2716
          bridges.add(full_nic[constants.NIC_LINK])
2717

    
2718
    if bridges:
2719
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2720

    
2721
    # Build our expected cluster state
2722
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2723
                                                 name=node.name,
2724
                                                 vm_capable=node.vm_capable))
2725
                      for node in node_data_list)
2726

    
2727
    # Gather OOB paths
2728
    oob_paths = []
2729
    for node in self.all_node_info.values():
2730
      path = _SupportsOob(self.cfg, node)
2731
      if path and path not in oob_paths:
2732
        oob_paths.append(path)
2733

    
2734
    if oob_paths:
2735
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2736

    
2737
    for instance in self.my_inst_names:
2738
      inst_config = self.my_inst_info[instance]
2739

    
2740
      for nname in inst_config.all_nodes:
2741
        if nname not in node_image:
2742
          gnode = self.NodeImage(name=nname)
2743
          gnode.ghost = (nname not in self.all_node_info)
2744
          node_image[nname] = gnode
2745

    
2746
      inst_config.MapLVsByNode(node_vol_should)
2747

    
2748
      pnode = inst_config.primary_node
2749
      node_image[pnode].pinst.append(instance)
2750

    
2751
      for snode in inst_config.secondary_nodes:
2752
        nimg = node_image[snode]
2753
        nimg.sinst.append(instance)
2754
        if pnode not in nimg.sbp:
2755
          nimg.sbp[pnode] = []
2756
        nimg.sbp[pnode].append(instance)
2757

    
2758
    # At this point, we have the in-memory data structures complete,
2759
    # except for the runtime information, which we'll gather next
2760

    
2761
    # Due to the way our RPC system works, exact response times cannot be
2762
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2763
    # time before and after executing the request, we can at least have a time
2764
    # window.
2765
    nvinfo_starttime = time.time()
2766
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2767
                                           node_verify_param,
2768
                                           self.cfg.GetClusterName())
2769
    nvinfo_endtime = time.time()
2770

    
2771
    if self.extra_lv_nodes and vg_name is not None:
2772
      extra_lv_nvinfo = \
2773
          self.rpc.call_node_verify(self.extra_lv_nodes,
2774
                                    {constants.NV_LVLIST: vg_name},
2775
                                    self.cfg.GetClusterName())
2776
    else:
2777
      extra_lv_nvinfo = {}
2778

    
2779
    all_drbd_map = self.cfg.ComputeDRBDMap()
2780

    
2781
    feedback_fn("* Gathering disk information (%s nodes)" %
2782
                len(self.my_node_names))
2783
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2784
                                     self.my_inst_info)
2785

    
2786
    feedback_fn("* Verifying configuration file consistency")
2787

    
2788
    # If not all nodes are being checked, we need to make sure the master node
2789
    # and a non-checked vm_capable node are in the list.
2790
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2791
    if absent_nodes:
2792
      vf_nvinfo = all_nvinfo.copy()
2793
      vf_node_info = list(self.my_node_info.values())
2794
      additional_nodes = []
2795
      if master_node not in self.my_node_info:
2796
        additional_nodes.append(master_node)
2797
        vf_node_info.append(self.all_node_info[master_node])
2798
      # Add the first vm_capable node we find which is not included
2799
      for node in absent_nodes:
2800
        nodeinfo = self.all_node_info[node]
2801
        if nodeinfo.vm_capable and not nodeinfo.offline:
2802
          additional_nodes.append(node)
2803
          vf_node_info.append(self.all_node_info[node])
2804
          break
2805
      key = constants.NV_FILELIST
2806
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2807
                                                 {key: node_verify_param[key]},
2808
                                                 self.cfg.GetClusterName()))
2809
    else:
2810
      vf_nvinfo = all_nvinfo
2811
      vf_node_info = self.my_node_info.values()
2812

    
2813
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2814

    
2815
    feedback_fn("* Verifying node status")
2816

    
2817
    refos_img = None
2818

    
2819
    for node_i in node_data_list:
2820
      node = node_i.name
2821
      nimg = node_image[node]
2822

    
2823
      if node_i.offline:
2824
        if verbose:
2825
          feedback_fn("* Skipping offline node %s" % (node,))
2826
        n_offline += 1
2827
        continue
2828

    
2829
      if node == master_node:
2830
        ntype = "master"
2831
      elif node_i.master_candidate:
2832
        ntype = "master candidate"
2833
      elif node_i.drained:
2834
        ntype = "drained"
2835
        n_drained += 1
2836
      else:
2837
        ntype = "regular"
2838
      if verbose:
2839
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2840

    
2841
      msg = all_nvinfo[node].fail_msg
2842
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2843
      if msg:
2844
        nimg.rpc_fail = True
2845
        continue
2846

    
2847
      nresult = all_nvinfo[node].payload
2848

    
2849
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2850
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2851
      self._VerifyNodeNetwork(node_i, nresult)
2852
      self._VerifyOob(node_i, nresult)
2853

    
2854
      if nimg.vm_capable:
2855
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2856
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2857
                             all_drbd_map)
2858

    
2859
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2860
        self._UpdateNodeInstances(node_i, nresult, nimg)
2861
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2862
        self._UpdateNodeOS(node_i, nresult, nimg)
2863

    
2864
        if not nimg.os_fail:
2865
          if refos_img is None:
2866
            refos_img = nimg
2867
          self._VerifyNodeOS(node_i, nimg, refos_img)
2868
        self._VerifyNodeBridges(node_i, nresult, bridges)
2869

    
2870
        # Check whether all running instancies are primary for the node. (This
2871
        # can no longer be done from _VerifyInstance below, since some of the
2872
        # wrong instances could be from other node groups.)
2873
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2874

    
2875
        for inst in non_primary_inst:
2876
          test = inst in self.all_inst_info
2877
          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
2878
                   "instance should not run on node %s", node_i.name)
2879
          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
2880
                   "node is running unknown instance %s", inst)
2881

    
2882
    for node, result in extra_lv_nvinfo.items():
2883
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
2884
                              node_image[node], vg_name)
2885

    
2886
    feedback_fn("* Verifying instance status")
2887
    for instance in self.my_inst_names:
2888
      if verbose:
2889
        feedback_fn("* Verifying instance %s" % instance)
2890
      inst_config = self.my_inst_info[instance]
2891
      self._VerifyInstance(instance, inst_config, node_image,
2892
                           instdisk[instance])
2893
      inst_nodes_offline = []
2894

    
2895
      pnode = inst_config.primary_node
2896
      pnode_img = node_image[pnode]
2897
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2898
               self.ENODERPC, pnode, "instance %s, connection to"
2899
               " primary node failed", instance)
2900

    
2901
      _ErrorIf(inst_config.admin_up and pnode_img.offline,
2902
               self.EINSTANCEBADNODE, instance,
2903
               "instance is marked as running and lives on offline node %s",
2904
               inst_config.primary_node)
2905

    
2906
      # If the instance is non-redundant we cannot survive losing its primary
2907
      # node, so we are not N+1 compliant. On the other hand we have no disk
2908
      # templates with more than one secondary so that situation is not well
2909
      # supported either.
2910
      # FIXME: does not support file-backed instances
2911
      if not inst_config.secondary_nodes:
2912
        i_non_redundant.append(instance)
2913

    
2914
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2915
               instance, "instance has multiple secondary nodes: %s",
2916
               utils.CommaJoin(inst_config.secondary_nodes),
2917
               code=self.ETYPE_WARNING)
2918

    
2919
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2920
        pnode = inst_config.primary_node
2921
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2922
        instance_groups = {}
2923

    
2924
        for node in instance_nodes:
2925
          instance_groups.setdefault(self.all_node_info[node].group,
2926
                                     []).append(node)
2927

    
2928
        pretty_list = [
2929
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2930
          # Sort so that we always list the primary node first.
2931
          for group, nodes in sorted(instance_groups.items(),
2932
                                     key=lambda (_, nodes): pnode in nodes,
2933
                                     reverse=True)]
2934

    
2935
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2936
                      instance, "instance has primary and secondary nodes in"
2937
                      " different groups: %s", utils.CommaJoin(pretty_list),
2938
                      code=self.ETYPE_WARNING)
2939

    
2940
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2941
        i_non_a_balanced.append(instance)
2942

    
2943
      for snode in inst_config.secondary_nodes:
2944
        s_img = node_image[snode]
2945
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2946
                 "instance %s, connection to secondary node failed", instance)
2947

    
2948
        if s_img.offline:
2949
          inst_nodes_offline.append(snode)
2950

    
2951
      # warn that the instance lives on offline nodes
2952
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2953
               "instance has offline secondary node(s) %s",
2954
               utils.CommaJoin(inst_nodes_offline))
2955
      # ... or ghost/non-vm_capable nodes
2956
      for node in inst_config.all_nodes:
2957
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2958
                 "instance lives on ghost node %s", node)
2959
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2960
                 instance, "instance lives on non-vm_capable node %s", node)
2961

    
2962
    feedback_fn("* Verifying orphan volumes")
2963
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2964

    
2965
    # We will get spurious "unknown volume" warnings if any node of this group
2966
    # is secondary for an instance whose primary is in another group. To avoid
2967
    # them, we find these instances and add their volumes to node_vol_should.
2968
    for inst in self.all_inst_info.values():
2969
      for secondary in inst.secondary_nodes:
2970
        if (secondary in self.my_node_info
2971
            and inst.name not in self.my_inst_info):
2972
          inst.MapLVsByNode(node_vol_should)
2973
          break
2974

    
2975
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2976

    
2977
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2978
      feedback_fn("* Verifying N+1 Memory redundancy")
2979
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
2980

    
2981
    feedback_fn("* Other Notes")
2982
    if i_non_redundant:
2983
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2984
                  % len(i_non_redundant))
2985

    
2986
    if i_non_a_balanced:
2987
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2988
                  % len(i_non_a_balanced))
2989

    
2990
    if n_offline:
2991
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2992

    
2993
    if n_drained:
2994
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2995

    
2996
    return not self.bad
2997

    
2998
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2999
    """Analyze the post-hooks' result
3000

3001
    This method analyses the hook result, handles it, and sends some
3002
    nicely-formatted feedback back to the user.
3003

3004
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
3005
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
3006
    @param hooks_results: the results of the multi-node hooks rpc call
3007
    @param feedback_fn: function used send feedback back to the caller
3008
    @param lu_result: previous Exec result
3009
    @return: the new Exec result, based on the previous result
3010
        and hook results
3011

3012
    """
3013
    # We only really run POST phase hooks, only for non-empty groups,
3014
    # and are only interested in their results
3015
    if not self.my_node_names:
3016
      # empty node group
3017
      pass
3018
    elif phase == constants.HOOKS_PHASE_POST:
3019
      # Used to change hooks' output to proper indentation
3020
      feedback_fn("* Hooks Results")
3021
      assert hooks_results, "invalid result from hooks"
3022

    
3023
      for node_name in hooks_results:
3024
        res = hooks_results[node_name]
3025
        msg = res.fail_msg
3026
        test = msg and not res.offline
3027
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
3028
                      "Communication failure in hooks execution: %s", msg)
3029
        if res.offline or msg:
3030
          # No need to investigate payload if node is offline or gave
3031
          # an error.
3032
          continue
3033
        for script, hkr, output in res.payload:
3034
          test = hkr == constants.HKR_FAIL
3035
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
3036
                        "Script %s failed, output:", script)
3037
          if test:
3038
            output = self._HOOKS_INDENT_RE.sub("      ", output)
3039
            feedback_fn("%s" % output)
3040
            lu_result = False
3041

    
3042
    return lu_result
3043

    
3044

    
3045
class LUClusterVerifyDisks(NoHooksLU):
3046
  """Verifies the cluster disks status.
3047

3048
  """
3049
  REQ_BGL = False
3050

    
3051
  def ExpandNames(self):
3052
    self.share_locks = _ShareAll()
3053
    self.needed_locks = {
3054
      locking.LEVEL_NODEGROUP: locking.ALL_SET,
3055
      }
3056

    
3057
  def Exec(self, feedback_fn):
3058
    group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3059

    
3060
    # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3061
    return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3062
                           for group in group_names])
3063

    
3064

    
3065
class LUGroupVerifyDisks(NoHooksLU):
3066
  """Verifies the status of all disks in a node group.
3067

3068
  """
3069
  REQ_BGL = False
3070

    
3071
  def ExpandNames(self):
3072
    # Raises errors.OpPrereqError on its own if group can't be found
3073
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3074

    
3075
    self.share_locks = _ShareAll()
3076
    self.needed_locks = {
3077
      locking.LEVEL_INSTANCE: [],
3078
      locking.LEVEL_NODEGROUP: [],
3079
      locking.LEVEL_NODE: [],
3080
      }
3081

    
3082
  def DeclareLocks(self, level):
3083
    if level == locking.LEVEL_INSTANCE:
3084
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
3085

    
3086
      # Lock instances optimistically, needs verification once node and group
3087
      # locks have been acquired
3088
      self.needed_locks[locking.LEVEL_INSTANCE] = \
3089
        self.cfg.GetNodeGroupInstances(self.group_uuid)
3090

    
3091
    elif level == locking.LEVEL_NODEGROUP:
3092
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3093

    
3094
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
3095
        set([self.group_uuid] +
3096
            # Lock all groups used by instances optimistically; this requires
3097
            # going via the node before it's locked, requiring verification
3098
            # later on
3099
            [group_uuid
3100
             for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3101
             for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3102

    
3103
    elif level == locking.LEVEL_NODE:
3104
      # This will only lock the nodes in the group to be verified which contain
3105
      # actual instances
3106
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3107
      self._LockInstancesNodes()
3108

    
3109
      # Lock all nodes in group to be verified
3110
      assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3111
      member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3112
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3113

    
3114
  def CheckPrereq(self):
3115
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3116
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3117
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3118

    
3119
    assert self.group_uuid in owned_groups
3120

    
3121
    # Check if locked instances are still correct
3122
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3123

    
3124
    # Get instance information
3125
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3126

    
3127
    # Check if node groups for locked instances are still correct
3128
    for (instance_name, inst) in self.instances.items():
3129
      assert owned_nodes.issuperset(inst.all_nodes), \
3130
        "Instance %s's nodes changed while we kept the lock" % instance_name
3131

    
3132
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
3133
                                             owned_groups)
3134

    
3135
      assert self.group_uuid in inst_groups, \
3136
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
3137

    
3138
  def Exec(self, feedback_fn):
3139
    """Verify integrity of cluster disks.
3140

3141
    @rtype: tuple of three items
3142
    @return: a tuple of (dict of node-to-node_error, list of instances
3143
        which need activate-disks, dict of instance: (node, volume) for
3144
        missing volumes
3145

3146
    """
3147
    res_nodes = {}
3148
    res_instances = set()
3149
    res_missing = {}
3150

    
3151
    nv_dict = _MapInstanceDisksToNodes([inst
3152
                                        for inst in self.instances.values()
3153
                                        if inst.admin_up])
3154

    
3155
    if nv_dict:
3156
      nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3157
                             set(self.cfg.GetVmCapableNodeList()))
3158

    
3159
      node_lvs = self.rpc.call_lv_list(nodes, [])
3160

    
3161
      for (node, node_res) in node_lvs.items():
3162
        if node_res.offline:
3163
          continue
3164

    
3165
        msg = node_res.fail_msg
3166
        if msg:
3167
          logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3168
          res_nodes[node] = msg
3169
          continue
3170

    
3171
        for lv_name, (_, _, lv_online) in node_res.payload.items():
3172
          inst = nv_dict.pop((node, lv_name), None)
3173
          if not (lv_online or inst is None):
3174
            res_instances.add(inst)
3175

    
3176
      # any leftover items in nv_dict are missing LVs, let's arrange the data
3177
      # better
3178
      for key, inst in nv_dict.iteritems():
3179
        res_missing.setdefault(inst, []).append(list(key))
3180

    
3181
    return (res_nodes, list(res_instances), res_missing)
3182

    
3183

    
3184
class LUClusterRepairDiskSizes(NoHooksLU):
3185
  """Verifies the cluster disks sizes.
3186

3187
  """
3188
  REQ_BGL = False
3189

    
3190
  def ExpandNames(self):
3191
    if self.op.instances:
3192
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
3193
      self.needed_locks = {
3194
        locking.LEVEL_NODE: [],
3195
        locking.LEVEL_INSTANCE: self.wanted_names,
3196
        }
3197
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3198
    else:
3199
      self.wanted_names = None
3200
      self.needed_locks = {
3201
        locking.LEVEL_NODE: locking.ALL_SET,
3202
        locking.LEVEL_INSTANCE: locking.ALL_SET,
3203
        }
3204
    self.share_locks = {
3205
      locking.LEVEL_NODE: 1,
3206
      locking.LEVEL_INSTANCE: 0,
3207
      }
3208

    
3209
  def DeclareLocks(self, level):
3210
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
3211
      self._LockInstancesNodes(primary_only=True)
3212

    
3213
  def CheckPrereq(self):
3214
    """Check prerequisites.
3215

3216
    This only checks the optional instance list against the existing names.
3217

3218
    """
3219
    if self.wanted_names is None:
3220
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3221

    
3222
    self.wanted_instances = \
3223
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3224

    
3225
  def _EnsureChildSizes(self, disk):
3226
    """Ensure children of the disk have the needed disk size.
3227

3228
    This is valid mainly for DRBD8 and fixes an issue where the
3229
    children have smaller disk size.
3230

3231
    @param disk: an L{ganeti.objects.Disk} object
3232

3233
    """
3234
    if disk.dev_type == constants.LD_DRBD8:
3235
      assert disk.children, "Empty children for DRBD8?"
3236
      fchild = disk.children[0]
3237
      mismatch = fchild.size < disk.size
3238
      if mismatch:
3239
        self.LogInfo("Child disk has size %d, parent %d, fixing",
3240
                     fchild.size, disk.size)
3241
        fchild.size = disk.size
3242

    
3243
      # and we recurse on this child only, not on the metadev
3244
      return self._EnsureChildSizes(fchild) or mismatch
3245
    else:
3246
      return False
3247

    
3248
  def Exec(self, feedback_fn):
3249
    """Verify the size of cluster disks.
3250

3251
    """
3252
    # TODO: check child disks too
3253
    # TODO: check differences in size between primary/secondary nodes
3254
    per_node_disks = {}
3255
    for instance in self.wanted_instances:
3256
      pnode = instance.primary_node
3257
      if pnode not in per_node_disks:
3258
        per_node_disks[pnode] = []
3259
      for idx, disk in enumerate(instance.disks):
3260
        per_node_disks[pnode].append((instance, idx, disk))
3261

    
3262
    changed = []
3263
    for node, dskl in per_node_disks.items():
3264
      newl = [v[2].Copy() for v in dskl]
3265
      for dsk in newl:
3266
        self.cfg.SetDiskID(dsk, node)
3267
      result = self.rpc.call_blockdev_getsize(node, newl)
3268
      if result.fail_msg:
3269
        self.LogWarning("Failure in blockdev_getsize call to node"
3270
                        " %s, ignoring", node)
3271
        continue
3272
      if len(result.payload) != len(dskl):
3273
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
3274
                        " result.payload=%s", node, len(dskl), result.payload)
3275
        self.LogWarning("Invalid result from node %s, ignoring node results",
3276
                        node)
3277
        continue
3278
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
3279
        if size is None:
3280
          self.LogWarning("Disk %d of instance %s did not return size"
3281
                          " information, ignoring", idx, instance.name)
3282
          continue
3283
        if not isinstance(size, (int, long)):
3284
          self.LogWarning("Disk %d of instance %s did not return valid"
3285
                          " size information, ignoring", idx, instance.name)
3286
          continue
3287
        size = size >> 20
3288
        if size != disk.size:
3289
          self.LogInfo("Disk %d of instance %s has mismatched size,"
3290
                       " correcting: recorded %d, actual %d", idx,
3291
                       instance.name, disk.size, size)
3292
          disk.size = size
3293
          self.cfg.Update(instance, feedback_fn)
3294
          changed.append((instance.name, idx, size))
3295
        if self._EnsureChildSizes(disk):
3296
          self.cfg.Update(instance, feedback_fn)
3297
          changed.append((instance.name, idx, disk.size))
3298
    return changed
3299

    
3300

    
3301
class LUClusterRename(LogicalUnit):
3302
  """Rename the cluster.
3303

3304
  """
3305
  HPATH = "cluster-rename"
3306
  HTYPE = constants.HTYPE_CLUSTER
3307

    
3308
  def BuildHooksEnv(self):
3309
    """Build hooks env.
3310

3311
    """
3312
    return {
3313
      "OP_TARGET": self.cfg.GetClusterName(),
3314
      "NEW_NAME": self.op.name,
3315
      }
3316

    
3317
  def BuildHooksNodes(self):
3318
    """Build hooks nodes.
3319

3320
    """
3321
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3322

    
3323
  def CheckPrereq(self):
3324
    """Verify that the passed name is a valid one.
3325

3326
    """
3327
    hostname = netutils.GetHostname(name=self.op.name,
3328
                                    family=self.cfg.GetPrimaryIPFamily())
3329

    
3330
    new_name = hostname.name
3331
    self.ip = new_ip = hostname.ip
3332
    old_name = self.cfg.GetClusterName()
3333
    old_ip = self.cfg.GetMasterIP()
3334
    if new_name == old_name and new_ip == old_ip:
3335
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3336
                                 " cluster has changed",
3337
                                 errors.ECODE_INVAL)
3338
    if new_ip != old_ip:
3339
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3340
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3341
                                   " reachable on the network" %
3342
                                   new_ip, errors.ECODE_NOTUNIQUE)
3343

    
3344
    self.op.name = new_name
3345

    
3346
  def Exec(self, feedback_fn):
3347
    """Rename the cluster.
3348

3349
    """
3350
    clustername = self.op.name
3351
    ip = self.ip
3352

    
3353
    # shutdown the master IP
3354
    master = self.cfg.GetMasterNode()
3355
    result = self.rpc.call_node_stop_master(master, False)
3356
    result.Raise("Could not disable the master role")
3357

    
3358
    try:
3359
      cluster = self.cfg.GetClusterInfo()
3360
      cluster.cluster_name = clustername
3361
      cluster.master_ip = ip
3362
      self.cfg.Update(cluster, feedback_fn)
3363

    
3364
      # update the known hosts file
3365
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3366
      node_list = self.cfg.GetOnlineNodeList()
3367
      try:
3368
        node_list.remove(master)
3369
      except ValueError:
3370
        pass
3371
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3372
    finally:
3373
      result = self.rpc.call_node_start_master(master, False, False)
3374
      msg = result.fail_msg
3375
      if msg:
3376
        self.LogWarning("Could not re-enable the master role on"
3377
                        " the master, please restart manually: %s", msg)
3378

    
3379
    return clustername
3380

    
3381

    
3382
class LUClusterSetParams(LogicalUnit):
3383
  """Change the parameters of the cluster.
3384

3385
  """
3386
  HPATH = "cluster-modify"
3387
  HTYPE = constants.HTYPE_CLUSTER
3388
  REQ_BGL = False
3389

    
3390
  def CheckArguments(self):
3391
    """Check parameters
3392

3393
    """
3394
    if self.op.uid_pool:
3395
      uidpool.CheckUidPool(self.op.uid_pool)
3396

    
3397
    if self.op.add_uids:
3398
      uidpool.CheckUidPool(self.op.add_uids)
3399

    
3400
    if self.op.remove_uids:
3401
      uidpool.CheckUidPool(self.op.remove_uids)
3402

    
3403
  def ExpandNames(self):
3404
    # FIXME: in the future maybe other cluster params won't require checking on
3405
    # all nodes to be modified.
3406
    self.needed_locks = {
3407
      locking.LEVEL_NODE: locking.ALL_SET,
3408
    }
3409
    self.share_locks[locking.LEVEL_NODE] = 1
3410

    
3411
  def BuildHooksEnv(self):
3412
    """Build hooks env.
3413

3414
    """
3415
    return {
3416
      "OP_TARGET": self.cfg.GetClusterName(),
3417
      "NEW_VG_NAME": self.op.vg_name,
3418
      }
3419

    
3420
  def BuildHooksNodes(self):
3421
    """Build hooks nodes.
3422

3423
    """
3424
    mn = self.cfg.GetMasterNode()
3425
    return ([mn], [mn])
3426

    
3427
  def CheckPrereq(self):
3428
    """Check prerequisites.
3429

3430
    This checks whether the given params don't conflict and
3431
    if the given volume group is valid.
3432

3433
    """
3434
    if self.op.vg_name is not None and not self.op.vg_name:
3435
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3436
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3437
                                   " instances exist", errors.ECODE_INVAL)
3438

    
3439
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3440
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3441
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3442
                                   " drbd-based instances exist",
3443
                                   errors.ECODE_INVAL)
3444

    
3445
    node_list = self.owned_locks(locking.LEVEL_NODE)
3446

    
3447
    # if vg_name not None, checks given volume group on all nodes
3448
    if self.op.vg_name:
3449
      vglist = self.rpc.call_vg_list(node_list)
3450
      for node in node_list:
3451
        msg = vglist[node].fail_msg
3452
        if msg:
3453
          # ignoring down node
3454
          self.LogWarning("Error while gathering data on node %s"
3455
                          " (ignoring node): %s", node, msg)
3456
          continue
3457
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3458
                                              self.op.vg_name,
3459
                                              constants.MIN_VG_SIZE)
3460
        if vgstatus:
3461
          raise errors.OpPrereqError("Error on node '%s': %s" %
3462
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3463

    
3464
    if self.op.drbd_helper:
3465
      # checks given drbd helper on all nodes
3466
      helpers = self.rpc.call_drbd_helper(node_list)
3467
      for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3468
        if ninfo.offline:
3469
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3470
          continue
3471
        msg = helpers[node].fail_msg
3472
        if msg:
3473
          raise errors.OpPrereqError("Error checking drbd helper on node"
3474
                                     " '%s': %s" % (node, msg),
3475
                                     errors.ECODE_ENVIRON)
3476
        node_helper = helpers[node].payload
3477
        if node_helper != self.op.drbd_helper:
3478
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3479
                                     (node, node_helper), errors.ECODE_ENVIRON)
3480

    
3481
    self.cluster = cluster = self.cfg.GetClusterInfo()
3482
    # validate params changes
3483
    if self.op.beparams:
3484
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3485
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3486

    
3487
    if self.op.ndparams:
3488
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3489
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3490

    
3491
      # TODO: we need a more general way to handle resetting
3492
      # cluster-level parameters to default values
3493
      if self.new_ndparams["oob_program"] == "":
3494
        self.new_ndparams["oob_program"] = \
3495
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3496

    
3497
    if self.op.nicparams:
3498
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3499
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3500
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3501
      nic_errors = []
3502

    
3503
      # check all instances for consistency
3504
      for instance in self.cfg.GetAllInstancesInfo().values():
3505
        for nic_idx, nic in enumerate(instance.nics):
3506
          params_copy = copy.deepcopy(nic.nicparams)
3507
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3508

    
3509
          # check parameter syntax
3510
          try:
3511
            objects.NIC.CheckParameterSyntax(params_filled)
3512
          except errors.ConfigurationError, err:
3513
            nic_errors.append("Instance %s, nic/%d: %s" %
3514
                              (instance.name, nic_idx, err))
3515

    
3516
          # if we're moving instances to routed, check that they have an ip
3517
          target_mode = params_filled[constants.NIC_MODE]
3518
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3519
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3520
                              " address" % (instance.name, nic_idx))
3521
      if nic_errors:
3522
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3523
                                   "\n".join(nic_errors))
3524

    
3525
    # hypervisor list/parameters
3526
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3527
    if self.op.hvparams:
3528
      for hv_name, hv_dict in self.op.hvparams.items():
3529
        if hv_name not in self.new_hvparams:
3530
          self.new_hvparams[hv_name] = hv_dict
3531
        else:
3532
          self.new_hvparams[hv_name].update(hv_dict)
3533

    
3534
    # os hypervisor parameters
3535
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3536
    if self.op.os_hvp:
3537
      for os_name, hvs in self.op.os_hvp.items():
3538
        if os_name not in self.new_os_hvp:
3539
          self.new_os_hvp[os_name] = hvs
3540
        else:
3541
          for hv_name, hv_dict in hvs.items():
3542
            if hv_name not in self.new_os_hvp[os_name]:
3543
              self.new_os_hvp[os_name][hv_name] = hv_dict
3544
            else:
3545
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3546

    
3547
    # os parameters
3548
    self.new_osp = objects.FillDict(cluster.osparams, {})
3549
    if self.op.osparams:
3550
      for os_name, osp in self.op.osparams.items():
3551
        if os_name not in self.new_osp:
3552
          self.new_osp[os_name] = {}
3553

    
3554
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3555
                                                  use_none=True)
3556

    
3557
        if not self.new_osp[os_name]:
3558
          # we removed all parameters
3559
          del self.new_osp[os_name]
3560
        else:
3561
          # check the parameter validity (remote check)
3562
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3563
                         os_name, self.new_osp[os_name])
3564

    
3565
    # changes to the hypervisor list
3566
    if self.op.enabled_hypervisors is not None:
3567
      self.hv_list = self.op.enabled_hypervisors
3568
      for hv in self.hv_list:
3569
        # if the hypervisor doesn't already exist in the cluster
3570
        # hvparams, we initialize it to empty, and then (in both
3571
        # cases) we make sure to fill the defaults, as we might not
3572
        # have a complete defaults list if the hypervisor wasn't
3573
        # enabled before
3574
        if hv not in new_hvp:
3575
          new_hvp[hv] = {}
3576
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3577
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3578
    else:
3579
      self.hv_list = cluster.enabled_hypervisors
3580

    
3581
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3582
      # either the enabled list has changed, or the parameters have, validate
3583
      for hv_name, hv_params in self.new_hvparams.items():
3584
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3585
            (self.op.enabled_hypervisors and
3586
             hv_name in self.op.enabled_hypervisors)):
3587
          # either this is a new hypervisor, or its parameters have changed
3588
          hv_class = hypervisor.GetHypervisor(hv_name)
3589
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3590
          hv_class.CheckParameterSyntax(hv_params)
3591
          _CheckHVParams(self, node_list, hv_name, hv_params)
3592

    
3593
    if self.op.os_hvp:
3594
      # no need to check any newly-enabled hypervisors, since the
3595
      # defaults have already been checked in the above code-block
3596
      for os_name, os_hvp in self.new_os_hvp.items():
3597
        for hv_name, hv_params in os_hvp.items():
3598
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3599
          # we need to fill in the new os_hvp on top of the actual hv_p
3600
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3601
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3602
          hv_class = hypervisor.GetHypervisor(hv_name)
3603
          hv_class.CheckParameterSyntax(new_osp)
3604
          _CheckHVParams(self, node_list, hv_name, new_osp)
3605

    
3606
    if self.op.default_iallocator:
3607
      alloc_script = utils.FindFile(self.op.default_iallocator,
3608
                                    constants.IALLOCATOR_SEARCH_PATH,
3609
                                    os.path.isfile)
3610
      if alloc_script is None:
3611
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3612
                                   " specified" % self.op.default_iallocator,
3613
                                   errors.ECODE_INVAL)
3614

    
3615
  def Exec(self, feedback_fn):
3616
    """Change the parameters of the cluster.
3617

3618
    """
3619
    if self.op.vg_name is not None:
3620
      new_volume = self.op.vg_name
3621
      if not new_volume:
3622
        new_volume = None
3623
      if new_volume != self.cfg.GetVGName():
3624
        self.cfg.SetVGName(new_volume)
3625
      else:
3626
        feedback_fn("Cluster LVM configuration already in desired"
3627
                    " state, not changing")
3628
    if self.op.drbd_helper is not None:
3629
      new_helper = self.op.drbd_helper
3630
      if not new_helper:
3631
        new_helper = None
3632
      if new_helper != self.cfg.GetDRBDHelper():
3633
        self.cfg.SetDRBDHelper(new_helper)
3634
      else:
3635
        feedback_fn("Cluster DRBD helper already in desired state,"
3636
                    " not changing")
3637
    if self.op.hvparams:
3638
      self.cluster.hvparams = self.new_hvparams
3639
    if self.op.os_hvp:
3640
      self.cluster.os_hvp = self.new_os_hvp
3641
    if self.op.enabled_hypervisors is not None:
3642
      self.cluster.hvparams = self.new_hvparams
3643
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3644
    if self.op.beparams:
3645
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3646
    if self.op.nicparams:
3647
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3648
    if self.op.osparams:
3649
      self.cluster.osparams = self.new_osp
3650
    if self.op.ndparams:
3651
      self.cluster.ndparams = self.new_ndparams
3652

    
3653
    if self.op.candidate_pool_size is not None:
3654
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3655
      # we need to update the pool size here, otherwise the save will fail
3656
      _AdjustCandidatePool(self, [])
3657

    
3658
    if self.op.maintain_node_health is not None:
3659
      self.cluster.maintain_node_health = self.op.maintain_node_health
3660

    
3661
    if self.op.prealloc_wipe_disks is not None:
3662
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3663

    
3664
    if self.op.add_uids is not None:
3665
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3666

    
3667
    if self.op.remove_uids is not None:
3668
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3669

    
3670
    if self.op.uid_pool is not None:
3671
      self.cluster.uid_pool = self.op.uid_pool
3672

    
3673
    if self.op.default_iallocator is not None:
3674
      self.cluster.default_iallocator = self.op.default_iallocator
3675

    
3676
    if self.op.reserved_lvs is not None:
3677
      self.cluster.reserved_lvs = self.op.reserved_lvs
3678

    
3679
    def helper_os(aname, mods, desc):
3680
      desc += " OS list"
3681
      lst = getattr(self.cluster, aname)
3682
      for key, val in mods:
3683
        if key == constants.DDM_ADD:
3684
          if val in lst:
3685
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3686
          else:
3687
            lst.append(val)
3688
        elif key == constants.DDM_REMOVE:
3689
          if val in lst:
3690
            lst.remove(val)
3691
          else:
3692
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3693
        else:
3694
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3695

    
3696
    if self.op.hidden_os:
3697
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3698

    
3699
    if self.op.blacklisted_os:
3700
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3701

    
3702
    if self.op.master_netdev:
3703
      master = self.cfg.GetMasterNode()
3704
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3705
                  self.cluster.master_netdev)
3706
      result = self.rpc.call_node_stop_master(master, False)
3707
      result.Raise("Could not disable the master ip")
3708
      feedback_fn("Changing master_netdev from %s to %s" %
3709
                  (self.cluster.master_netdev, self.op.master_netdev))
3710
      self.cluster.master_netdev = self.op.master_netdev
3711

    
3712
    self.cfg.Update(self.cluster, feedback_fn)
3713

    
3714
    if self.op.master_netdev:
3715
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3716
                  self.op.master_netdev)
3717
      result = self.rpc.call_node_start_master(master, False, False)
3718
      if result.fail_msg:
3719
        self.LogWarning("Could not re-enable the master ip on"
3720
                        " the master, please restart manually: %s",
3721
                        result.fail_msg)
3722

    
3723

    
3724
def _UploadHelper(lu, nodes, fname):
3725
  """Helper for uploading a file and showing warnings.
3726

3727
  """
3728
  if os.path.exists(fname):
3729
    result = lu.rpc.call_upload_file(nodes, fname)
3730
    for to_node, to_result in result.items():
3731
      msg = to_result.fail_msg
3732
      if msg:
3733
        msg = ("Copy of file %s to node %s failed: %s" %
3734
               (fname, to_node, msg))
3735
        lu.proc.LogWarning(msg)
3736

    
3737

    
3738
def _ComputeAncillaryFiles(cluster, redist):
3739
  """Compute files external to Ganeti which need to be consistent.
3740

3741
  @type redist: boolean
3742
  @param redist: Whether to include files which need to be redistributed
3743

3744
  """
3745
  # Compute files for all nodes
3746
  files_all = set([
3747
    constants.SSH_KNOWN_HOSTS_FILE,
3748
    constants.CONFD_HMAC_KEY,
3749
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3750
    ])
3751

    
3752
  if not redist:
3753
    files_all.update(constants.ALL_CERT_FILES)
3754
    files_all.update(ssconf.SimpleStore().GetFileList())
3755
  else:
3756
    # we need to ship at least the RAPI certificate
3757
    files_all.add(constants.RAPI_CERT_FILE)
3758

    
3759
  if cluster.modify_etc_hosts:
3760
    files_all.add(constants.ETC_HOSTS)
3761

    
3762
  # Files which must either exist on all nodes or on none
3763
  files_all_opt = set([
3764
    constants.RAPI_USERS_FILE,
3765
    ])
3766

    
3767
  # Files which should only be on master candidates
3768
  files_mc = set()
3769
  if not redist:
3770
    files_mc.add(constants.CLUSTER_CONF_FILE)
3771

    
3772
  # Files which should only be on VM-capable nodes
3773
  files_vm = set(filename
3774
    for hv_name in cluster.enabled_hypervisors
3775
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
3776

    
3777
  # Filenames must be unique
3778
  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
3779
          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
3780
         "Found file listed in more than one file list"
3781

    
3782
  return (files_all, files_all_opt, files_mc, files_vm)
3783

    
3784

    
3785
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3786
  """Distribute additional files which are part of the cluster configuration.
3787

3788
  ConfigWriter takes care of distributing the config and ssconf files, but
3789
  there are more files which should be distributed to all nodes. This function
3790
  makes sure those are copied.
3791

3792
  @param lu: calling logical unit
3793
  @param additional_nodes: list of nodes not in the config to distribute to
3794
  @type additional_vm: boolean
3795
  @param additional_vm: whether the additional nodes are vm-capable or not
3796

3797
  """
3798
  # Gather target nodes
3799
  cluster = lu.cfg.GetClusterInfo()
3800
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3801

    
3802
  online_nodes = lu.cfg.GetOnlineNodeList()
3803
  vm_nodes = lu.cfg.GetVmCapableNodeList()
3804

    
3805
  if additional_nodes is not None:
3806
    online_nodes.extend(additional_nodes)
3807
    if additional_vm:
3808
      vm_nodes.extend(additional_nodes)
3809

    
3810
  # Never distribute to master node
3811
  for nodelist in [online_nodes, vm_nodes]:
3812
    if master_info.name in nodelist:
3813
      nodelist.remove(master_info.name)
3814

    
3815
  # Gather file lists
3816
  (files_all, files_all_opt, files_mc, files_vm) = \
3817
    _ComputeAncillaryFiles(cluster, True)
3818

    
3819
  # Never re-distribute configuration file from here
3820
  assert not (constants.CLUSTER_CONF_FILE in files_all or
3821
              constants.CLUSTER_CONF_FILE in files_vm)
3822
  assert not files_mc, "Master candidates not handled in this function"
3823

    
3824
  filemap = [
3825
    (online_nodes, files_all),
3826
    (online_nodes, files_all_opt),
3827
    (vm_nodes, files_vm),
3828
    ]
3829

    
3830
  # Upload the files
3831
  for (node_list, files) in filemap:
3832
    for fname in files:
3833
      _UploadHelper(lu, node_list, fname)
3834

    
3835

    
3836
class LUClusterRedistConf(NoHooksLU):
3837
  """Force the redistribution of cluster configuration.
3838

3839
  This is a very simple LU.
3840

3841
  """
3842
  REQ_BGL = False
3843

    
3844
  def ExpandNames(self):
3845
    self.needed_locks = {
3846
      locking.LEVEL_NODE: locking.ALL_SET,
3847
    }
3848
    self.share_locks[locking.LEVEL_NODE] = 1
3849

    
3850
  def Exec(self, feedback_fn):
3851
    """Redistribute the configuration.
3852

3853
    """
3854
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3855
    _RedistributeAncillaryFiles(self)
3856

    
3857

    
3858
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3859
  """Sleep and poll for an instance's disk to sync.
3860

3861
  """
3862
  if not instance.disks or disks is not None and not disks:
3863
    return True
3864

    
3865
  disks = _ExpandCheckDisks(instance, disks)
3866

    
3867
  if not oneshot:
3868
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3869

    
3870
  node = instance.primary_node
3871

    
3872
  for dev in disks:
3873
    lu.cfg.SetDiskID(dev, node)
3874

    
3875
  # TODO: Convert to utils.Retry
3876

    
3877
  retries = 0
3878
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3879
  while True:
3880
    max_time = 0
3881
    done = True
3882
    cumul_degraded = False
3883
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3884
    msg = rstats.fail_msg
3885
    if msg:
3886
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3887
      retries += 1
3888
      if retries >= 10:
3889
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3890
                                 " aborting." % node)
3891
      time.sleep(6)
3892
      continue
3893
    rstats = rstats.payload
3894
    retries = 0
3895
    for i, mstat in enumerate(rstats):
3896
      if mstat is None:
3897
        lu.LogWarning("Can't compute data for node %s/%s",
3898
                           node, disks[i].iv_name)
3899
        continue
3900

    
3901
      cumul_degraded = (cumul_degraded or
3902
                        (mstat.is_degraded and mstat.sync_percent is None))
3903
      if mstat.sync_percent is not None:
3904
        done = False
3905
        if mstat.estimated_time is not None:
3906
          rem_time = ("%s remaining (estimated)" %
3907
                      utils.FormatSeconds(mstat.estimated_time))
3908
          max_time = mstat.estimated_time
3909
        else:
3910
          rem_time = "no time estimate"
3911
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3912
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3913

    
3914
    # if we're done but degraded, let's do a few small retries, to
3915
    # make sure we see a stable and not transient situation; therefore
3916
    # we force restart of the loop
3917
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3918
      logging.info("Degraded disks found, %d retries left", degr_retries)
3919
      degr_retries -= 1
3920
      time.sleep(1)
3921
      continue
3922

    
3923
    if done or oneshot:
3924
      break
3925

    
3926
    time.sleep(min(60, max_time))
3927

    
3928
  if done:
3929
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3930
  return not cumul_degraded
3931

    
3932

    
3933
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3934
  """Check that mirrors are not degraded.
3935

3936
  The ldisk parameter, if True, will change the test from the
3937
  is_degraded attribute (which represents overall non-ok status for
3938
  the device(s)) to the ldisk (representing the local storage status).
3939

3940
  """
3941
  lu.cfg.SetDiskID(dev, node)
3942

    
3943
  result = True
3944

    
3945
  if on_primary or dev.AssembleOnSecondary():
3946
    rstats = lu.rpc.call_blockdev_find(node, dev)
3947
    msg = rstats.fail_msg
3948
    if msg:
3949
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3950
      result = False
3951
    elif not rstats.payload:
3952
      lu.LogWarning("Can't find disk on node %s", node)
3953
      result = False
3954
    else:
3955
      if ldisk:
3956
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3957
      else:
3958
        result = result and not rstats.payload.is_degraded
3959

    
3960
  if dev.children:
3961
    for child in dev.children:
3962
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3963

    
3964
  return result
3965

    
3966

    
3967
class LUOobCommand(NoHooksLU):
3968
  """Logical unit for OOB handling.
3969

3970
  """
3971
  REQ_BGL = False
3972
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
3973

    
3974
  def ExpandNames(self):
3975
    """Gather locks we need.
3976

3977
    """
3978
    if self.op.node_names:
3979
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
3980
      lock_names = self.op.node_names
3981
    else:
3982
      lock_names = locking.ALL_SET
3983

    
3984
    self.needed_locks = {
3985
      locking.LEVEL_NODE: lock_names,
3986
      }
3987

    
3988
  def CheckPrereq(self):
3989
    """Check prerequisites.
3990

3991
    This checks:
3992
     - the node exists in the configuration
3993
     - OOB is supported
3994

3995
    Any errors are signaled by raising errors.OpPrereqError.
3996

3997
    """
3998
    self.nodes = []
3999
    self.master_node = self.cfg.GetMasterNode()
4000

    
4001
    assert self.op.power_delay >= 0.0
4002

    
4003
    if self.op.node_names:
4004
      if (self.op.command in self._SKIP_MASTER and
4005
          self.master_node in self.op.node_names):
4006
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
4007
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
4008

    
4009
        if master_oob_handler:
4010
          additional_text = ("run '%s %s %s' if you want to operate on the"
4011
                             " master regardless") % (master_oob_handler,
4012
                                                      self.op.command,
4013
                                                      self.master_node)
4014
        else:
4015
          additional_text = "it does not support out-of-band operations"
4016

    
4017
        raise errors.OpPrereqError(("Operating on the master node %s is not"
4018
                                    " allowed for %s; %s") %
4019
                                   (self.master_node, self.op.command,
4020
                                    additional_text), errors.ECODE_INVAL)
4021
    else:
4022
      self.op.node_names = self.cfg.GetNodeList()
4023
      if self.op.command in self._SKIP_MASTER:
4024
        self.op.node_names.remove(self.master_node)
4025

    
4026
    if self.op.command in self._SKIP_MASTER:
4027
      assert self.master_node not in self.op.node_names
4028

    
4029
    for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
4030
      if node is None:
4031
        raise errors.OpPrereqError("Node %s not found" % node_name,
4032
                                   errors.ECODE_NOENT)
4033
      else:
4034
        self.nodes.append(node)
4035

    
4036
      if (not self.op.ignore_status and
4037
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
4038
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
4039
                                    " not marked offline") % node_name,
4040
                                   errors.ECODE_STATE)
4041

    
4042
  def Exec(self, feedback_fn):
4043
    """Execute OOB and return result if we expect any.
4044

4045
    """
4046
    master_node = self.master_node
4047
    ret = []
4048

    
4049
    for idx, node in enumerate(utils.NiceSort(self.nodes,
4050
                                              key=lambda node: node.name)):
4051
      node_entry = [(constants.RS_NORMAL, node.name)]
4052
      ret.append(node_entry)
4053

    
4054
      oob_program = _SupportsOob(self.cfg, node)
4055

    
4056
      if not oob_program:
4057
        node_entry.append((constants.RS_UNAVAIL, None))
4058
        continue
4059

    
4060
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
4061
                   self.op.command, oob_program, node.name)
4062
      result = self.rpc.call_run_oob(master_node, oob_program,
4063
                                     self.op.command, node.name,
4064
                                     self.op.timeout)
4065

    
4066
      if result.fail_msg:
4067
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4068
                        node.name, result.fail_msg)
4069
        node_entry.append((constants.RS_NODATA, None))
4070
      else:
4071
        try:
4072
          self._CheckPayload(result)
4073
        except errors.OpExecError, err:
4074
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
4075
                          node.name, err)
4076
          node_entry.append((constants.RS_NODATA, None))
4077
        else:
4078
          if self.op.command == constants.OOB_HEALTH:
4079
            # For health we should log important events
4080
            for item, status in result.payload:
4081
              if status in [constants.OOB_STATUS_WARNING,
4082
                            constants.OOB_STATUS_CRITICAL]:
4083
                self.LogWarning("Item '%s' on node '%s' has status '%s'",
4084
                                item, node.name, status)
4085

    
4086
          if self.op.command == constants.OOB_POWER_ON:
4087
            node.powered = True
4088
          elif self.op.command == constants.OOB_POWER_OFF:
4089
            node.powered = False
4090
          elif self.op.command == constants.OOB_POWER_STATUS:
4091
            powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
4092
            if powered != node.powered:
4093
              logging.warning(("Recorded power state (%s) of node '%s' does not"
4094
                               " match actual power state (%s)"), node.powered,
4095
                              node.name, powered)
4096

    
4097
          # For configuration changing commands we should update the node
4098
          if self.op.command in (constants.OOB_POWER_ON,
4099
                                 constants.OOB_POWER_OFF):
4100
            self.cfg.Update(node, feedback_fn)
4101

    
4102
          node_entry.append((constants.RS_NORMAL, result.payload))
4103

    
4104
          if (self.op.command == constants.OOB_POWER_ON and
4105
              idx < len(self.nodes) - 1):
4106
            time.sleep(self.op.power_delay)
4107

    
4108
    return ret
4109

    
4110
  def _CheckPayload(self, result):
4111
    """Checks if the payload is valid.
4112

4113
    @param result: RPC result
4114
    @raises errors.OpExecError: If payload is not valid
4115

4116
    """
4117
    errs = []
4118
    if self.op.command == constants.OOB_HEALTH:
4119
      if not isinstance(result.payload, list):
4120
        errs.append("command 'health' is expected to return a list but got %s" %
4121
                    type(result.payload))
4122
      else:
4123
        for item, status in result.payload:
4124
          if status not in constants.OOB_STATUSES:
4125
            errs.append("health item '%s' has invalid status '%s'" %
4126
                        (item, status))
4127

    
4128
    if self.op.command == constants.OOB_POWER_STATUS:
4129
      if not isinstance(result.payload, dict):
4130
        errs.append("power-status is expected to return a dict but got %s" %
4131
                    type(result.payload))
4132

    
4133
    if self.op.command in [
4134
        constants.OOB_POWER_ON,
4135
        constants.OOB_POWER_OFF,
4136
        constants.OOB_POWER_CYCLE,
4137
        ]:
4138
      if result.payload is not None:
4139
        errs.append("%s is expected to not return payload but got '%s'" %
4140
                    (self.op.command, result.payload))
4141

    
4142
    if errs:
4143
      raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
4144
                               utils.CommaJoin(errs))
4145

    
4146

    
4147
class _OsQuery(_QueryBase):
4148
  FIELDS = query.OS_FIELDS
4149

    
4150
  def ExpandNames(self, lu):
4151
    # Lock all nodes in shared mode
4152
    # Temporary removal of locks, should be reverted later
4153
    # TODO: reintroduce locks when they are lighter-weight
4154
    lu.needed_locks = {}
4155
    #self.share_locks[locking.LEVEL_NODE] = 1
4156
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4157

    
4158
    # The following variables interact with _QueryBase._GetNames
4159
    if self.names:
4160
      self.wanted = self.names
4161
    else:
4162
      self.wanted = locking.ALL_SET
4163

    
4164
    self.do_locking = self.use_locking
4165

    
4166
  def DeclareLocks(self, lu, level):
4167
    pass
4168

    
4169
  @staticmethod
4170
  def _DiagnoseByOS(rlist):
4171
    """Remaps a per-node return list into an a per-os per-node dictionary
4172

4173
    @param rlist: a map with node names as keys and OS objects as values
4174

4175
    @rtype: dict
4176
    @return: a dictionary with osnames as keys and as value another
4177
        map, with nodes as keys and tuples of (path, status, diagnose,
4178
        variants, parameters, api_versions) as values, eg::
4179

4180
          {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
4181
                                     (/srv/..., False, "invalid api")],
4182
                           "node2": [(/srv/..., True, "", [], [])]}
4183
          }
4184

4185
    """
4186
    all_os = {}
4187
    # we build here the list of nodes that didn't fail the RPC (at RPC
4188
    # level), so that nodes with a non-responding node daemon don't
4189
    # make all OSes invalid
4190
    good_nodes = [node_name for node_name in rlist
4191
                  if not rlist[node_name].fail_msg]
4192
    for node_name, nr in rlist.items():
4193
      if nr.fail_msg or not nr.payload:
4194
        continue
4195
      for (name, path, status, diagnose, variants,
4196
           params, api_versions) in nr.payload:
4197
        if name not in all_os:
4198
          # build a list of nodes for this os containing empty lists
4199
          # for each node in node_list
4200
          all_os[name] = {}
4201
          for nname in good_nodes:
4202
            all_os[name][nname] = []
4203
        # convert params from [name, help] to (name, help)
4204
        params = [tuple(v) for v in params]
4205
        all_os[name][node_name].append((path, status, diagnose,
4206
                                        variants, params, api_versions))
4207
    return all_os
4208

    
4209
  def _GetQueryData(self, lu):
4210
    """Computes the list of nodes and their attributes.
4211

4212
    """
4213
    # Locking is not used
4214
    assert not (compat.any(lu.glm.is_owned(level)
4215
                           for level in locking.LEVELS
4216
                           if level != locking.LEVEL_CLUSTER) or
4217
                self.do_locking or self.use_locking)
4218

    
4219
    valid_nodes = [node.name
4220
                   for node in lu.cfg.GetAllNodesInfo().values()
4221
                   if not node.offline and node.vm_capable]
4222
    pol = self._DiagnoseByOS(lu.rpc.call_os_diagnose(valid_nodes))
4223
    cluster = lu.cfg.GetClusterInfo()
4224

    
4225
    data = {}
4226

    
4227
    for (os_name, os_data) in pol.items():
4228
      info = query.OsInfo(name=os_name, valid=True, node_status=os_data,
4229
                          hidden=(os_name in cluster.hidden_os),
4230
                          blacklisted=(os_name in cluster.blacklisted_os))
4231

    
4232
      variants = set()
4233
      parameters = set()
4234
      api_versions = set()
4235

    
4236
      for idx, osl in enumerate(os_data.values()):
4237
        info.valid = bool(info.valid and osl and osl[0][1])
4238
        if not info.valid:
4239
          break
4240

    
4241
        (node_variants, node_params, node_api) = osl[0][3:6]
4242
        if idx == 0:
4243
          # First entry
4244
          variants.update(node_variants)
4245
          parameters.update(node_params)
4246
          api_versions.update(node_api)
4247
        else:
4248
          # Filter out inconsistent values
4249
          variants.intersection_update(node_variants)
4250
          parameters.intersection_update(node_params)
4251
          api_versions.intersection_update(node_api)
4252

    
4253
      info.variants = list(variants)
4254
      info.parameters = list(parameters)
4255
      info.api_versions = list(api_versions)
4256

    
4257
      data[os_name] = info
4258

    
4259
    # Prepare data in requested order
4260
    return [data[name] for name in self._GetNames(lu, pol.keys(), None)
4261
            if name in data]
4262

    
4263

    
4264
class LUOsDiagnose(NoHooksLU):
4265
  """Logical unit for OS diagnose/query.
4266

4267
  """
4268
  REQ_BGL = False
4269

    
4270
  @staticmethod
4271
  def _BuildFilter(fields, names):
4272
    """Builds a filter for querying OSes.
4273

4274
    """
4275
    name_filter = qlang.MakeSimpleFilter("name", names)
4276

    
4277
    # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the
4278
    # respective field is not requested
4279
    status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]
4280
                     for fname in ["hidden", "blacklisted"]
4281
                     if fname not in fields]
4282
    if "valid" not in fields:
4283
      status_filter.append([qlang.OP_TRUE, "valid"])
4284

    
4285
    if status_filter:
4286
      status_filter.insert(0, qlang.OP_AND)
4287
    else:
4288
      status_filter = None
4289

    
4290
    if name_filter and status_filter:
4291
      return [qlang.OP_AND, name_filter, status_filter]
4292
    elif name_filter:
4293
      return name_filter
4294
    else:
4295
      return status_filter
4296

    
4297
  def CheckArguments(self):
4298
    self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),
4299
                       self.op.output_fields, False)
4300

    
4301
  def ExpandNames(self):
4302
    self.oq.ExpandNames(self)
4303

    
4304
  def Exec(self, feedback_fn):
4305
    return self.oq.OldStyleQuery(self)
4306

    
4307

    
4308
class LUNodeRemove(LogicalUnit):
4309
  """Logical unit for removing a node.
4310

4311
  """
4312
  HPATH = "node-remove"
4313
  HTYPE = constants.HTYPE_NODE
4314

    
4315
  def BuildHooksEnv(self):
4316
    """Build hooks env.
4317

4318
    This doesn't run on the target node in the pre phase as a failed
4319
    node would then be impossible to remove.
4320

4321
    """
4322
    return {
4323
      "OP_TARGET": self.op.node_name,
4324
      "NODE_NAME": self.op.node_name,
4325
      }
4326

    
4327
  def BuildHooksNodes(self):
4328
    """Build hooks nodes.
4329

4330
    """
4331
    all_nodes = self.cfg.GetNodeList()
4332
    try:
4333
      all_nodes.remove(self.op.node_name)
4334
    except ValueError:
4335
      logging.warning("Node '%s', which is about to be removed, was not found"
4336
                      " in the list of all nodes", self.op.node_name)
4337
    return (all_nodes, all_nodes)
4338

    
4339
  def CheckPrereq(self):
4340
    """Check prerequisites.
4341

4342
    This checks:
4343
     - the node exists in the configuration
4344
     - it does not have primary or secondary instances
4345
     - it's not the master
4346

4347
    Any errors are signaled by raising errors.OpPrereqError.
4348

4349
    """
4350
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4351
    node = self.cfg.GetNodeInfo(self.op.node_name)
4352
    assert node is not None
4353

    
4354
    masternode = self.cfg.GetMasterNode()
4355
    if node.name == masternode:
4356
      raise errors.OpPrereqError("Node is the master node, failover to another"
4357
                                 " node is required", errors.ECODE_INVAL)
4358

    
4359
    for instance_name, instance in self.cfg.GetAllInstancesInfo().items():
4360
      if node.name in instance.all_nodes:
4361
        raise errors.OpPrereqError("Instance %s is still running on the node,"
4362
                                   " please remove first" % instance_name,
4363
                                   errors.ECODE_INVAL)
4364
    self.op.node_name = node.name
4365
    self.node = node
4366

    
4367
  def Exec(self, feedback_fn):
4368
    """Removes the node from the cluster.
4369

4370
    """
4371
    node = self.node
4372
    logging.info("Stopping the node daemon and removing configs from node %s",
4373
                 node.name)
4374

    
4375
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
4376

    
4377
    # Promote nodes to master candidate as needed
4378
    _AdjustCandidatePool(self, exceptions=[node.name])
4379
    self.context.RemoveNode(node.name)
4380

    
4381
    # Run post hooks on the node before it's removed
4382
    _RunPostHook(self, node.name)
4383

    
4384
    result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
4385
    msg = result.fail_msg
4386
    if msg:
4387
      self.LogWarning("Errors encountered on the remote node while leaving"
4388
                      " the cluster: %s", msg)
4389

    
4390
    # Remove node from our /etc/hosts
4391
    if self.cfg.GetClusterInfo().modify_etc_hosts:
4392
      master_node = self.cfg.GetMasterNode()
4393
      result = self.rpc.call_etc_hosts_modify(master_node,
4394
                                              constants.ETC_HOSTS_REMOVE,
4395
                                              node.name, None)
4396
      result.Raise("Can't update hosts file with new host data")
4397
      _RedistributeAncillaryFiles(self)
4398

    
4399

    
4400
class _NodeQuery(_QueryBase):
4401
  FIELDS = query.NODE_FIELDS
4402

    
4403
  def ExpandNames(self, lu):
4404
    lu.needed_locks = {}
4405
    lu.share_locks = _ShareAll()
4406

    
4407
    if self.names:
4408
      self.wanted = _GetWantedNodes(lu, self.names)
4409
    else:
4410
      self.wanted = locking.ALL_SET
4411

    
4412
    self.do_locking = (self.use_locking and
4413
                       query.NQ_LIVE in self.requested_data)
4414

    
4415
    if self.do_locking:
4416
      # If any non-static field is requested we need to lock the nodes
4417
      lu.needed_locks[locking.LEVEL_NODE] = self.wanted
4418

    
4419
  def DeclareLocks(self, lu, level):
4420
    pass
4421

    
4422
  def _GetQueryData(self, lu):
4423
    """Computes the list of nodes and their attributes.
4424

4425
    """
4426
    all_info = lu.cfg.GetAllNodesInfo()
4427

    
4428
    nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
4429

    
4430
    # Gather data as requested
4431
    if query.NQ_LIVE in self.requested_data:
4432
      # filter out non-vm_capable nodes
4433
      toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
4434

    
4435
      node_data = lu.rpc.call_node_info(toquery_nodes, lu.cfg.GetVGName(),
4436
                                        lu.cfg.GetHypervisorType())
4437
      live_data = dict((name, nresult.payload)
4438
                       for (name, nresult) in node_data.items()
4439
                       if not nresult.fail_msg and nresult.payload)
4440
    else:
4441
      live_data = None
4442

    
4443
    if query.NQ_INST in self.requested_data:
4444
      node_to_primary = dict([(name, set()) for name in nodenames])
4445
      node_to_secondary = dict([(name, set()) for name in nodenames])
4446

    
4447
      inst_data = lu.cfg.GetAllInstancesInfo()
4448

    
4449
      for inst in inst_data.values():
4450
        if inst.primary_node in node_to_primary:
4451
          node_to_primary[inst.primary_node].add(inst.name)
4452
        for secnode in inst.secondary_nodes:
4453
          if secnode in node_to_secondary:
4454
            node_to_secondary[secnode].add(inst.name)
4455
    else:
4456
      node_to_primary = None
4457
      node_to_secondary = None
4458

    
4459
    if query.NQ_OOB in self.requested_data:
4460
      oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
4461
                         for name, node in all_info.iteritems())
4462
    else:
4463
      oob_support = None
4464

    
4465
    if query.NQ_GROUP in self.requested_data:
4466
      groups = lu.cfg.GetAllNodeGroupsInfo()
4467
    else:
4468
      groups = {}
4469

    
4470
    return query.NodeQueryData([all_info[name] for name in nodenames],
4471
                               live_data, lu.cfg.GetMasterNode(),
4472
                               node_to_primary, node_to_secondary, groups,
4473
                               oob_support, lu.cfg.GetClusterInfo())
4474

    
4475

    
4476
class LUNodeQuery(NoHooksLU):
4477
  """Logical unit for querying nodes.
4478

4479
  """
4480
  # pylint: disable=W0142
4481
  REQ_BGL = False
4482

    
4483
  def CheckArguments(self):
4484
    self.nq = _NodeQuery(qlang.MakeSimpleFilter("name", self.op.names),
4485
                         self.op.output_fields, self.op.use_locking)
4486

    
4487
  def ExpandNames(self):
4488
    self.nq.ExpandNames(self)
4489

    
4490
  def Exec(self, feedback_fn):
4491
    return self.nq.OldStyleQuery(self)
4492

    
4493

    
4494
class LUNodeQueryvols(NoHooksLU):
4495
  """Logical unit for getting volumes on node(s).
4496

4497
  """
4498
  REQ_BGL = False
4499
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
4500
  _FIELDS_STATIC = utils.FieldSet("node")
4501

    
4502
  def CheckArguments(self):
4503
    _CheckOutputFields(static=self._FIELDS_STATIC,
4504
                       dynamic=self._FIELDS_DYNAMIC,
4505
                       selected=self.op.output_fields)
4506

    
4507
  def ExpandNames(self):
4508
    self.needed_locks = {}
4509
    self.share_locks[locking.LEVEL_NODE] = 1
4510
    if not self.op.nodes:
4511
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4512
    else:
4513
      self.needed_locks[locking.LEVEL_NODE] = \
4514
        _GetWantedNodes(self, self.op.nodes)
4515

    
4516
  def Exec(self, feedback_fn):
4517
    """Computes the list of nodes and their attributes.
4518

4519
    """
4520
    nodenames = self.owned_locks(locking.LEVEL_NODE)
4521
    volumes = self.rpc.call_node_volumes(nodenames)
4522

    
4523
    ilist = self.cfg.GetAllInstancesInfo()
4524
    vol2inst = _MapInstanceDisksToNodes(ilist.values())
4525

    
4526
    output = []
4527
    for node in nodenames:
4528
      nresult = volumes[node]
4529
      if nresult.offline:
4530
        continue
4531
      msg = nresult.fail_msg
4532
      if msg:
4533
        self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
4534
        continue
4535

    
4536
      node_vols = sorted(nresult.payload,
4537
                         key=operator.itemgetter("dev"))
4538

    
4539
      for vol in node_vols:
4540
        node_output = []
4541
        for field in self.op.output_fields:
4542
          if field == "node":
4543
            val = node
4544
          elif field == "phys":
4545
            val = vol["dev"]
4546
          elif field == "vg":
4547
            val = vol["vg"]
4548
          elif field == "name":
4549
            val = vol["name"]
4550
          elif field == "size":
4551
            val = int(float(vol["size"]))
4552
          elif field == "instance":
4553
            val = vol2inst.get((node, vol["vg"] + "/" + vol["name"]), "-")
4554
          else:
4555
            raise errors.ParameterError(field)
4556
          node_output.append(str(val))
4557

    
4558
        output.append(node_output)
4559

    
4560
    return output
4561

    
4562

    
4563
class LUNodeQueryStorage(NoHooksLU):
4564
  """Logical unit for getting information on storage units on node(s).
4565

4566
  """
4567
  _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
4568
  REQ_BGL = False
4569

    
4570
  def CheckArguments(self):
4571
    _CheckOutputFields(static=self._FIELDS_STATIC,
4572
                       dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
4573
                       selected=self.op.output_fields)
4574

    
4575
  def ExpandNames(self):
4576
    self.needed_locks = {}
4577
    self.share_locks[locking.LEVEL_NODE] = 1
4578

    
4579
    if self.op.nodes:
4580
      self.needed_locks[locking.LEVEL_NODE] = \
4581
        _GetWantedNodes(self, self.op.nodes)
4582
    else:
4583
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4584

    
4585
  def Exec(self, feedback_fn):
4586
    """Computes the list of nodes and their attributes.
4587

4588
    """
4589
    self.nodes = self.owned_locks(locking.LEVEL_NODE)
4590

    
4591
    # Always get name to sort by
4592
    if constants.SF_NAME in self.op.output_fields:
4593
      fields = self.op.output_fields[:]
4594
    else:
4595
      fields = [constants.SF_NAME] + self.op.output_fields
4596

    
4597
    # Never ask for node or type as it's only known to the LU
4598
    for extra in [constants.SF_NODE, constants.SF_TYPE]:
4599
      while extra in fields:
4600
        fields.remove(extra)
4601

    
4602
    field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
4603
    name_idx = field_idx[constants.SF_NAME]
4604

    
4605
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4606
    data = self.rpc.call_storage_list(self.nodes,
4607
                                      self.op.storage_type, st_args,
4608
                                      self.op.name, fields)
4609

    
4610
    result = []
4611

    
4612
    for node in utils.NiceSort(self.nodes):
4613
      nresult = data[node]
4614
      if nresult.offline:
4615
        continue
4616

    
4617
      msg = nresult.fail_msg
4618
      if msg:
4619
        self.LogWarning("Can't get storage data from node %s: %s", node, msg)
4620
        continue
4621

    
4622
      rows = dict([(row[name_idx], row) for row in nresult.payload])
4623

    
4624
      for name in utils.NiceSort(rows.keys()):
4625
        row = rows[name]
4626

    
4627
        out = []
4628

    
4629
        for field in self.op.output_fields:
4630
          if field == constants.SF_NODE:
4631
            val = node
4632
          elif field == constants.SF_TYPE:
4633
            val = self.op.storage_type
4634
          elif field in field_idx:
4635
            val = row[field_idx[field]]
4636
          else:
4637
            raise errors.ParameterError(field)
4638

    
4639
          out.append(val)
4640

    
4641
        result.append(out)
4642

    
4643
    return result
4644

    
4645

    
4646
class _InstanceQuery(_QueryBase):
4647
  FIELDS = query.INSTANCE_FIELDS
4648

    
4649
  def ExpandNames(self, lu):
4650
    lu.needed_locks = {}
4651
    lu.share_locks = _ShareAll()
4652

    
4653
    if self.names:
4654
      self.wanted = _GetWantedInstances(lu, self.names)
4655
    else:
4656
      self.wanted = locking.ALL_SET
4657

    
4658
    self.do_locking = (self.use_locking and
4659
                       query.IQ_LIVE in self.requested_data)
4660
    if self.do_locking:
4661
      lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
4662
      lu.needed_locks[locking.LEVEL_NODEGROUP] = []
4663
      lu.needed_locks[locking.LEVEL_NODE] = []
4664
      lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4665

    
4666
    self.do_grouplocks = (self.do_locking and
4667
                          query.IQ_NODES in self.requested_data)
4668

    
4669
  def DeclareLocks(self, lu, level):
4670
    if self.do_locking:
4671
      if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:
4672
        assert not lu.needed_locks[locking.LEVEL_NODEGROUP]
4673

    
4674
        # Lock all groups used by instances optimistically; this requires going
4675
        # via the node before it's locked, requiring verification later on
4676
        lu.needed_locks[locking.LEVEL_NODEGROUP] = \
4677
          set(group_uuid
4678
              for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
4679
              for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
4680
      elif level == locking.LEVEL_NODE:
4681
        lu._LockInstancesNodes() # pylint: disable=W0212
4682

    
4683
  @staticmethod
4684
  def _CheckGroupLocks(lu):
4685
    owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))
4686
    owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))
4687

    
4688
    # Check if node groups for locked instances are still correct
4689
    for instance_name in owned_instances:
4690
      _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
4691

    
4692
  def _GetQueryData(self, lu):
4693
    """Computes the list of instances and their attributes.
4694

4695
    """
4696
    if self.do_grouplocks:
4697
      self._CheckGroupLocks(lu)
4698

    
4699
    cluster = lu.cfg.GetClusterInfo()
4700
    all_info = lu.cfg.GetAllInstancesInfo()
4701

    
4702
    instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
4703

    
4704
    instance_list = [all_info[name] for name in instance_names]
4705
    nodes = frozenset(itertools.chain(*(inst.all_nodes
4706
                                        for inst in instance_list)))
4707
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
4708
    bad_nodes = []
4709
    offline_nodes = []
4710
    wrongnode_inst = set()
4711

    
4712
    # Gather data as requested
4713
    if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
4714
      live_data = {}
4715
      node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
4716
      for name in nodes:
4717
        result = node_data[name]
4718
        if result.offline:
4719
          # offline nodes will be in both lists
4720
          assert result.fail_msg
4721
          offline_nodes.append(name)
4722
        if result.fail_msg:
4723
          bad_nodes.append(name)
4724
        elif result.payload:
4725
          for inst in result.payload:
4726
            if inst in all_info:
4727
              if all_info[inst].primary_node == name:
4728
                live_data.update(result.payload)
4729
              else:
4730
                wrongnode_inst.add(inst)
4731
            else:
4732
              # orphan instance; we don't list it here as we don't
4733
              # handle this case yet in the output of instance listing
4734
              logging.warning("Orphan instance '%s' found on node %s",
4735
                              inst, name)
4736
        # else no instance is alive
4737
    else:
4738
      live_data = {}
4739

    
4740
    if query.IQ_DISKUSAGE in self.requested_data:
4741
      disk_usage = dict((inst.name,
4742
                         _ComputeDiskSize(inst.disk_template,
4743
                                          [{constants.IDISK_SIZE: disk.size}
4744
                                           for disk in inst.disks]))
4745
                        for inst in instance_list)
4746
    else:
4747
      disk_usage = None
4748

    
4749
    if query.IQ_CONSOLE in self.requested_data:
4750
      consinfo = {}
4751
      for inst in instance_list:
4752
        if inst.name in live_data:
4753
          # Instance is running
4754
          consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
4755
        else:
4756
          consinfo[inst.name] = None
4757
      assert set(consinfo.keys()) == set(instance_names)
4758
    else:
4759
      consinfo = None
4760

    
4761
    if query.IQ_NODES in self.requested_data:
4762
      node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),
4763
                                            instance_list)))
4764
      nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))
4765
      groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))
4766
                    for uuid in set(map(operator.attrgetter("group"),
4767
                                        nodes.values())))
4768
    else:
4769
      nodes = None
4770
      groups = None
4771

    
4772
    return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
4773
                                   disk_usage, offline_nodes, bad_nodes,
4774
                                   live_data, wrongnode_inst, consinfo,
4775
                                   nodes, groups)
4776

    
4777

    
4778
class LUQuery(NoHooksLU):
4779
  """Query for resources/items of a certain kind.
4780

4781
  """
4782
  # pylint: disable=W0142
4783
  REQ_BGL = False
4784

    
4785
  def CheckArguments(self):
4786
    qcls = _GetQueryImplementation(self.op.what)
4787

    
4788
    self.impl = qcls(self.op.filter, self.op.fields, self.op.use_locking)
4789

    
4790
  def ExpandNames(self):
4791
    self.impl.ExpandNames(self)
4792

    
4793
  def DeclareLocks(self, level):
4794
    self.impl.DeclareLocks(self, level)
4795

    
4796
  def Exec(self, feedback_fn):
4797
    return self.impl.NewStyleQuery(self)
4798

    
4799

    
4800
class LUQueryFields(NoHooksLU):
4801
  """Query for resources/items of a certain kind.
4802

4803
  """
4804
  # pylint: disable=W0142
4805
  REQ_BGL = False
4806

    
4807
  def CheckArguments(self):
4808
    self.qcls = _GetQueryImplementation(self.op.what)
4809

    
4810
  def ExpandNames(self):
4811
    self.needed_locks = {}
4812

    
4813
  def Exec(self, feedback_fn):
4814
    return query.QueryFields(self.qcls.FIELDS, self.op.fields)
4815

    
4816

    
4817
class LUNodeModifyStorage(NoHooksLU):
4818
  """Logical unit for modifying a storage volume on a node.
4819

4820
  """
4821
  REQ_BGL = False
4822

    
4823
  def CheckArguments(self):
4824
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4825

    
4826
    storage_type = self.op.storage_type
4827

    
4828
    try:
4829
      modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
4830
    except KeyError:
4831
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
4832
                                 " modified" % storage_type,
4833
                                 errors.ECODE_INVAL)
4834

    
4835
    diff = set(self.op.changes.keys()) - modifiable
4836
    if diff:
4837
      raise errors.OpPrereqError("The following fields can not be modified for"
4838
                                 " storage units of type '%s': %r" %
4839
                                 (storage_type, list(diff)),
4840
                                 errors.ECODE_INVAL)
4841

    
4842
  def ExpandNames(self):
4843
    self.needed_locks = {
4844
      locking.LEVEL_NODE: self.op.node_name,
4845
      }
4846

    
4847
  def Exec(self, feedback_fn):
4848
    """Computes the list of nodes and their attributes.
4849

4850
    """
4851
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4852
    result = self.rpc.call_storage_modify(self.op.node_name,
4853
                                          self.op.storage_type, st_args,
4854
                                          self.op.name, self.op.changes)
4855
    result.Raise("Failed to modify storage unit '%s' on %s" %
4856
                 (self.op.name, self.op.node_name))
4857

    
4858

    
4859
class LUNodeAdd(LogicalUnit):
4860
  """Logical unit for adding node to the cluster.
4861

4862
  """
4863
  HPATH = "node-add"
4864
  HTYPE = constants.HTYPE_NODE
4865
  _NFLAGS = ["master_capable", "vm_capable"]
4866

    
4867
  def CheckArguments(self):
4868
    self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
4869
    # validate/normalize the node name
4870
    self.hostname = netutils.GetHostname(name=self.op.node_name,
4871
                                         family=self.primary_ip_family)
4872
    self.op.node_name = self.hostname.name
4873

    
4874
    if self.op.readd and self.op.node_name == self.cfg.GetMasterNode():
4875
      raise errors.OpPrereqError("Cannot readd the master node",
4876
                                 errors.ECODE_STATE)
4877

    
4878
    if self.op.readd and self.op.group:
4879
      raise errors.OpPrereqError("Cannot pass a node group when a node is"
4880
                                 " being readded", errors.ECODE_INVAL)
4881

    
4882
  def BuildHooksEnv(self):
4883
    """Build hooks env.
4884

4885
    This will run on all nodes before, and on all nodes + the new node after.
4886

4887
    """
4888
    return {
4889
      "OP_TARGET": self.op.node_name,
4890
      "NODE_NAME": self.op.node_name,
4891
      "NODE_PIP": self.op.primary_ip,
4892
      "NODE_SIP": self.op.secondary_ip,
4893
      "MASTER_CAPABLE": str(self.op.master_capable),
4894
      "VM_CAPABLE": str(self.op.vm_capable),
4895
      }
4896

    
4897
  def BuildHooksNodes(self):
4898
    """Build hooks nodes.
4899

4900
    """
4901
    # Exclude added node
4902
    pre_nodes = list(set(self.cfg.GetNodeList()) - set([self.op.node_name]))
4903
    post_nodes = pre_nodes + [self.op.node_name, ]
4904

    
4905
    return (pre_nodes, post_nodes)
4906

    
4907
  def CheckPrereq(self):
4908
    """Check prerequisites.
4909

4910
    This checks:
4911
     - the new node is not already in the config
4912
     - it is resolvable
4913
     - its parameters (single/dual homed) matches the cluster
4914

4915
    Any errors are signaled by raising errors.OpPrereqError.
4916

4917
    """
4918
    cfg = self.cfg
4919
    hostname = self.hostname
4920
    node = hostname.name
4921
    primary_ip = self.op.primary_ip = hostname.ip
4922
    if self.op.secondary_ip is None:
4923
      if self.primary_ip_family == netutils.IP6Address.family:
4924
        raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
4925
                                   " IPv4 address must be given as secondary",
4926
                                   errors.ECODE_INVAL)
4927
      self.op.secondary_ip = primary_ip
4928

    
4929
    secondary_ip = self.op.secondary_ip
4930
    if not netutils.IP4Address.IsValid(secondary_ip):
4931
      raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
4932
                                 " address" % secondary_ip, errors.ECODE_INVAL)
4933

    
4934
    node_list = cfg.GetNodeList()
4935
    if not self.op.readd and node in node_list:
4936
      raise errors.OpPrereqError("Node %s is already in the configuration" %
4937
                                 node, errors.ECODE_EXISTS)
4938
    elif self.op.readd and node not in node_list:
4939
      raise errors.OpPrereqError("Node %s is not in the configuration" % node,
4940
                                 errors.ECODE_NOENT)
4941

    
4942
    self.changed_primary_ip = False
4943

    
4944
    for existing_node_name, existing_node in cfg.GetMultiNodeInfo(node_list):
4945
      if self.op.readd and node == existing_node_name:
4946
        if existing_node.secondary_ip != secondary_ip:
4947
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
4948
                                     " address configuration as before",
4949
                                     errors.ECODE_INVAL)
4950
        if existing_node.primary_ip != primary_ip:
4951
          self.changed_primary_ip = True
4952

    
4953
        continue
4954

    
4955
      if (existing_node.primary_ip == primary_ip or
4956
          existing_node.secondary_ip == primary_ip or
4957
          existing_node.primary_ip == secondary_ip or
4958
          existing_node.secondary_ip == secondary_ip):
4959
        raise errors.OpPrereqError("New node ip address(es) conflict with"
4960
                                   " existing node %s" % existing_node.name,
4961
                                   errors.ECODE_NOTUNIQUE)
4962

    
4963
    # After this 'if' block, None is no longer a valid value for the
4964
    # _capable op attributes
4965
    if self.op.readd:
4966
      old_node = self.cfg.GetNodeInfo(node)
4967
      assert old_node is not None, "Can't retrieve locked node %s" % node
4968
      for attr in self._NFLAGS:
4969
        if getattr(self.op, attr) is None:
4970
          setattr(self.op, attr, getattr(old_node, attr))
4971
    else:
4972
      for attr in self._NFLAGS:
4973
        if getattr(self.op, attr) is None:
4974
          setattr(self.op, attr, True)
4975

    
4976
    if self.op.readd and not self.op.vm_capable:
4977
      pri, sec = cfg.GetNodeInstances(node)
4978
      if pri or sec:
4979
        raise errors.OpPrereqError("Node %s being re-added with vm_capable"
4980
                                   " flag set to false, but it already holds"
4981
                                   " instances" % node,
4982
                                   errors.ECODE_STATE)
4983

    
4984
    # check that the type of the node (single versus dual homed) is the
4985
    # same as for the master
4986
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
4987
    master_singlehomed = myself.secondary_ip == myself.primary_ip
4988
    newbie_singlehomed = secondary_ip == primary_ip
4989
    if master_singlehomed != newbie_singlehomed:
4990
      if master_singlehomed:
4991
        raise errors.OpPrereqError("The master has no secondary ip but the"
4992
                                   " new node has one",
4993
                                   errors.ECODE_INVAL)
4994
      else:
4995
        raise errors.OpPrereqError("The master has a secondary ip but the"
4996
                                   " new node doesn't have one",
4997
                                   errors.ECODE_INVAL)
4998

    
4999
    # checks reachability
5000
    if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
5001
      raise errors.OpPrereqError("Node not reachable by ping",
5002
                                 errors.ECODE_ENVIRON)
5003

    
5004
    if not newbie_singlehomed:
5005
      # check reachability from my secondary ip to newbie's secondary ip
5006
      if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
5007
                           source=myself.secondary_ip):
5008
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5009
                                   " based ping to node daemon port",
5010
                                   errors.ECODE_ENVIRON)
5011

    
5012
    if self.op.readd:
5013
      exceptions = [node]
5014
    else:
5015
      exceptions = []
5016

    
5017
    if self.op.master_capable:
5018
      self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
5019
    else:
5020
      self.master_candidate = False
5021

    
5022
    if self.op.readd:
5023
      self.new_node = old_node
5024
    else:
5025
      node_group = cfg.LookupNodeGroup(self.op.group)
5026
      self.new_node = objects.Node(name=node,
5027
                                   primary_ip=primary_ip,
5028
                                   secondary_ip=secondary_ip,
5029
                                   master_candidate=self.master_candidate,
5030
                                   offline=False, drained=False,
5031
                                   group=node_group)
5032

    
5033
    if self.op.ndparams:
5034
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
5035

    
5036
    # check connectivity
5037
    result = self.rpc.call_version([self.new_node.name])[self.new_node.name]
5038
    result.Raise("Can't get version information from node %s" % node)
5039
    if constants.PROTOCOL_VERSION == result.payload:
5040
      logging.info("Communication to node %s fine, sw version %s match",
5041
                   node, result.payload)
5042
    else:
5043
      raise errors.OpPrereqError("Version mismatch master version %s,"
5044
                                 " node version %s" %
5045
                                 (constants.PROTOCOL_VERSION, result.payload),
5046
                                 errors.ECODE_ENVIRON)
5047

    
5048
  def Exec(self, feedback_fn):
5049
    """Adds the new node to the cluster.
5050

5051
    """
5052
    new_node = self.new_node
5053
    node = new_node.name
5054

    
5055
    # We adding a new node so we assume it's powered
5056
    new_node.powered = True
5057

    
5058
    # for re-adds, reset the offline/drained/master-candidate flags;
5059
    # we need to reset here, otherwise offline would prevent RPC calls
5060
    # later in the procedure; this also means that if the re-add
5061
    # fails, we are left with a non-offlined, broken node
5062
    if self.op.readd:
5063
      new_node.drained = new_node.offline = False # pylint: disable=W0201
5064
      self.LogInfo("Readding a node, the offline/drained flags were reset")
5065
      # if we demote the node, we do cleanup later in the procedure
5066
      new_node.master_candidate = self.master_candidate
5067
      if self.changed_primary_ip:
5068
        new_node.primary_ip = self.op.primary_ip
5069

    
5070
    # copy the master/vm_capable flags
5071
    for attr in self._NFLAGS:
5072
      setattr(new_node, attr, getattr(self.op, attr))
5073

    
5074
    # notify the user about any possible mc promotion
5075
    if new_node.master_candidate:
5076
      self.LogInfo("Node will be a master candidate")
5077

    
5078
    if self.op.ndparams:
5079
      new_node.ndparams = self.op.ndparams
5080
    else:
5081
      new_node.ndparams = {}
5082

    
5083
    # Add node to our /etc/hosts, and add key to known_hosts
5084
    if self.cfg.GetClusterInfo().modify_etc_hosts:
5085
      master_node = self.cfg.GetMasterNode()
5086
      result = self.rpc.call_etc_hosts_modify(master_node,
5087
                                              constants.ETC_HOSTS_ADD,
5088
                                              self.hostname.name,
5089
                                              self.hostname.ip)
5090
      result.Raise("Can't update hosts file with new host data")
5091

    
5092
    if new_node.secondary_ip != new_node.primary_ip:
5093
      _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
5094
                               False)
5095

    
5096
    node_verify_list = [self.cfg.GetMasterNode()]
5097
    node_verify_param = {
5098
      constants.NV_NODELIST: ([node], {}),
5099
      # TODO: do a node-net-test as well?
5100
    }
5101

    
5102
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
5103
                                       self.cfg.GetClusterName())
5104
    for verifier in node_verify_list:
5105
      result[verifier].Raise("Cannot communicate with node %s" % verifier)
5106
      nl_payload = result[verifier].payload[constants.NV_NODELIST]
5107
      if nl_payload:
5108
        for failed in nl_payload:
5109
          feedback_fn("ssh/hostname verification failed"
5110
                      " (checking from %s): %s" %
5111
                      (verifier, nl_payload[failed]))
5112
        raise errors.OpExecError("ssh/hostname verification failed")
5113

    
5114
    if self.op.readd:
5115
      _RedistributeAncillaryFiles(self)
5116
      self.context.ReaddNode(new_node)
5117
      # make sure we redistribute the config
5118
      self.cfg.Update(new_node, feedback_fn)
5119
      # and make sure the new node will not have old files around
5120
      if not new_node.master_candidate:
5121
        result = self.rpc.call_node_demote_from_mc(new_node.name)
5122
        msg = result.fail_msg
5123
        if msg:
5124
          self.LogWarning("Node failed to demote itself from master"
5125
                          " candidate status: %s" % msg)
5126
    else:
5127
      _RedistributeAncillaryFiles(self, additional_nodes=[node],
5128
                                  additional_vm=self.op.vm_capable)
5129
      self.context.AddNode(new_node, self.proc.GetECId())
5130

    
5131

    
5132
class LUNodeSetParams(LogicalUnit):
5133
  """Modifies the parameters of a node.
5134

5135
  @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
5136
      to the node role (as _ROLE_*)
5137
  @cvar _R2F: a dictionary from node role to tuples of flags
5138
  @cvar _FLAGS: a list of attribute names corresponding to the flags
5139

5140
  """
5141
  HPATH = "node-modify"
5142
  HTYPE = constants.HTYPE_NODE
5143
  REQ_BGL = False
5144
  (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
5145
  _F2R = {
5146
    (True, False, False): _ROLE_CANDIDATE,
5147
    (False, True, False): _ROLE_DRAINED,
5148
    (False, False, True): _ROLE_OFFLINE,
5149
    (False, False, False): _ROLE_REGULAR,
5150
    }
5151
  _R2F = dict((v, k) for k, v in _F2R.items())
5152
  _FLAGS = ["master_candidate", "drained", "offline"]
5153

    
5154
  def CheckArguments(self):
5155
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5156
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
5157
                self.op.master_capable, self.op.vm_capable,
5158
                self.op.secondary_ip, self.op.ndparams]
5159
    if all_mods.count(None) == len(all_mods):
5160
      raise errors.OpPrereqError("Please pass at least one modification",
5161
                                 errors.ECODE_INVAL)
5162
    if all_mods.count(True) > 1:
5163
      raise errors.OpPrereqError("Can't set the node into more than one"
5164
                                 " state at the same time",
5165
                                 errors.ECODE_INVAL)
5166

    
5167
    # Boolean value that tells us whether we might be demoting from MC
5168
    self.might_demote = (self.op.master_candidate == False or
5169
                         self.op.offline == True or
5170
                         self.op.drained == True or
5171
                         self.op.master_capable == False)
5172

    
5173
    if self.op.secondary_ip:
5174
      if not netutils.IP4Address.IsValid(self.op.secondary_ip):
5175
        raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
5176
                                   " address" % self.op.secondary_ip,
5177
                                   errors.ECODE_INVAL)
5178

    
5179
    self.lock_all = self.op.auto_promote and self.might_demote
5180
    self.lock_instances = self.op.secondary_ip is not None
5181

    
5182
  def ExpandNames(self):
5183
    if self.lock_all:
5184
      self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
5185
    else:
5186
      self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
5187

    
5188
    if self.lock_instances:
5189
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5190

    
5191
  def DeclareLocks(self, level):
5192
    # If we have locked all instances, before waiting to lock nodes, release
5193
    # all the ones living on nodes unrelated to the current operation.
5194
    if level == locking.LEVEL_NODE and self.lock_instances:
5195
      self.affected_instances = []
5196
      if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
5197
        instances_keep = []
5198

    
5199
        # Build list of instances to release
5200
        locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
5201
        for instance_name, instance in self.cfg.GetMultiInstanceInfo(locked_i):
5202
          if (instance.disk_template in constants.DTS_INT_MIRROR and
5203
              self.op.node_name in instance.all_nodes):
5204
            instances_keep.append(instance_name)
5205
            self.affected_instances.append(instance)
5206

    
5207
        _ReleaseLocks(self, locking.LEVEL_INSTANCE, keep=instances_keep)
5208

    
5209
        assert (set(self.owned_locks(locking.LEVEL_INSTANCE)) ==
5210
                set(instances_keep))
5211

    
5212
  def BuildHooksEnv(self):
5213
    """Build hooks env.
5214

5215
    This runs on the master node.
5216

5217
    """
5218
    return {
5219
      "OP_TARGET": self.op.node_name,
5220
      "MASTER_CANDIDATE": str(self.op.master_candidate),
5221
      "OFFLINE": str(self.op.offline),
5222
      "DRAINED": str(self.op.drained),
5223
      "MASTER_CAPABLE": str(self.op.master_capable),
5224
      "VM_CAPABLE": str(self.op.vm_capable),
5225
      }
5226

    
5227
  def BuildHooksNodes(self):
5228
    """Build hooks nodes.
5229

5230
    """
5231
    nl = [self.cfg.GetMasterNode(), self.op.node_name]
5232
    return (nl, nl)
5233

    
5234
  def CheckPrereq(self):
5235
    """Check prerequisites.
5236

5237
    This only checks the instance list against the existing names.
5238

5239
    """
5240
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
5241

    
5242
    if (self.op.master_candidate is not None or
5243
        self.op.drained is not None or
5244
        self.op.offline is not None):
5245
      # we can't change the master's node flags
5246
      if self.op.node_name == self.cfg.GetMasterNode():
5247
        raise errors.OpPrereqError("The master role can be changed"
5248
                                   " only via master-failover",
5249
                                   errors.ECODE_INVAL)
5250

    
5251
    if self.op.master_candidate and not node.master_capable:
5252
      raise errors.OpPrereqError("Node %s is not master capable, cannot make"
5253
                                 " it a master candidate" % node.name,
5254
                                 errors.ECODE_STATE)
5255

    
5256
    if self.op.vm_capable == False:
5257
      (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
5258
      if ipri or isec:
5259
        raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
5260
                                   " the vm_capable flag" % node.name,
5261
                                   errors.ECODE_STATE)
5262

    
5263
    if node.master_candidate and self.might_demote and not self.lock_all:
5264
      assert not self.op.auto_promote, "auto_promote set but lock_all not"
5265
      # check if after removing the current node, we're missing master
5266
      # candidates
5267
      (mc_remaining, mc_should, _) = \
5268
          self.cfg.GetMasterCandidateStats(exceptions=[node.name])
5269
      if mc_remaining < mc_should:
5270
        raise errors.OpPrereqError("Not enough master candidates, please"
5271
                                   " pass auto promote option to allow"
5272
                                   " promotion", errors.ECODE_STATE)
5273

    
5274
    self.old_flags = old_flags = (node.master_candidate,
5275
                                  node.drained, node.offline)
5276
    assert old_flags in self._F2R, "Un-handled old flags %s" % str(old_flags)
5277
    self.old_role = old_role = self._F2R[old_flags]
5278

    
5279
    # Check for ineffective changes
5280
    for attr in self._FLAGS:
5281
      if (getattr(self.op, attr) == False and getattr(node, attr) == False):
5282
        self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
5283
        setattr(self.op, attr, None)
5284

    
5285
    # Past this point, any flag change to False means a transition
5286
    # away from the respective state, as only real changes are kept
5287

    
5288
    # TODO: We might query the real power state if it supports OOB
5289
    if _SupportsOob(self.cfg, node):
5290
      if self.op.offline is False and not (node.powered or
5291
                                           self.op.powered == True):
5292
        raise errors.OpPrereqError(("Node %s needs to be turned on before its"
5293
                                    " offline status can be reset") %
5294
                                   self.op.node_name)
5295
    elif self.op.powered is not None:
5296
      raise errors.OpPrereqError(("Unable to change powered state for node %s"
5297
                                  " as it does not support out-of-band"
5298
                                  " handling") % self.op.node_name)
5299

    
5300
    # If we're being deofflined/drained, we'll MC ourself if needed
5301
    if (self.op.drained == False or self.op.offline == False or
5302
        (self.op.master_capable and not node.master_capable)):
5303
      if _DecideSelfPromotion(self):
5304
        self.op.master_candidate = True
5305
        self.LogInfo("Auto-promoting node to master candidate")
5306

    
5307
    # If we're no longer master capable, we'll demote ourselves from MC
5308
    if self.op.master_capable == False and node.master_candidate:
5309
      self.LogInfo("Demoting from master candidate")
5310
      self.op.master_candidate = False
5311

    
5312
    # Compute new role
5313
    assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
5314
    if self.op.master_candidate:
5315
      new_role = self._ROLE_CANDIDATE
5316
    elif self.op.drained:
5317
      new_role = self._ROLE_DRAINED
5318
    elif self.op.offline:
5319
      new_role = self._ROLE_OFFLINE
5320
    elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
5321
      # False is still in new flags, which means we're un-setting (the
5322
      # only) True flag
5323
      new_role = self._ROLE_REGULAR
5324
    else: # no new flags, nothing, keep old role
5325
      new_role = old_role
5326

    
5327
    self.new_role = new_role
5328

    
5329
    if old_role == self._ROLE_OFFLINE and new_role != old_role:
5330
      # Trying to transition out of offline status
5331
      result = self.rpc.call_version([node.name])[node.name]
5332
      if result.fail_msg:
5333
        raise errors.OpPrereqError("Node %s is being de-offlined but fails"
5334
                                   " to report its version: %s" %
5335
                                   (node.name, result.fail_msg),
5336
                                   errors.ECODE_STATE)
5337
      else:
5338
        self.LogWarning("Transitioning node from offline to online state"
5339
                        " without using re-add. Please make sure the node"
5340
                        " is healthy!")
5341

    
5342
    if self.op.secondary_ip:
5343
      # Ok even without locking, because this can't be changed by any LU
5344
      master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
5345
      master_singlehomed = master.secondary_ip == master.primary_ip
5346
      if master_singlehomed and self.op.secondary_ip:
5347
        raise errors.OpPrereqError("Cannot change the secondary ip on a single"
5348
                                   " homed cluster", errors.ECODE_INVAL)
5349

    
5350
      if node.offline:
5351
        if self.affected_instances:
5352
          raise errors.OpPrereqError("Cannot change secondary ip: offline"
5353
                                     " node has instances (%s) configured"
5354
                                     " to use it" % self.affected_instances)
5355
      else:
5356
        # On online nodes, check that no instances are running, and that
5357
        # the node has the new ip and we can reach it.
5358
        for instance in self.affected_instances:
5359
          _CheckInstanceDown(self, instance, "cannot change secondary ip")
5360

    
5361
        _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
5362
        if master.name != node.name:
5363
          # check reachability from master secondary ip to new secondary ip
5364
          if not netutils.TcpPing(self.op.secondary_ip,
5365
                                  constants.DEFAULT_NODED_PORT,
5366
                                  source=master.secondary_ip):
5367
            raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5368
                                       " based ping to node daemon port",
5369
                                       errors.ECODE_ENVIRON)
5370

    
5371
    if self.op.ndparams:
5372
      new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
5373
      utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
5374
      self.new_ndparams = new_ndparams
5375

    
5376
  def Exec(self, feedback_fn):
5377
    """Modifies a node.
5378

5379
    """
5380
    node = self.node
5381
    old_role = self.old_role
5382
    new_role = self.new_role
5383

    
5384
    result = []
5385

    
5386
    if self.op.ndparams:
5387
      node.ndparams = self.new_ndparams
5388

    
5389
    if self.op.powered is not None:
5390
      node.powered = self.op.powered
5391

    
5392
    for attr in ["master_capable", "vm_capable"]:
5393
      val = getattr(self.op, attr)
5394
      if val is not None:
5395
        setattr(node, attr, val)
5396
        result.append((attr, str(val)))
5397

    
5398
    if new_role != old_role:
5399
      # Tell the node to demote itself, if no longer MC and not offline
5400
      if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
5401
        msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
5402
        if msg:
5403
          self.LogWarning("Node failed to demote itself: %s", msg)
5404

    
5405
      new_flags = self._R2F[new_role]
5406
      for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
5407
        if of != nf:
5408
          result.append((desc, str(nf)))
5409
      (node.master_candidate, node.drained, node.offline) = new_flags
5410

    
5411
      # we locked all nodes, we adjust the CP before updating this node
5412
      if self.lock_all:
5413
        _AdjustCandidatePool(self, [node.name])
5414

    
5415
    if self.op.secondary_ip:
5416
      node.secondary_ip = self.op.secondary_ip
5417
      result.append(("secondary_ip", self.op.secondary_ip))
5418

    
5419
    # this will trigger configuration file update, if needed
5420
    self.cfg.Update(node, feedback_fn)
5421

    
5422
    # this will trigger job queue propagation or cleanup if the mc
5423
    # flag changed
5424
    if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
5425
      self.context.ReaddNode(node)
5426

    
5427
    return result
5428

    
5429

    
5430
class LUNodePowercycle(NoHooksLU):
5431
  """Powercycles a node.
5432

5433
  """
5434
  REQ_BGL = False
5435

    
5436
  def CheckArguments(self):
5437
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5438
    if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
5439
      raise errors.OpPrereqError("The node is the master and the force"
5440
                                 " parameter was not set",
5441
                                 errors.ECODE_INVAL)
5442

    
5443
  def ExpandNames(self):
5444
    """Locking for PowercycleNode.
5445

5446
    This is a last-resort option and shouldn't block on other
5447
    jobs. Therefore, we grab no locks.
5448

5449
    """
5450
    self.needed_locks = {}
5451

    
5452
  def Exec(self, feedback_fn):
5453
    """Reboots a node.
5454

5455
    """
5456
    result = self.rpc.call_node_powercycle(self.op.node_name,
5457
                                           self.cfg.GetHypervisorType())
5458
    result.Raise("Failed to schedule the reboot")
5459
    return result.payload
5460

    
5461

    
5462
class LUClusterQuery(NoHooksLU):
5463
  """Query cluster configuration.
5464

5465
  """
5466
  REQ_BGL = False
5467

    
5468
  def ExpandNames(self):
5469
    self.needed_locks = {}
5470

    
5471
  def Exec(self, feedback_fn):
5472
    """Return cluster config.
5473

5474
    """
5475
    cluster = self.cfg.GetClusterInfo()
5476
    os_hvp = {}
5477

    
5478
    # Filter just for enabled hypervisors
5479
    for os_name, hv_dict in cluster.os_hvp.items():
5480
      os_hvp[os_name] = {}
5481
      for hv_name, hv_params in hv_dict.items():
5482
        if hv_name in cluster.enabled_hypervisors:
5483
          os_hvp[os_name][hv_name] = hv_params
5484

    
5485
    # Convert ip_family to ip_version
5486
    primary_ip_version = constants.IP4_VERSION
5487
    if cluster.primary_ip_family == netutils.IP6Address.family:
5488
      primary_ip_version = constants.IP6_VERSION
5489

    
5490
    result = {
5491
      "software_version": constants.RELEASE_VERSION,
5492
      "protocol_version": constants.PROTOCOL_VERSION,
5493
      "config_version": constants.CONFIG_VERSION,
5494
      "os_api_version": max(constants.OS_API_VERSIONS),
5495
      "export_version": constants.EXPORT_VERSION,
5496
      "architecture": (platform.architecture()[0], platform.machine()),
5497
      "name": cluster.cluster_name,
5498
      "master": cluster.master_node,
5499
      "default_hypervisor": cluster.enabled_hypervisors[0],
5500
      "enabled_hypervisors": cluster.enabled_hypervisors,
5501
      "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
5502
                        for hypervisor_name in cluster.enabled_hypervisors]),
5503
      "os_hvp": os_hvp,
5504
      "beparams": cluster.beparams,
5505
      "osparams": cluster.osparams,
5506
      "nicparams": cluster.nicparams,
5507
      "ndparams": cluster.ndparams,
5508
      "candidate_pool_size": cluster.candidate_pool_size,
5509
      "master_netdev": cluster.master_netdev,
5510
      "volume_group_name": cluster.volume_group_name,
5511
      "drbd_usermode_helper": cluster.drbd_usermode_helper,
5512
      "file_storage_dir": cluster.file_storage_dir,
5513
      "shared_file_storage_dir": cluster.shared_file_storage_dir,
5514
      "maintain_node_health": cluster.maintain_node_health,
5515
      "ctime": cluster.ctime,
5516
      "mtime": cluster.mtime,
5517
      "uuid": cluster.uuid,
5518
      "tags": list(cluster.GetTags()),
5519
      "uid_pool": cluster.uid_pool,
5520
      "default_iallocator": cluster.default_iallocator,
5521
      "reserved_lvs": cluster.reserved_lvs,
5522
      "primary_ip_version": primary_ip_version,
5523
      "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
5524
      "hidden_os": cluster.hidden_os,
5525
      "blacklisted_os": cluster.blacklisted_os,
5526
      }
5527

    
5528
    return result
5529

    
5530

    
5531
class LUClusterConfigQuery(NoHooksLU):
5532
  """Return configuration values.
5533

5534
  """
5535
  REQ_BGL = False
5536
  _FIELDS_DYNAMIC = utils.FieldSet()
5537
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag",
5538
                                  "watcher_pause", "volume_group_name")
5539

    
5540
  def CheckArguments(self):
5541
    _CheckOutputFields(static=self._FIELDS_STATIC,
5542
                       dynamic=self._FIELDS_DYNAMIC,
5543
                       selected=self.op.output_fields)
5544

    
5545
  def ExpandNames(self):
5546
    self.needed_locks = {}
5547

    
5548
  def Exec(self, feedback_fn):
5549
    """Dump a representation of the cluster config to the standard output.
5550

5551
    """
5552
    values = []
5553
    for field in self.op.output_fields:
5554
      if field == "cluster_name":
5555
        entry = self.cfg.GetClusterName()
5556
      elif field == "master_node":
5557
        entry = self.cfg.GetMasterNode()
5558
      elif field == "drain_flag":
5559
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
5560
      elif field == "watcher_pause":
5561
        entry = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
5562
      elif field == "volume_group_name":
5563
        entry = self.cfg.GetVGName()
5564
      else:
5565
        raise errors.ParameterError(field)
5566
      values.append(entry)
5567
    return values
5568

    
5569

    
5570
class LUInstanceActivateDisks(NoHooksLU):
5571
  """Bring up an instance's disks.
5572

5573
  """
5574
  REQ_BGL = False
5575

    
5576
  def ExpandNames(self):
5577
    self._ExpandAndLockInstance()
5578
    self.needed_locks[locking.LEVEL_NODE] = []
5579
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5580

    
5581
  def DeclareLocks(self, level):
5582
    if level == locking.LEVEL_NODE:
5583
      self._LockInstancesNodes()
5584

    
5585
  def CheckPrereq(self):
5586
    """Check prerequisites.
5587

5588
    This checks that the instance is in the cluster.
5589

5590
    """
5591
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5592
    assert self.instance is not None, \
5593
      "Cannot retrieve locked instance %s" % self.op.instance_name
5594
    _CheckNodeOnline(self, self.instance.primary_node)
5595

    
5596
  def Exec(self, feedback_fn):
5597
    """Activate the disks.
5598

5599
    """
5600
    disks_ok, disks_info = \
5601
              _AssembleInstanceDisks(self, self.instance,
5602
                                     ignore_size=self.op.ignore_size)
5603
    if not disks_ok:
5604
      raise errors.OpExecError("Cannot activate block devices")
5605

    
5606
    return disks_info
5607

    
5608

    
5609
def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
5610
                           ignore_size=False):
5611
  """Prepare the block devices for an instance.
5612

5613
  This sets up the block devices on all nodes.
5614

5615
  @type lu: L{LogicalUnit}
5616
  @param lu: the logical unit on whose behalf we execute
5617
  @type instance: L{objects.Instance}
5618
  @param instance: the instance for whose disks we assemble
5619
  @type disks: list of L{objects.Disk} or None
5620
  @param disks: which disks to assemble (or all, if None)
5621
  @type ignore_secondaries: boolean
5622
  @param ignore_secondaries: if true, errors on secondary nodes
5623
      won't result in an error return from the function
5624
  @type ignore_size: boolean
5625
  @param ignore_size: if true, the current known size of the disk
5626
      will not be used during the disk activation, useful for cases
5627
      when the size is wrong
5628
  @return: False if the operation failed, otherwise a list of
5629
      (host, instance_visible_name, node_visible_name)
5630
      with the mapping from node devices to instance devices
5631

5632
  """
5633
  device_info = []
5634
  disks_ok = True
5635
  iname = instance.name
5636
  disks = _ExpandCheckDisks(instance, disks)
5637

    
5638
  # With the two passes mechanism we try to reduce the window of
5639
  # opportunity for the race condition of switching DRBD to primary
5640
  # before handshaking occured, but we do not eliminate it
5641

    
5642
  # The proper fix would be to wait (with some limits) until the
5643
  # connection has been made and drbd transitions from WFConnection
5644
  # into any other network-connected state (Connected, SyncTarget,
5645
  # SyncSource, etc.)
5646

    
5647
  # 1st pass, assemble on all nodes in secondary mode
5648
  for idx, inst_disk in enumerate(disks):
5649
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
5650
      if ignore_size:
5651
        node_disk = node_disk.Copy()
5652
        node_disk.UnsetSize()
5653
      lu.cfg.SetDiskID(node_disk, node)
5654
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False, idx)
5655
      msg = result.fail_msg
5656
      if msg:
5657
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
5658
                           " (is_primary=False, pass=1): %s",
5659
                           inst_disk.iv_name, node, msg)
5660
        if not ignore_secondaries:
5661
          disks_ok = False
5662

    
5663
  # FIXME: race condition on drbd migration to primary
5664

    
5665
  # 2nd pass, do only the primary node
5666
  for idx, inst_disk in enumerate(disks):
5667
    dev_path = None
5668

    
5669
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
5670
      if node != instance.primary_node:
5671
        continue
5672
      if ignore_size:
5673
        node_disk = node_disk.Copy()
5674
        node_disk.UnsetSize()
5675
      lu.cfg.SetDiskID(node_disk, node)
5676
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True, idx)
5677
      msg = result.fail_msg
5678
      if msg:
5679
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
5680
                           " (is_primary=True, pass=2): %s",
5681
                           inst_disk.iv_name, node, msg)
5682
        disks_ok = False
5683
      else:
5684
        dev_path = result.payload
5685

    
5686
    device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
5687

    
5688
  # leave the disks configured for the primary node
5689
  # this is a workaround that would be fixed better by
5690
  # improving the logical/physical id handling
5691
  for disk in disks:
5692
    lu.cfg.SetDiskID(disk, instance.primary_node)
5693

    
5694
  return disks_ok, device_info
5695

    
5696

    
5697
def _StartInstanceDisks(lu, instance, force):
5698
  """Start the disks of an instance.
5699

5700
  """
5701
  disks_ok, _ = _AssembleInstanceDisks(lu, instance,
5702
                                           ignore_secondaries=force)
5703
  if not disks_ok:
5704
    _ShutdownInstanceDisks(lu, instance)
5705
    if force is not None and not force:
5706
      lu.proc.LogWarning("", hint="If the message above refers to a"
5707
                         " secondary node,"
5708
                         " you can retry the operation using '--force'.")
5709
    raise errors.OpExecError("Disk consistency error")
5710

    
5711

    
5712
class LUInstanceDeactivateDisks(NoHooksLU):
5713
  """Shutdown an instance's disks.
5714

5715
  """
5716
  REQ_BGL = False
5717

    
5718
  def ExpandNames(self):
5719
    self._ExpandAndLockInstance()
5720
    self.needed_locks[locking.LEVEL_NODE] = []
5721
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5722

    
5723
  def DeclareLocks(self, level):
5724
    if level == locking.LEVEL_NODE:
5725
      self._LockInstancesNodes()
5726

    
5727
  def CheckPrereq(self):
5728
    """Check prerequisites.
5729

5730
    This checks that the instance is in the cluster.
5731

5732
    """
5733
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5734
    assert self.instance is not None, \
5735
      "Cannot retrieve locked instance %s" % self.op.instance_name
5736

    
5737
  def Exec(self, feedback_fn):
5738
    """Deactivate the disks
5739

5740
    """
5741
    instance = self.instance
5742
    if self.op.force:
5743
      _ShutdownInstanceDisks(self, instance)
5744
    else:
5745
      _SafeShutdownInstanceDisks(self, instance)
5746

    
5747

    
5748
def _SafeShutdownInstanceDisks(lu, instance, disks=None):
5749
  """Shutdown block devices of an instance.
5750

5751
  This function checks if an instance is running, before calling
5752
  _ShutdownInstanceDisks.
5753

5754
  """
5755
  _CheckInstanceDown(lu, instance, "cannot shutdown disks")
5756
  _ShutdownInstanceDisks(lu, instance, disks=disks)
5757

    
5758

    
5759
def _ExpandCheckDisks(instance, disks):
5760
  """Return the instance disks selected by the disks list
5761

5762
  @type disks: list of L{objects.Disk} or None
5763
  @param disks: selected disks
5764
  @rtype: list of L{objects.Disk}
5765
  @return: selected instance disks to act on
5766

5767
  """
5768
  if disks is None:
5769
    return instance.disks
5770
  else:
5771
    if not set(disks).issubset(instance.disks):
5772
      raise errors.ProgrammerError("Can only act on disks belonging to the"
5773
                                   " target instance")
5774
    return disks
5775

    
5776

    
5777
def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
5778
  """Shutdown block devices of an instance.
5779

5780
  This does the shutdown on all nodes of the instance.
5781

5782
  If the ignore_primary is false, errors on the primary node are
5783
  ignored.
5784

5785
  """
5786
  all_result = True
5787
  disks = _ExpandCheckDisks(instance, disks)
5788

    
5789
  for disk in disks:
5790
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
5791
      lu.cfg.SetDiskID(top_disk, node)
5792
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
5793
      msg = result.fail_msg
5794
      if msg:
5795
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
5796
                      disk.iv_name, node, msg)
5797
        if ((node == instance.primary_node and not ignore_primary) or
5798
            (node != instance.primary_node and not result.offline)):
5799
          all_result = False
5800
  return all_result
5801

    
5802

    
5803
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
5804
  """Checks if a node has enough free memory.
5805

5806
  This function check if a given node has the needed amount of free
5807
  memory. In case the node has less memory or we cannot get the
5808
  information from the node, this function raise an OpPrereqError
5809
  exception.
5810

5811
  @type lu: C{LogicalUnit}
5812
  @param lu: a logical unit from which we get configuration data
5813
  @type node: C{str}
5814
  @param node: the node to check
5815
  @type reason: C{str}
5816
  @param reason: string to use in the error message
5817
  @type requested: C{int}
5818
  @param requested: the amount of memory in MiB to check for
5819
  @type hypervisor_name: C{str}
5820
  @param hypervisor_name: the hypervisor to ask for memory stats
5821
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
5822
      we cannot check the node
5823

5824
  """
5825
  nodeinfo = lu.rpc.call_node_info([node], None, hypervisor_name)
5826
  nodeinfo[node].Raise("Can't get data from node %s" % node,
5827
                       prereq=True, ecode=errors.ECODE_ENVIRON)
5828
  free_mem = nodeinfo[node].payload.get("memory_free", None)
5829
  if not isinstance(free_mem, int):
5830
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
5831
                               " was '%s'" % (node, free_mem),
5832
                               errors.ECODE_ENVIRON)
5833
  if requested > free_mem:
5834
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
5835
                               " needed %s MiB, available %s MiB" %
5836
                               (node, reason, requested, free_mem),
5837
                               errors.ECODE_NORES)
5838

    
5839

    
5840
def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
5841
  """Checks if nodes have enough free disk space in the all VGs.
5842

5843
  This function check if all given nodes have the needed amount of
5844
  free disk. In case any node has less disk or we cannot get the
5845
  information from the node, this function raise an OpPrereqError
5846
  exception.
5847

5848
  @type lu: C{LogicalUnit}
5849
  @param lu: a logical unit from which we get configuration data
5850
  @type nodenames: C{list}
5851
  @param nodenames: the list of node names to check
5852
  @type req_sizes: C{dict}
5853
  @param req_sizes: the hash of vg and corresponding amount of disk in
5854
      MiB to check for
5855
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5856
      or we cannot check the node
5857

5858
  """
5859
  for vg, req_size in req_sizes.items():
5860
    _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
5861

    
5862

    
5863
def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
5864
  """Checks if nodes have enough free disk space in the specified VG.
5865

5866
  This function check if all given nodes have the needed amount of
5867
  free disk. In case any node has less disk or we cannot get the
5868
  information from the node, this function raise an OpPrereqError
5869
  exception.
5870

5871
  @type lu: C{LogicalUnit}
5872
  @param lu: a logical unit from which we get configuration data
5873
  @type nodenames: C{list}
5874
  @param nodenames: the list of node names to check
5875
  @type vg: C{str}
5876
  @param vg: the volume group to check
5877
  @type requested: C{int}
5878
  @param requested: the amount of disk in MiB to check for
5879
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5880
      or we cannot check the node
5881

5882
  """
5883
  nodeinfo = lu.rpc.call_node_info(nodenames, vg, None)
5884
  for node in nodenames:
5885
    info = nodeinfo[node]
5886
    info.Raise("Cannot get current information from node %s" % node,
5887
               prereq=True, ecode=errors.ECODE_ENVIRON)
5888
    vg_free = info.payload.get("vg_free", None)
5889
    if not isinstance(vg_free, int):
5890
      raise errors.OpPrereqError("Can't compute free disk space on node"
5891
                                 " %s for vg %s, result was '%s'" %
5892
                                 (node, vg, vg_free), errors.ECODE_ENVIRON)
5893
    if requested > vg_free:
5894
      raise errors.OpPrereqError("Not enough disk space on target node %s"
5895
                                 " vg %s: required %d MiB, available %d MiB" %
5896
                                 (node, vg, requested, vg_free),
5897
                                 errors.ECODE_NORES)
5898

    
5899

    
5900
class LUInstanceStartup(LogicalUnit):
5901
  """Starts an instance.
5902

5903
  """
5904
  HPATH = "instance-start"
5905
  HTYPE = constants.HTYPE_INSTANCE
5906
  REQ_BGL = False
5907

    
5908
  def CheckArguments(self):
5909
    # extra beparams
5910
    if self.op.beparams:
5911
      # fill the beparams dict
5912
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
5913

    
5914
  def ExpandNames(self):
5915
    self._ExpandAndLockInstance()
5916

    
5917
  def BuildHooksEnv(self):
5918
    """Build hooks env.
5919

5920
    This runs on master, primary and secondary nodes of the instance.
5921

5922
    """
5923
    env = {
5924
      "FORCE": self.op.force,
5925
      }
5926

    
5927
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5928

    
5929
    return env
5930

    
5931
  def BuildHooksNodes(self):
5932
    """Build hooks nodes.
5933

5934
    """
5935
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5936
    return (nl, nl)
5937

    
5938
  def CheckPrereq(self):
5939
    """Check prerequisites.
5940

5941
    This checks that the instance is in the cluster.
5942

5943
    """
5944
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5945
    assert self.instance is not None, \
5946
      "Cannot retrieve locked instance %s" % self.op.instance_name
5947

    
5948
    # extra hvparams
5949
    if self.op.hvparams:
5950
      # check hypervisor parameter syntax (locally)
5951
      cluster = self.cfg.GetClusterInfo()
5952
      utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
5953
      filled_hvp = cluster.FillHV(instance)
5954
      filled_hvp.update(self.op.hvparams)
5955
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
5956
      hv_type.CheckParameterSyntax(filled_hvp)
5957
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
5958

    
5959
    self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
5960

    
5961
    if self.primary_offline and self.op.ignore_offline_nodes:
5962
      self.proc.LogWarning("Ignoring offline primary node")
5963

    
5964
      if self.op.hvparams or self.op.beparams:
5965
        self.proc.LogWarning("Overridden parameters are ignored")
5966
    else:
5967
      _CheckNodeOnline(self, instance.primary_node)
5968

    
5969
      bep = self.cfg.GetClusterInfo().FillBE(instance)
5970

    
5971
      # check bridges existence
5972
      _CheckInstanceBridgesExist(self, instance)
5973

    
5974
      remote_info = self.rpc.call_instance_info(instance.primary_node,
5975
                                                instance.name,
5976
                                                instance.hypervisor)
5977
      remote_info.Raise("Error checking node %s" % instance.primary_node,
5978
                        prereq=True, ecode=errors.ECODE_ENVIRON)
5979
      if not remote_info.payload: # not running already
5980
        _CheckNodeFreeMemory(self, instance.primary_node,
5981
                             "starting instance %s" % instance.name,
5982
                             bep[constants.BE_MEMORY], instance.hypervisor)
5983

    
5984
  def Exec(self, feedback_fn):
5985
    """Start the instance.
5986

5987
    """
5988
    instance = self.instance
5989
    force = self.op.force
5990

    
5991
    if not self.op.no_remember:
5992
      self.cfg.MarkInstanceUp(instance.name)
5993

    
5994
    if self.primary_offline:
5995
      assert self.op.ignore_offline_nodes
5996
      self.proc.LogInfo("Primary node offline, marked instance as started")
5997
    else:
5998
      node_current = instance.primary_node
5999

    
6000
      _StartInstanceDisks(self, instance, force)
6001

    
6002
      result = self.rpc.call_instance_start(node_current, instance,
6003
                                            self.op.hvparams, self.op.beparams,
6004
                                            self.op.startup_paused)
6005
      msg = result.fail_msg
6006
      if msg:
6007
        _ShutdownInstanceDisks(self, instance)
6008
        raise errors.OpExecError("Could not start instance: %s" % msg)
6009

    
6010

    
6011
class LUInstanceReboot(LogicalUnit):
6012
  """Reboot an instance.
6013

6014
  """
6015
  HPATH = "instance-reboot"
6016
  HTYPE = constants.HTYPE_INSTANCE
6017
  REQ_BGL = False
6018

    
6019
  def ExpandNames(self):
6020
    self._ExpandAndLockInstance()
6021

    
6022
  def BuildHooksEnv(self):
6023
    """Build hooks env.
6024

6025
    This runs on master, primary and secondary nodes of the instance.
6026

6027
    """
6028
    env = {
6029
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
6030
      "REBOOT_TYPE": self.op.reboot_type,
6031
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6032
      }
6033

    
6034
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6035

    
6036
    return env
6037

    
6038
  def BuildHooksNodes(self):
6039
    """Build hooks nodes.
6040

6041
    """
6042
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6043
    return (nl, nl)
6044

    
6045
  def CheckPrereq(self):
6046
    """Check prerequisites.
6047

6048
    This checks that the instance is in the cluster.
6049

6050
    """
6051
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6052
    assert self.instance is not None, \
6053
      "Cannot retrieve locked instance %s" % self.op.instance_name
6054

    
6055
    _CheckNodeOnline(self, instance.primary_node)
6056

    
6057
    # check bridges existence
6058
    _CheckInstanceBridgesExist(self, instance)
6059

    
6060
  def Exec(self, feedback_fn):
6061
    """Reboot the instance.
6062

6063
    """
6064
    instance = self.instance
6065
    ignore_secondaries = self.op.ignore_secondaries
6066
    reboot_type = self.op.reboot_type
6067

    
6068
    remote_info = self.rpc.call_instance_info(instance.primary_node,
6069
                                              instance.name,
6070
                                              instance.hypervisor)
6071
    remote_info.Raise("Error checking node %s" % instance.primary_node)
6072
    instance_running = bool(remote_info.payload)
6073

    
6074
    node_current = instance.primary_node
6075

    
6076
    if instance_running and reboot_type in [constants.INSTANCE_REBOOT_SOFT,
6077
                                            constants.INSTANCE_REBOOT_HARD]:
6078
      for disk in instance.disks:
6079
        self.cfg.SetDiskID(disk, node_current)
6080
      result = self.rpc.call_instance_reboot(node_current, instance,
6081
                                             reboot_type,
6082
                                             self.op.shutdown_timeout)
6083
      result.Raise("Could not reboot instance")
6084
    else:
6085
      if instance_running:
6086
        result = self.rpc.call_instance_shutdown(node_current, instance,
6087
                                                 self.op.shutdown_timeout)
6088
        result.Raise("Could not shutdown instance for full reboot")
6089
        _ShutdownInstanceDisks(self, instance)
6090
      else:
6091
        self.LogInfo("Instance %s was already stopped, starting now",
6092
                     instance.name)
6093
      _StartInstanceDisks(self, instance, ignore_secondaries)
6094
      result = self.rpc.call_instance_start(node_current, instance,
6095
                                            None, None, False)
6096
      msg = result.fail_msg
6097
      if msg:
6098
        _ShutdownInstanceDisks(self, instance)
6099
        raise errors.OpExecError("Could not start instance for"
6100
                                 " full reboot: %s" % msg)
6101

    
6102
    self.cfg.MarkInstanceUp(instance.name)
6103

    
6104

    
6105
class LUInstanceShutdown(LogicalUnit):
6106
  """Shutdown an instance.
6107

6108
  """
6109
  HPATH = "instance-stop"
6110
  HTYPE = constants.HTYPE_INSTANCE
6111
  REQ_BGL = False
6112

    
6113
  def ExpandNames(self):
6114
    self._ExpandAndLockInstance()
6115

    
6116
  def BuildHooksEnv(self):
6117
    """Build hooks env.
6118

6119
    This runs on master, primary and secondary nodes of the instance.
6120

6121
    """
6122
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6123
    env["TIMEOUT"] = self.op.timeout
6124
    return env
6125

    
6126
  def BuildHooksNodes(self):
6127
    """Build hooks nodes.
6128

6129
    """
6130
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6131
    return (nl, nl)
6132

    
6133
  def CheckPrereq(self):
6134
    """Check prerequisites.
6135

6136
    This checks that the instance is in the cluster.
6137

6138
    """
6139
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6140
    assert self.instance is not None, \
6141
      "Cannot retrieve locked instance %s" % self.op.instance_name
6142

    
6143
    self.primary_offline = \
6144
      self.cfg.GetNodeInfo(self.instance.primary_node).offline
6145

    
6146
    if self.primary_offline and self.op.ignore_offline_nodes:
6147
      self.proc.LogWarning("Ignoring offline primary node")
6148
    else:
6149
      _CheckNodeOnline(self, self.instance.primary_node)
6150

    
6151
  def Exec(self, feedback_fn):
6152
    """Shutdown the instance.
6153

6154
    """
6155
    instance = self.instance
6156
    node_current = instance.primary_node
6157
    timeout = self.op.timeout
6158

    
6159
    if not self.op.no_remember:
6160
      self.cfg.MarkInstanceDown(instance.name)
6161

    
6162
    if self.primary_offline:
6163
      assert self.op.ignore_offline_nodes
6164
      self.proc.LogInfo("Primary node offline, marked instance as stopped")
6165
    else:
6166
      result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
6167
      msg = result.fail_msg
6168
      if msg:
6169
        self.proc.LogWarning("Could not shutdown instance: %s" % msg)
6170

    
6171
      _ShutdownInstanceDisks(self, instance)
6172

    
6173

    
6174
class LUInstanceReinstall(LogicalUnit):
6175
  """Reinstall an instance.
6176

6177
  """
6178
  HPATH = "instance-reinstall"
6179
  HTYPE = constants.HTYPE_INSTANCE
6180
  REQ_BGL = False
6181

    
6182
  def ExpandNames(self):
6183
    self._ExpandAndLockInstance()
6184

    
6185
  def BuildHooksEnv(self):
6186
    """Build hooks env.
6187

6188
    This runs on master, primary and secondary nodes of the instance.
6189

6190
    """
6191
    return _BuildInstanceHookEnvByObject(self, self.instance)
6192

    
6193
  def BuildHooksNodes(self):
6194
    """Build hooks nodes.
6195

6196
    """
6197
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6198
    return (nl, nl)
6199

    
6200
  def CheckPrereq(self):
6201
    """Check prerequisites.
6202

6203
    This checks that the instance is in the cluster and is not running.
6204

6205
    """
6206
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6207
    assert instance is not None, \
6208
      "Cannot retrieve locked instance %s" % self.op.instance_name
6209
    _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
6210
                     " offline, cannot reinstall")
6211
    for node in instance.secondary_nodes:
6212
      _CheckNodeOnline(self, node, "Instance secondary node offline,"
6213
                       " cannot reinstall")
6214

    
6215
    if instance.disk_template == constants.DT_DISKLESS:
6216
      raise errors.OpPrereqError("Instance '%s' has no disks" %
6217
                                 self.op.instance_name,
6218
                                 errors.ECODE_INVAL)
6219
    _CheckInstanceDown(self, instance, "cannot reinstall")
6220

    
6221
    if self.op.os_type is not None:
6222
      # OS verification
6223
      pnode = _ExpandNodeName(self.cfg, instance.primary_node)
6224
      _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
6225
      instance_os = self.op.os_type
6226
    else:
6227
      instance_os = instance.os
6228

    
6229
    nodelist = list(instance.all_nodes)
6230

    
6231
    if self.op.osparams:
6232
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
6233
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
6234
      self.os_inst = i_osdict # the new dict (without defaults)
6235
    else:
6236
      self.os_inst = None
6237

    
6238
    self.instance = instance
6239

    
6240
  def Exec(self, feedback_fn):
6241
    """Reinstall the instance.
6242

6243
    """
6244
    inst = self.instance
6245

    
6246
    if self.op.os_type is not None:
6247
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
6248
      inst.os = self.op.os_type
6249
      # Write to configuration
6250
      self.cfg.Update(inst, feedback_fn)
6251

    
6252
    _StartInstanceDisks(self, inst, None)
6253
    try:
6254
      feedback_fn("Running the instance OS create scripts...")
6255
      # FIXME: pass debug option from opcode to backend
6256
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True,
6257
                                             self.op.debug_level,
6258
                                             osparams=self.os_inst)
6259
      result.Raise("Could not install OS for instance %s on node %s" %
6260
                   (inst.name, inst.primary_node))
6261
    finally:
6262
      _ShutdownInstanceDisks(self, inst)
6263

    
6264

    
6265
class LUInstanceRecreateDisks(LogicalUnit):
6266
  """Recreate an instance's missing disks.
6267

6268
  """
6269
  HPATH = "instance-recreate-disks"
6270
  HTYPE = constants.HTYPE_INSTANCE
6271
  REQ_BGL = False
6272

    
6273
  def CheckArguments(self):
6274
    # normalise the disk list
6275
    self.op.disks = sorted(frozenset(self.op.disks))
6276

    
6277
  def ExpandNames(self):
6278
    self._ExpandAndLockInstance()
6279
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6280
    if self.op.nodes:
6281
      self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
6282
      self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
6283
    else:
6284
      self.needed_locks[locking.LEVEL_NODE] = []
6285

    
6286
  def DeclareLocks(self, level):
6287
    if level == locking.LEVEL_NODE:
6288
      # if we replace the nodes, we only need to lock the old primary,
6289
      # otherwise we need to lock all nodes for disk re-creation
6290
      primary_only = bool(self.op.nodes)
6291
      self._LockInstancesNodes(primary_only=primary_only)
6292

    
6293
  def BuildHooksEnv(self):
6294
    """Build hooks env.
6295

6296
    This runs on master, primary and secondary nodes of the instance.
6297

6298
    """
6299
    return _BuildInstanceHookEnvByObject(self, self.instance)
6300

    
6301
  def BuildHooksNodes(self):
6302
    """Build hooks nodes.
6303

6304
    """
6305
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6306
    return (nl, nl)
6307

    
6308
  def CheckPrereq(self):
6309
    """Check prerequisites.
6310

6311
    This checks that the instance is in the cluster and is not running.
6312

6313
    """
6314
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6315
    assert instance is not None, \
6316
      "Cannot retrieve locked instance %s" % self.op.instance_name
6317
    if self.op.nodes:
6318
      if len(self.op.nodes) != len(instance.all_nodes):
6319
        raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
6320
                                   " %d replacement nodes were specified" %
6321
                                   (instance.name, len(instance.all_nodes),
6322
                                    len(self.op.nodes)),
6323
                                   errors.ECODE_INVAL)
6324
      assert instance.disk_template != constants.DT_DRBD8 or \
6325
          len(self.op.nodes) == 2
6326
      assert instance.disk_template != constants.DT_PLAIN or \
6327
          len(self.op.nodes) == 1
6328
      primary_node = self.op.nodes[0]
6329
    else:
6330
      primary_node = instance.primary_node
6331
    _CheckNodeOnline(self, primary_node)
6332

    
6333
    if instance.disk_template == constants.DT_DISKLESS:
6334
      raise errors.OpPrereqError("Instance '%s' has no disks" %
6335
                                 self.op.instance_name, errors.ECODE_INVAL)
6336
    # if we replace nodes *and* the old primary is offline, we don't
6337
    # check
6338
    assert instance.primary_node in self.needed_locks[locking.LEVEL_NODE]
6339
    old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
6340
    if not (self.op.nodes and old_pnode.offline):
6341
      _CheckInstanceDown(self, instance, "cannot recreate disks")
6342

    
6343
    if not self.op.disks:
6344
      self.op.disks = range(len(instance.disks))
6345
    else:
6346
      for idx in self.op.disks:
6347
        if idx >= len(instance.disks):
6348
          raise errors.OpPrereqError("Invalid disk index '%s'" % idx,
6349
                                     errors.ECODE_INVAL)
6350
    if self.op.disks != range(len(instance.disks)) and self.op.nodes:
6351
      raise errors.OpPrereqError("Can't recreate disks partially and"
6352
                                 " change the nodes at the same time",
6353
                                 errors.ECODE_INVAL)
6354
    self.instance = instance
6355

    
6356
  def Exec(self, feedback_fn):
6357
    """Recreate the disks.
6358

6359
    """
6360
    instance = self.instance
6361

    
6362
    to_skip = []
6363
    mods = [] # keeps track of needed logical_id changes
6364

    
6365
    for idx, disk in enumerate(instance.disks):
6366
      if idx not in self.op.disks: # disk idx has not been passed in
6367
        to_skip.append(idx)
6368
        continue
6369
      # update secondaries for disks, if needed
6370
      if self.op.nodes:
6371
        if disk.dev_type == constants.LD_DRBD8:
6372
          # need to update the nodes and minors
6373
          assert len(self.op.nodes) == 2
6374
          assert len(disk.logical_id) == 6 # otherwise disk internals
6375
                                           # have changed
6376
          (_, _, old_port, _, _, old_secret) = disk.logical_id
6377
          new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
6378
          new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
6379
                    new_minors[0], new_minors[1], old_secret)
6380
          assert len(disk.logical_id) == len(new_id)
6381
          mods.append((idx, new_id))
6382

    
6383
    # now that we have passed all asserts above, we can apply the mods
6384
    # in a single run (to avoid partial changes)
6385
    for idx, new_id in mods:
6386
      instance.disks[idx].logical_id = new_id
6387

    
6388
    # change primary node, if needed
6389
    if self.op.nodes:
6390
      instance.primary_node = self.op.nodes[0]
6391
      self.LogWarning("Changing the instance's nodes, you will have to"
6392
                      " remove any disks left on the older nodes manually")
6393

    
6394
    if self.op.nodes:
6395
      self.cfg.Update(instance, feedback_fn)
6396

    
6397
    _CreateDisks(self, instance, to_skip=to_skip)
6398

    
6399

    
6400
class LUInstanceRename(LogicalUnit):
6401
  """Rename an instance.
6402

6403
  """
6404
  HPATH = "instance-rename"
6405
  HTYPE = constants.HTYPE_INSTANCE
6406

    
6407
  def CheckArguments(self):
6408
    """Check arguments.
6409

6410
    """
6411
    if self.op.ip_check and not self.op.name_check:
6412
      # TODO: make the ip check more flexible and not depend on the name check
6413
      raise errors.OpPrereqError("IP address check requires a name check",
6414
                                 errors.ECODE_INVAL)
6415

    
6416
  def BuildHooksEnv(self):
6417
    """Build hooks env.
6418

6419
    This runs on master, primary and secondary nodes of the instance.
6420

6421
    """
6422
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6423
    env["INSTANCE_NEW_NAME"] = self.op.new_name
6424
    return env
6425

    
6426
  def BuildHooksNodes(self):
6427
    """Build hooks nodes.
6428

6429
    """
6430
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6431
    return (nl, nl)
6432

    
6433
  def CheckPrereq(self):
6434
    """Check prerequisites.
6435

6436
    This checks that the instance is in the cluster and is not running.
6437

6438
    """
6439
    self.op.instance_name = _ExpandInstanceName(self.cfg,
6440
                                                self.op.instance_name)
6441
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6442
    assert instance is not None
6443
    _CheckNodeOnline(self, instance.primary_node)
6444
    _CheckInstanceDown(self, instance, "cannot rename")
6445
    self.instance = instance
6446

    
6447
    new_name = self.op.new_name
6448
    if self.op.name_check:
6449
      hostname = netutils.GetHostname(name=new_name)
6450
      if hostname.name != new_name:
6451
        self.LogInfo("Resolved given name '%s' to '%s'", new_name,
6452
                     hostname.name)
6453
      if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
6454
        raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
6455
                                    " same as given hostname '%s'") %
6456
                                    (hostname.name, self.op.new_name),
6457
                                    errors.ECODE_INVAL)
6458
      new_name = self.op.new_name = hostname.name
6459
      if (self.op.ip_check and
6460
          netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
6461
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
6462
                                   (hostname.ip, new_name),
6463
                                   errors.ECODE_NOTUNIQUE)
6464

    
6465
    instance_list = self.cfg.GetInstanceList()
6466
    if new_name in instance_list and new_name != instance.name:
6467
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
6468
                                 new_name, errors.ECODE_EXISTS)
6469

    
6470
  def Exec(self, feedback_fn):
6471
    """Rename the instance.
6472

6473
    """
6474
    inst = self.instance
6475
    old_name = inst.name
6476

    
6477
    rename_file_storage = False
6478
    if (inst.disk_template in constants.DTS_FILEBASED and
6479
        self.op.new_name != inst.name):
6480
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
6481
      rename_file_storage = True
6482

    
6483
    self.cfg.RenameInstance(inst.name, self.op.new_name)
6484
    # Change the instance lock. This is definitely safe while we hold the BGL.
6485
    # Otherwise the new lock would have to be added in acquired mode.
6486
    assert self.REQ_BGL
6487
    self.glm.remove(locking.LEVEL_INSTANCE, old_name)
6488
    self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
6489

    
6490
    # re-read the instance from the configuration after rename
6491
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
6492

    
6493
    if rename_file_storage:
6494
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
6495
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
6496
                                                     old_file_storage_dir,
6497
                                                     new_file_storage_dir)
6498
      result.Raise("Could not rename on node %s directory '%s' to '%s'"
6499
                   " (but the instance has been renamed in Ganeti)" %
6500
                   (inst.primary_node, old_file_storage_dir,
6501
                    new_file_storage_dir))
6502

    
6503
    _StartInstanceDisks(self, inst, None)
6504
    try:
6505
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
6506
                                                 old_name, self.op.debug_level)
6507
      msg = result.fail_msg
6508
      if msg:
6509
        msg = ("Could not run OS rename script for instance %s on node %s"
6510
               " (but the instance has been renamed in Ganeti): %s" %
6511
               (inst.name, inst.primary_node, msg))
6512
        self.proc.LogWarning(msg)
6513
    finally:
6514
      _ShutdownInstanceDisks(self, inst)
6515

    
6516
    return inst.name
6517

    
6518

    
6519
class LUInstanceRemove(LogicalUnit):
6520
  """Remove an instance.
6521

6522
  """
6523
  HPATH = "instance-remove"
6524
  HTYPE = constants.HTYPE_INSTANCE
6525
  REQ_BGL = False
6526

    
6527
  def ExpandNames(self):
6528
    self._ExpandAndLockInstance()
6529
    self.needed_locks[locking.LEVEL_NODE] = []
6530
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6531

    
6532
  def DeclareLocks(self, level):
6533
    if level == locking.LEVEL_NODE:
6534
      self._LockInstancesNodes()
6535

    
6536
  def BuildHooksEnv(self):
6537
    """Build hooks env.
6538

6539
    This runs on master, primary and secondary nodes of the instance.
6540

6541
    """
6542
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6543
    env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
6544
    return env
6545

    
6546
  def BuildHooksNodes(self):
6547
    """Build hooks nodes.
6548

6549
    """
6550
    nl = [self.cfg.GetMasterNode()]
6551
    nl_post = list(self.instance.all_nodes) + nl
6552
    return (nl, nl_post)
6553

    
6554
  def CheckPrereq(self):
6555
    """Check prerequisites.
6556

6557
    This checks that the instance is in the cluster.
6558

6559
    """
6560
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6561
    assert self.instance is not None, \
6562
      "Cannot retrieve locked instance %s" % self.op.instance_name
6563

    
6564
  def Exec(self, feedback_fn):
6565
    """Remove the instance.
6566

6567
    """
6568
    instance = self.instance
6569
    logging.info("Shutting down instance %s on node %s",
6570
                 instance.name, instance.primary_node)
6571

    
6572
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
6573
                                             self.op.shutdown_timeout)
6574
    msg = result.fail_msg
6575
    if msg:
6576
      if self.op.ignore_failures:
6577
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
6578
      else:
6579
        raise errors.OpExecError("Could not shutdown instance %s on"
6580
                                 " node %s: %s" %
6581
                                 (instance.name, instance.primary_node, msg))
6582

    
6583
    _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
6584

    
6585

    
6586
def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
6587
  """Utility function to remove an instance.
6588

6589
  """
6590
  logging.info("Removing block devices for instance %s", instance.name)
6591

    
6592
  if not _RemoveDisks(lu, instance):
6593
    if not ignore_failures:
6594
      raise errors.OpExecError("Can't remove instance's disks")
6595
    feedback_fn("Warning: can't remove instance's disks")
6596

    
6597
  logging.info("Removing instance %s out of cluster config", instance.name)
6598

    
6599
  lu.cfg.RemoveInstance(instance.name)
6600

    
6601
  assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
6602
    "Instance lock removal conflict"
6603

    
6604
  # Remove lock for the instance
6605
  lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
6606

    
6607

    
6608
class LUInstanceQuery(NoHooksLU):
6609
  """Logical unit for querying instances.
6610

6611
  """
6612
  # pylint: disable=W0142
6613
  REQ_BGL = False
6614

    
6615
  def CheckArguments(self):
6616
    self.iq = _InstanceQuery(qlang.MakeSimpleFilter("name", self.op.names),
6617
                             self.op.output_fields, self.op.use_locking)
6618

    
6619
  def ExpandNames(self):
6620
    self.iq.ExpandNames(self)
6621

    
6622
  def DeclareLocks(self, level):
6623
    self.iq.DeclareLocks(self, level)
6624

    
6625
  def Exec(self, feedback_fn):
6626
    return self.iq.OldStyleQuery(self)
6627

    
6628

    
6629
class LUInstanceFailover(LogicalUnit):
6630
  """Failover an instance.
6631

6632
  """
6633
  HPATH = "instance-failover"
6634
  HTYPE = constants.HTYPE_INSTANCE
6635
  REQ_BGL = False
6636

    
6637
  def CheckArguments(self):
6638
    """Check the arguments.
6639

6640
    """
6641
    self.iallocator = getattr(self.op, "iallocator", None)
6642
    self.target_node = getattr(self.op, "target_node", None)
6643

    
6644
  def ExpandNames(self):
6645
    self._ExpandAndLockInstance()
6646

    
6647
    if self.op.target_node is not None:
6648
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6649

    
6650
    self.needed_locks[locking.LEVEL_NODE] = []
6651
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6652

    
6653
    ignore_consistency = self.op.ignore_consistency
6654
    shutdown_timeout = self.op.shutdown_timeout
6655
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
6656
                                       cleanup=False,
6657
                                       failover=True,
6658
                                       ignore_consistency=ignore_consistency,
6659
                                       shutdown_timeout=shutdown_timeout)
6660
    self.tasklets = [self._migrater]
6661

    
6662
  def DeclareLocks(self, level):
6663
    if level == locking.LEVEL_NODE:
6664
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
6665
      if instance.disk_template in constants.DTS_EXT_MIRROR:
6666
        if self.op.target_node is None:
6667
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6668
        else:
6669
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
6670
                                                   self.op.target_node]
6671
        del self.recalculate_locks[locking.LEVEL_NODE]
6672
      else:
6673
        self._LockInstancesNodes()
6674

    
6675
  def BuildHooksEnv(self):
6676
    """Build hooks env.
6677

6678
    This runs on master, primary and secondary nodes of the instance.
6679

6680
    """
6681
    instance = self._migrater.instance
6682
    source_node = instance.primary_node
6683
    target_node = self.op.target_node
6684
    env = {
6685
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
6686
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6687
      "OLD_PRIMARY": source_node,
6688
      "NEW_PRIMARY": target_node,
6689
      }
6690

    
6691
    if instance.disk_template in constants.DTS_INT_MIRROR:
6692
      env["OLD_SECONDARY"] = instance.secondary_nodes[0]
6693
      env["NEW_SECONDARY"] = source_node
6694
    else:
6695
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = ""
6696

    
6697
    env.update(_BuildInstanceHookEnvByObject(self, instance))
6698

    
6699
    return env
6700

    
6701
  def BuildHooksNodes(self):
6702
    """Build hooks nodes.
6703

6704
    """
6705
    instance = self._migrater.instance
6706
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
6707
    return (nl, nl + [instance.primary_node])
6708

    
6709

    
6710
class LUInstanceMigrate(LogicalUnit):
6711
  """Migrate an instance.
6712

6713
  This is migration without shutting down, compared to the failover,
6714
  which is done with shutdown.
6715

6716
  """
6717
  HPATH = "instance-migrate"
6718
  HTYPE = constants.HTYPE_INSTANCE
6719
  REQ_BGL = False
6720

    
6721
  def ExpandNames(self):
6722
    self._ExpandAndLockInstance()
6723

    
6724
    if self.op.target_node is not None:
6725
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6726

    
6727
    self.needed_locks[locking.LEVEL_NODE] = []
6728
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6729

    
6730
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
6731
                                       cleanup=self.op.cleanup,
6732
                                       failover=False,
6733
                                       fallback=self.op.allow_failover)
6734
    self.tasklets = [self._migrater]
6735

    
6736
  def DeclareLocks(self, level):
6737
    if level == locking.LEVEL_NODE:
6738
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
6739
      if instance.disk_template in constants.DTS_EXT_MIRROR:
6740
        if self.op.target_node is None:
6741
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6742
        else:
6743
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
6744
                                                   self.op.target_node]
6745
        del self.recalculate_locks[locking.LEVEL_NODE]
6746
      else:
6747
        self._LockInstancesNodes()
6748

    
6749
  def BuildHooksEnv(self):
6750
    """Build hooks env.
6751

6752
    This runs on master, primary and secondary nodes of the instance.
6753

6754
    """
6755
    instance = self._migrater.instance
6756
    source_node = instance.primary_node
6757
    target_node = self.op.target_node
6758
    env = _BuildInstanceHookEnvByObject(self, instance)
6759
    env.update({
6760
      "MIGRATE_LIVE": self._migrater.live,
6761
      "MIGRATE_CLEANUP": self.op.cleanup,
6762
      "OLD_PRIMARY": source_node,
6763
      "NEW_PRIMARY": target_node,
6764
      })
6765

    
6766
    if instance.disk_template in constants.DTS_INT_MIRROR:
6767
      env["OLD_SECONDARY"] = target_node
6768
      env["NEW_SECONDARY"] = source_node
6769
    else:
6770
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
6771

    
6772
    return env
6773

    
6774
  def BuildHooksNodes(self):
6775
    """Build hooks nodes.
6776

6777
    """
6778
    instance = self._migrater.instance
6779
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
6780
    return (nl, nl + [instance.primary_node])
6781

    
6782

    
6783
class LUInstanceMove(LogicalUnit):
6784
  """Move an instance by data-copying.
6785

6786
  """
6787
  HPATH = "instance-move"
6788
  HTYPE = constants.HTYPE_INSTANCE
6789
  REQ_BGL = False
6790

    
6791
  def ExpandNames(self):
6792
    self._ExpandAndLockInstance()
6793
    target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6794
    self.op.target_node = target_node
6795
    self.needed_locks[locking.LEVEL_NODE] = [target_node]
6796
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6797

    
6798
  def DeclareLocks(self, level):
6799
    if level == locking.LEVEL_NODE:
6800
      self._LockInstancesNodes(primary_only=True)
6801

    
6802
  def BuildHooksEnv(self):
6803
    """Build hooks env.
6804

6805
    This runs on master, primary and secondary nodes of the instance.
6806

6807
    """
6808
    env = {
6809
      "TARGET_NODE": self.op.target_node,
6810
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6811
      }
6812
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6813
    return env
6814

    
6815
  def BuildHooksNodes(self):
6816
    """Build hooks nodes.
6817

6818
    """
6819
    nl = [
6820
      self.cfg.GetMasterNode(),
6821
      self.instance.primary_node,
6822
      self.op.target_node,
6823
      ]
6824
    return (nl, nl)
6825

    
6826
  def CheckPrereq(self):
6827
    """Check prerequisites.
6828

6829
    This checks that the instance is in the cluster.
6830

6831
    """
6832
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6833
    assert self.instance is not None, \
6834
      "Cannot retrieve locked instance %s" % self.op.instance_name
6835

    
6836
    node = self.cfg.GetNodeInfo(self.op.target_node)
6837
    assert node is not None, \
6838
      "Cannot retrieve locked node %s" % self.op.target_node
6839

    
6840
    self.target_node = target_node = node.name
6841

    
6842
    if target_node == instance.primary_node:
6843
      raise errors.OpPrereqError("Instance %s is already on the node %s" %
6844
                                 (instance.name, target_node),
6845
                                 errors.ECODE_STATE)
6846

    
6847
    bep = self.cfg.GetClusterInfo().FillBE(instance)
6848

    
6849
    for idx, dsk in enumerate(instance.disks):
6850
      if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
6851
        raise errors.OpPrereqError("Instance disk %d has a complex layout,"
6852
                                   " cannot copy" % idx, errors.ECODE_STATE)
6853

    
6854
    _CheckNodeOnline(self, target_node)
6855
    _CheckNodeNotDrained(self, target_node)
6856
    _CheckNodeVmCapable(self, target_node)
6857

    
6858
    if instance.admin_up:
6859
      # check memory requirements on the secondary node
6860
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
6861
                           instance.name, bep[constants.BE_MEMORY],
6862
                           instance.hypervisor)
6863
    else:
6864
      self.LogInfo("Not checking memory on the secondary node as"
6865
                   " instance will not be started")
6866

    
6867
    # check bridge existance
6868
    _CheckInstanceBridgesExist(self, instance, node=target_node)
6869

    
6870
  def Exec(self, feedback_fn):
6871
    """Move an instance.
6872

6873
    The move is done by shutting it down on its present node, copying
6874
    the data over (slow) and starting it on the new node.
6875

6876
    """
6877
    instance = self.instance
6878

    
6879
    source_node = instance.primary_node
6880
    target_node = self.target_node
6881

    
6882
    self.LogInfo("Shutting down instance %s on source node %s",
6883
                 instance.name, source_node)
6884

    
6885
    result = self.rpc.call_instance_shutdown(source_node, instance,
6886
                                             self.op.shutdown_timeout)
6887
    msg = result.fail_msg
6888
    if msg:
6889
      if self.op.ignore_consistency:
6890
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
6891
                             " Proceeding anyway. Please make sure node"
6892
                             " %s is down. Error details: %s",
6893
                             instance.name, source_node, source_node, msg)
6894
      else:
6895
        raise errors.OpExecError("Could not shutdown instance %s on"
6896
                                 " node %s: %s" %
6897
                                 (instance.name, source_node, msg))
6898

    
6899
    # create the target disks
6900
    try:
6901
      _CreateDisks(self, instance, target_node=target_node)
6902
    except errors.OpExecError:
6903
      self.LogWarning("Device creation failed, reverting...")
6904
      try:
6905
        _RemoveDisks(self, instance, target_node=target_node)
6906
      finally:
6907
        self.cfg.ReleaseDRBDMinors(instance.name)
6908
        raise
6909

    
6910
    cluster_name = self.cfg.GetClusterInfo().cluster_name
6911

    
6912
    errs = []
6913
    # activate, get path, copy the data over
6914
    for idx, disk in enumerate(instance.disks):
6915
      self.LogInfo("Copying data for disk %d", idx)
6916
      result = self.rpc.call_blockdev_assemble(target_node, disk,
6917
                                               instance.name, True, idx)
6918
      if result.fail_msg:
6919
        self.LogWarning("Can't assemble newly created disk %d: %s",
6920
                        idx, result.fail_msg)
6921
        errs.append(result.fail_msg)
6922
        break
6923
      dev_path = result.payload
6924
      result = self.rpc.call_blockdev_export(source_node, disk,
6925
                                             target_node, dev_path,
6926
                                             cluster_name)
6927
      if result.fail_msg:
6928
        self.LogWarning("Can't copy data over for disk %d: %s",
6929
                        idx, result.fail_msg)
6930
        errs.append(result.fail_msg)
6931
        break
6932

    
6933
    if errs:
6934
      self.LogWarning("Some disks failed to copy, aborting")
6935
      try:
6936
        _RemoveDisks(self, instance, target_node=target_node)
6937
      finally:
6938
        self.cfg.ReleaseDRBDMinors(instance.name)
6939
        raise errors.OpExecError("Errors during disk copy: %s" %
6940
                                 (",".join(errs),))
6941

    
6942
    instance.primary_node = target_node
6943
    self.cfg.Update(instance, feedback_fn)
6944

    
6945
    self.LogInfo("Removing the disks on the original node")
6946
    _RemoveDisks(self, instance, target_node=source_node)
6947

    
6948
    # Only start the instance if it's marked as up
6949
    if instance.admin_up:
6950
      self.LogInfo("Starting instance %s on node %s",
6951
                   instance.name, target_node)
6952

    
6953
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
6954
                                           ignore_secondaries=True)
6955
      if not disks_ok:
6956
        _ShutdownInstanceDisks(self, instance)
6957
        raise errors.OpExecError("Can't activate the instance's disks")
6958

    
6959
      result = self.rpc.call_instance_start(target_node, instance,
6960
                                            None, None, False)
6961
      msg = result.fail_msg
6962
      if msg:
6963
        _ShutdownInstanceDisks(self, instance)
6964
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
6965
                                 (instance.name, target_node, msg))
6966

    
6967

    
6968
class LUNodeMigrate(LogicalUnit):
6969
  """Migrate all instances from a node.
6970

6971
  """
6972
  HPATH = "node-migrate"
6973
  HTYPE = constants.HTYPE_NODE
6974
  REQ_BGL = False
6975

    
6976
  def CheckArguments(self):
6977
    pass
6978

    
6979
  def ExpandNames(self):
6980
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6981

    
6982
    self.share_locks = _ShareAll()
6983
    self.needed_locks = {
6984
      locking.LEVEL_NODE: [self.op.node_name],
6985
      }
6986

    
6987
  def BuildHooksEnv(self):
6988
    """Build hooks env.
6989

6990
    This runs on the master, the primary and all the secondaries.
6991

6992
    """
6993
    return {
6994
      "NODE_NAME": self.op.node_name,
6995
      }
6996

    
6997
  def BuildHooksNodes(self):
6998
    """Build hooks nodes.
6999

7000
    """
7001
    nl = [self.cfg.GetMasterNode()]
7002
    return (nl, nl)
7003

    
7004
  def CheckPrereq(self):
7005
    pass
7006

    
7007
  def Exec(self, feedback_fn):
7008
    # Prepare jobs for migration instances
7009
    jobs = [
7010
      [opcodes.OpInstanceMigrate(instance_name=inst.name,
7011
                                 mode=self.op.mode,
7012
                                 live=self.op.live,
7013
                                 iallocator=self.op.iallocator,
7014
                                 target_node=self.op.target_node)]
7015
      for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)
7016
      ]
7017

    
7018
    # TODO: Run iallocator in this opcode and pass correct placement options to
7019
    # OpInstanceMigrate. Since other jobs can modify the cluster between
7020
    # running the iallocator and the actual migration, a good consistency model
7021
    # will have to be found.
7022

    
7023
    assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
7024
            frozenset([self.op.node_name]))
7025

    
7026
    return ResultWithJobs(jobs)
7027

    
7028

    
7029
class TLMigrateInstance(Tasklet):
7030
  """Tasklet class for instance migration.
7031

7032
  @type live: boolean
7033
  @ivar live: whether the migration will be done live or non-live;
7034
      this variable is initalized only after CheckPrereq has run
7035
  @type cleanup: boolean
7036
  @ivar cleanup: Wheater we cleanup from a failed migration
7037
  @type iallocator: string
7038
  @ivar iallocator: The iallocator used to determine target_node
7039
  @type target_node: string
7040
  @ivar target_node: If given, the target_node to reallocate the instance to
7041
  @type failover: boolean
7042
  @ivar failover: Whether operation results in failover or migration
7043
  @type fallback: boolean
7044
  @ivar fallback: Whether fallback to failover is allowed if migration not
7045
                  possible
7046
  @type ignore_consistency: boolean
7047
  @ivar ignore_consistency: Wheter we should ignore consistency between source
7048
                            and target node
7049
  @type shutdown_timeout: int
7050
  @ivar shutdown_timeout: In case of failover timeout of the shutdown
7051

7052
  """
7053
  def __init__(self, lu, instance_name, cleanup=False,
7054
               failover=False, fallback=False,
7055
               ignore_consistency=False,
7056
               shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT):
7057
    """Initializes this class.
7058

7059
    """
7060
    Tasklet.__init__(self, lu)
7061

    
7062
    # Parameters
7063
    self.instance_name = instance_name
7064
    self.cleanup = cleanup
7065
    self.live = False # will be overridden later
7066
    self.failover = failover
7067
    self.fallback = fallback
7068
    self.ignore_consistency = ignore_consistency
7069
    self.shutdown_timeout = shutdown_timeout
7070

    
7071
  def CheckPrereq(self):
7072
    """Check prerequisites.
7073

7074
    This checks that the instance is in the cluster.
7075

7076
    """
7077
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
7078
    instance = self.cfg.GetInstanceInfo(instance_name)
7079
    assert instance is not None
7080
    self.instance = instance
7081

    
7082
    if (not self.cleanup and not instance.admin_up and not self.failover and
7083
        self.fallback):
7084
      self.lu.LogInfo("Instance is marked down, fallback allowed, switching"
7085
                      " to failover")
7086
      self.failover = True
7087

    
7088
    if instance.disk_template not in constants.DTS_MIRRORED:
7089
      if self.failover:
7090
        text = "failovers"
7091
      else:
7092
        text = "migrations"
7093
      raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
7094
                                 " %s" % (instance.disk_template, text),
7095
                                 errors.ECODE_STATE)
7096

    
7097
    if instance.disk_template in constants.DTS_EXT_MIRROR:
7098
      _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
7099

    
7100
      if self.lu.op.iallocator:
7101
        self._RunAllocator()
7102
      else:
7103
        # We set set self.target_node as it is required by
7104
        # BuildHooksEnv
7105
        self.target_node = self.lu.op.target_node
7106

    
7107
      # self.target_node is already populated, either directly or by the
7108
      # iallocator run
7109
      target_node = self.target_node
7110
      if self.target_node == instance.primary_node:
7111
        raise errors.OpPrereqError("Cannot migrate instance %s"
7112
                                   " to its primary (%s)" %
7113
                                   (instance.name, instance.primary_node))
7114

    
7115
      if len(self.lu.tasklets) == 1:
7116
        # It is safe to release locks only when we're the only tasklet
7117
        # in the LU
7118
        _ReleaseLocks(self.lu, locking.LEVEL_NODE,
7119
                      keep=[instance.primary_node, self.target_node])
7120

    
7121
    else:
7122
      secondary_nodes = instance.secondary_nodes
7123
      if not secondary_nodes:
7124
        raise errors.ConfigurationError("No secondary node but using"
7125
                                        " %s disk template" %
7126
                                        instance.disk_template)
7127
      target_node = secondary_nodes[0]
7128
      if self.lu.op.iallocator or (self.lu.op.target_node and
7129
                                   self.lu.op.target_node != target_node):
7130
        if self.failover:
7131
          text = "failed over"
7132
        else:
7133
          text = "migrated"
7134
        raise errors.OpPrereqError("Instances with disk template %s cannot"
7135
                                   " be %s to arbitrary nodes"
7136
                                   " (neither an iallocator nor a target"
7137
                                   " node can be passed)" %
7138
                                   (instance.disk_template, text),
7139
                                   errors.ECODE_INVAL)
7140

    
7141
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
7142

    
7143
    # check memory requirements on the secondary node
7144
    if not self.cleanup and (not self.failover or instance.admin_up):
7145
      _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
7146
                           instance.name, i_be[constants.BE_MEMORY],
7147
                           instance.hypervisor)
7148
    else:
7149
      self.lu.LogInfo("Not checking memory on the secondary node as"
7150
                      " instance will not be started")
7151

    
7152
    # check bridge existance
7153
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
7154

    
7155
    if not self.cleanup:
7156
      _CheckNodeNotDrained(self.lu, target_node)
7157
      if not self.failover:
7158
        result = self.rpc.call_instance_migratable(instance.primary_node,
7159
                                                   instance)
7160
        if result.fail_msg and self.fallback:
7161
          self.lu.LogInfo("Can't migrate, instance offline, fallback to"
7162
                          " failover")
7163
          self.failover = True
7164
        else:
7165
          result.Raise("Can't migrate, please use failover",
7166
                       prereq=True, ecode=errors.ECODE_STATE)
7167

    
7168
    assert not (self.failover and self.cleanup)
7169

    
7170
    if not self.failover:
7171
      if self.lu.op.live is not None and self.lu.op.mode is not None:
7172
        raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
7173
                                   " parameters are accepted",
7174
                                   errors.ECODE_INVAL)
7175
      if self.lu.op.live is not None:
7176
        if self.lu.op.live:
7177
          self.lu.op.mode = constants.HT_MIGRATION_LIVE
7178
        else:
7179
          self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
7180
        # reset the 'live' parameter to None so that repeated
7181
        # invocations of CheckPrereq do not raise an exception
7182
        self.lu.op.live = None
7183
      elif self.lu.op.mode is None:
7184
        # read the default value from the hypervisor
7185
        i_hv = self.cfg.GetClusterInfo().FillHV(self.instance,
7186
                                                skip_globals=False)
7187
        self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
7188

    
7189
      self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
7190
    else:
7191
      # Failover is never live
7192
      self.live = False
7193

    
7194
  def _RunAllocator(self):
7195
    """Run the allocator based on input opcode.
7196

7197
    """
7198
    ial = IAllocator(self.cfg, self.rpc,
7199
                     mode=constants.IALLOCATOR_MODE_RELOC,
7200
                     name=self.instance_name,
7201
                     # TODO See why hail breaks with a single node below
7202
                     relocate_from=[self.instance.primary_node,
7203
                                    self.instance.primary_node],
7204
                     )
7205

    
7206
    ial.Run(self.lu.op.iallocator)
7207

    
7208
    if not ial.success:
7209
      raise errors.OpPrereqError("Can't compute nodes using"
7210
                                 " iallocator '%s': %s" %
7211
                                 (self.lu.op.iallocator, ial.info),
7212
                                 errors.ECODE_NORES)
7213
    if len(ial.result) != ial.required_nodes:
7214
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
7215
                                 " of nodes (%s), required %s" %
7216
                                 (self.lu.op.iallocator, len(ial.result),
7217
                                  ial.required_nodes), errors.ECODE_FAULT)
7218
    self.target_node = ial.result[0]
7219
    self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
7220
                 self.instance_name, self.lu.op.iallocator,
7221
                 utils.CommaJoin(ial.result))
7222

    
7223
  def _WaitUntilSync(self):
7224
    """Poll with custom rpc for disk sync.
7225

7226
    This uses our own step-based rpc call.
7227

7228
    """
7229
    self.feedback_fn("* wait until resync is done")
7230
    all_done = False
7231
    while not all_done:
7232
      all_done = True
7233
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
7234
                                            self.nodes_ip,
7235
                                            self.instance.disks)
7236
      min_percent = 100
7237
      for node, nres in result.items():
7238
        nres.Raise("Cannot resync disks on node %s" % node)
7239
        node_done, node_percent = nres.payload
7240
        all_done = all_done and node_done
7241
        if node_percent is not None:
7242
          min_percent = min(min_percent, node_percent)
7243
      if not all_done:
7244
        if min_percent < 100:
7245
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
7246
        time.sleep(2)
7247

    
7248
  def _EnsureSecondary(self, node):
7249
    """Demote a node to secondary.
7250

7251
    """
7252
    self.feedback_fn("* switching node %s to secondary mode" % node)
7253

    
7254
    for dev in self.instance.disks:
7255
      self.cfg.SetDiskID(dev, node)
7256

    
7257
    result = self.rpc.call_blockdev_close(node, self.instance.name,
7258
                                          self.instance.disks)
7259
    result.Raise("Cannot change disk to secondary on node %s" % node)
7260

    
7261
  def _GoStandalone(self):
7262
    """Disconnect from the network.
7263

7264
    """
7265
    self.feedback_fn("* changing into standalone mode")
7266
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
7267
                                               self.instance.disks)
7268
    for node, nres in result.items():
7269
      nres.Raise("Cannot disconnect disks node %s" % node)
7270

    
7271
  def _GoReconnect(self, multimaster):
7272
    """Reconnect to the network.
7273

7274
    """
7275
    if multimaster:
7276
      msg = "dual-master"
7277
    else:
7278
      msg = "single-master"
7279
    self.feedback_fn("* changing disks into %s mode" % msg)
7280
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
7281
                                           self.instance.disks,
7282
                                           self.instance.name, multimaster)
7283
    for node, nres in result.items():
7284
      nres.Raise("Cannot change disks config on node %s" % node)
7285

    
7286
  def _ExecCleanup(self):
7287
    """Try to cleanup after a failed migration.
7288

7289
    The cleanup is done by:
7290
      - check that the instance is running only on one node
7291
        (and update the config if needed)
7292
      - change disks on its secondary node to secondary
7293
      - wait until disks are fully synchronized
7294
      - disconnect from the network
7295
      - change disks into single-master mode
7296
      - wait again until disks are fully synchronized
7297

7298
    """
7299
    instance = self.instance
7300
    target_node = self.target_node
7301
    source_node = self.source_node
7302

    
7303
    # check running on only one node
7304
    self.feedback_fn("* checking where the instance actually runs"
7305
                     " (if this hangs, the hypervisor might be in"
7306
                     " a bad state)")
7307
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
7308
    for node, result in ins_l.items():
7309
      result.Raise("Can't contact node %s" % node)
7310

    
7311
    runningon_source = instance.name in ins_l[source_node].payload
7312
    runningon_target = instance.name in ins_l[target_node].payload
7313

    
7314
    if runningon_source and runningon_target:
7315
      raise errors.OpExecError("Instance seems to be running on two nodes,"
7316
                               " or the hypervisor is confused; you will have"
7317
                               " to ensure manually that it runs only on one"
7318
                               " and restart this operation")
7319

    
7320
    if not (runningon_source or runningon_target):
7321
      raise errors.OpExecError("Instance does not seem to be running at all;"
7322
                               " in this case it's safer to repair by"
7323
                               " running 'gnt-instance stop' to ensure disk"
7324
                               " shutdown, and then restarting it")
7325

    
7326
    if runningon_target:
7327
      # the migration has actually succeeded, we need to update the config
7328
      self.feedback_fn("* instance running on secondary node (%s),"
7329
                       " updating config" % target_node)
7330
      instance.primary_node = target_node
7331
      self.cfg.Update(instance, self.feedback_fn)
7332
      demoted_node = source_node
7333
    else:
7334
      self.feedback_fn("* instance confirmed to be running on its"
7335
                       " primary node (%s)" % source_node)
7336
      demoted_node = target_node
7337

    
7338
    if instance.disk_template in constants.DTS_INT_MIRROR:
7339
      self._EnsureSecondary(demoted_node)
7340
      try:
7341
        self._WaitUntilSync()
7342
      except errors.OpExecError:
7343
        # we ignore here errors, since if the device is standalone, it
7344
        # won't be able to sync
7345
        pass
7346
      self._GoStandalone()
7347
      self._GoReconnect(False)
7348
      self._WaitUntilSync()
7349

    
7350
    self.feedback_fn("* done")
7351

    
7352
  def _RevertDiskStatus(self):
7353
    """Try to revert the disk status after a failed migration.
7354

7355
    """
7356
    target_node = self.target_node
7357
    if self.instance.disk_template in constants.DTS_EXT_MIRROR:
7358
      return
7359

    
7360
    try:
7361
      self._EnsureSecondary(target_node)
7362
      self._GoStandalone()
7363
      self._GoReconnect(False)
7364
      self._WaitUntilSync()
7365
    except errors.OpExecError, err:
7366
      self.lu.LogWarning("Migration failed and I can't reconnect the drives,"
7367
                         " please try to recover the instance manually;"
7368
                         " error '%s'" % str(err))
7369

    
7370
  def _AbortMigration(self):
7371
    """Call the hypervisor code to abort a started migration.
7372

7373
    """
7374
    instance = self.instance
7375
    target_node = self.target_node
7376
    migration_info = self.migration_info
7377

    
7378
    abort_result = self.rpc.call_finalize_migration(target_node,
7379
                                                    instance,
7380
                                                    migration_info,
7381
                                                    False)
7382
    abort_msg = abort_result.fail_msg
7383
    if abort_msg:
7384
      logging.error("Aborting migration failed on target node %s: %s",
7385
                    target_node, abort_msg)
7386
      # Don't raise an exception here, as we stil have to try to revert the
7387
      # disk status, even if this step failed.
7388

    
7389
  def _ExecMigration(self):
7390
    """Migrate an instance.
7391

7392
    The migrate is done by:
7393
      - change the disks into dual-master mode
7394
      - wait until disks are fully synchronized again
7395
      - migrate the instance
7396
      - change disks on the new secondary node (the old primary) to secondary
7397
      - wait until disks are fully synchronized
7398
      - change disks into single-master mode
7399

7400
    """
7401
    instance = self.instance
7402
    target_node = self.target_node
7403
    source_node = self.source_node
7404

    
7405
    self.feedback_fn("* checking disk consistency between source and target")
7406
    for dev in instance.disks:
7407
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
7408
        raise errors.OpExecError("Disk %s is degraded or not fully"
7409
                                 " synchronized on target node,"
7410
                                 " aborting migration" % dev.iv_name)
7411

    
7412
    # First get the migration information from the remote node
7413
    result = self.rpc.call_migration_info(source_node, instance)
7414
    msg = result.fail_msg
7415
    if msg:
7416
      log_err = ("Failed fetching source migration information from %s: %s" %
7417
                 (source_node, msg))
7418
      logging.error(log_err)
7419
      raise errors.OpExecError(log_err)
7420

    
7421
    self.migration_info = migration_info = result.payload
7422

    
7423
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
7424
      # Then switch the disks to master/master mode
7425
      self._EnsureSecondary(target_node)
7426
      self._GoStandalone()
7427
      self._GoReconnect(True)
7428
      self._WaitUntilSync()
7429

    
7430
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
7431
    result = self.rpc.call_accept_instance(target_node,
7432
                                           instance,
7433
                                           migration_info,
7434
                                           self.nodes_ip[target_node])
7435

    
7436
    msg = result.fail_msg
7437
    if msg:
7438
      logging.error("Instance pre-migration failed, trying to revert"
7439
                    " disk status: %s", msg)
7440
      self.feedback_fn("Pre-migration failed, aborting")
7441
      self._AbortMigration()
7442
      self._RevertDiskStatus()
7443
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
7444
                               (instance.name, msg))
7445

    
7446
    self.feedback_fn("* migrating instance to %s" % target_node)
7447
    result = self.rpc.call_instance_migrate(source_node, instance,
7448
                                            self.nodes_ip[target_node],
7449
                                            self.live)
7450
    msg = result.fail_msg
7451
    if msg:
7452
      logging.error("Instance migration failed, trying to revert"
7453
                    " disk status: %s", msg)
7454
      self.feedback_fn("Migration failed, aborting")
7455
      self._AbortMigration()
7456
      self._RevertDiskStatus()
7457
      raise errors.OpExecError("Could not migrate instance %s: %s" %
7458
                               (instance.name, msg))
7459

    
7460
    instance.primary_node = target_node
7461
    # distribute new instance config to the other nodes
7462
    self.cfg.Update(instance, self.feedback_fn)
7463

    
7464
    result = self.rpc.call_finalize_migration(target_node,
7465
                                              instance,
7466
                                              migration_info,
7467
                                              True)
7468
    msg = result.fail_msg
7469
    if msg:
7470
      logging.error("Instance migration succeeded, but finalization failed:"
7471
                    " %s", msg)
7472
      raise errors.OpExecError("Could not finalize instance migration: %s" %
7473
                               msg)
7474

    
7475
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
7476
      self._EnsureSecondary(source_node)
7477
      self._WaitUntilSync()
7478
      self._GoStandalone()
7479
      self._GoReconnect(False)
7480
      self._WaitUntilSync()
7481

    
7482
    self.feedback_fn("* done")
7483

    
7484
  def _ExecFailover(self):
7485
    """Failover an instance.
7486

7487
    The failover is done by shutting it down on its present node and
7488
    starting it on the secondary.
7489

7490
    """
7491
    instance = self.instance
7492
    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
7493

    
7494
    source_node = instance.primary_node
7495
    target_node = self.target_node
7496

    
7497
    if instance.admin_up:
7498
      self.feedback_fn("* checking disk consistency between source and target")
7499
      for dev in instance.disks:
7500
        # for drbd, these are drbd over lvm
7501
        if not _CheckDiskConsistency(self.lu, dev, target_node, False):
7502
          if primary_node.offline:
7503
            self.feedback_fn("Node %s is offline, ignoring degraded disk %s on"
7504
                             " target node %s" %
7505
                             (primary_node.name, dev.iv_name, target_node))
7506
          elif not self.ignore_consistency:
7507
            raise errors.OpExecError("Disk %s is degraded on target node,"
7508
                                     " aborting failover" % dev.iv_name)
7509
    else:
7510
      self.feedback_fn("* not checking disk consistency as instance is not"
7511
                       " running")
7512

    
7513
    self.feedback_fn("* shutting down instance on source node")
7514
    logging.info("Shutting down instance %s on node %s",
7515
                 instance.name, source_node)
7516

    
7517
    result = self.rpc.call_instance_shutdown(source_node, instance,
7518
                                             self.shutdown_timeout)
7519
    msg = result.fail_msg
7520
    if msg:
7521
      if self.ignore_consistency or primary_node.offline:
7522
        self.lu.LogWarning("Could not shutdown instance %s on node %s,"
7523
                           " proceeding anyway; please make sure node"
7524
                           " %s is down; error details: %s",
7525
                           instance.name, source_node, source_node, msg)
7526
      else:
7527
        raise errors.OpExecError("Could not shutdown instance %s on"
7528
                                 " node %s: %s" %
7529
                                 (instance.name, source_node, msg))
7530

    
7531
    self.feedback_fn("* deactivating the instance's disks on source node")
7532
    if not _ShutdownInstanceDisks(self.lu, instance, ignore_primary=True):
7533
      raise errors.OpExecError("Can't shut down the instance's disks")
7534

    
7535
    instance.primary_node = target_node
7536
    # distribute new instance config to the other nodes
7537
    self.cfg.Update(instance, self.feedback_fn)
7538

    
7539
    # Only start the instance if it's marked as up
7540
    if instance.admin_up:
7541
      self.feedback_fn("* activating the instance's disks on target node %s" %
7542
                       target_node)
7543
      logging.info("Starting instance %s on node %s",
7544
                   instance.name, target_node)
7545

    
7546
      disks_ok, _ = _AssembleInstanceDisks(self.lu, instance,
7547
                                           ignore_secondaries=True)
7548
      if not disks_ok:
7549
        _ShutdownInstanceDisks(self.lu, instance)
7550
        raise errors.OpExecError("Can't activate the instance's disks")
7551

    
7552
      self.feedback_fn("* starting the instance on the target node %s" %
7553
                       target_node)
7554
      result = self.rpc.call_instance_start(target_node, instance, None, None,
7555
                                            False)
7556
      msg = result.fail_msg
7557
      if msg:
7558
        _ShutdownInstanceDisks(self.lu, instance)
7559
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
7560
                                 (instance.name, target_node, msg))
7561

    
7562
  def Exec(self, feedback_fn):
7563
    """Perform the migration.
7564

7565
    """
7566
    self.feedback_fn = feedback_fn
7567
    self.source_node = self.instance.primary_node
7568

    
7569
    # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
7570
    if self.instance.disk_template in constants.DTS_INT_MIRROR:
7571
      self.target_node = self.instance.secondary_nodes[0]
7572
      # Otherwise self.target_node has been populated either
7573
      # directly, or through an iallocator.
7574

    
7575
    self.all_nodes = [self.source_node, self.target_node]
7576
    self.nodes_ip = dict((name, node.secondary_ip) for (name, node)
7577
                         in self.cfg.GetMultiNodeInfo(self.all_nodes))
7578

    
7579
    if self.failover:
7580
      feedback_fn("Failover instance %s" % self.instance.name)
7581
      self._ExecFailover()
7582
    else:
7583
      feedback_fn("Migrating instance %s" % self.instance.name)
7584

    
7585
      if self.cleanup:
7586
        return self._ExecCleanup()
7587
      else:
7588
        return self._ExecMigration()
7589

    
7590

    
7591
def _CreateBlockDev(lu, node, instance, device, force_create,
7592
                    info, force_open):
7593
  """Create a tree of block devices on a given node.
7594

7595
  If this device type has to be created on secondaries, create it and
7596
  all its children.
7597

7598
  If not, just recurse to children keeping the same 'force' value.
7599

7600
  @param lu: the lu on whose behalf we execute
7601
  @param node: the node on which to create the device
7602
  @type instance: L{objects.Instance}
7603
  @param instance: the instance which owns the device
7604
  @type device: L{objects.Disk}
7605
  @param device: the device to create
7606
  @type force_create: boolean
7607
  @param force_create: whether to force creation of this device; this
7608
      will be change to True whenever we find a device which has
7609
      CreateOnSecondary() attribute
7610
  @param info: the extra 'metadata' we should attach to the device
7611
      (this will be represented as a LVM tag)
7612
  @type force_open: boolean
7613
  @param force_open: this parameter will be passes to the
7614
      L{backend.BlockdevCreate} function where it specifies
7615
      whether we run on primary or not, and it affects both
7616
      the child assembly and the device own Open() execution
7617

7618
  """
7619
  if device.CreateOnSecondary():
7620
    force_create = True
7621

    
7622
  if device.children:
7623
    for child in device.children:
7624
      _CreateBlockDev(lu, node, instance, child, force_create,
7625
                      info, force_open)
7626

    
7627
  if not force_create:
7628
    return
7629

    
7630
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
7631

    
7632

    
7633
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
7634
  """Create a single block device on a given node.
7635

7636
  This will not recurse over children of the device, so they must be
7637
  created in advance.
7638

7639
  @param lu: the lu on whose behalf we execute
7640
  @param node: the node on which to create the device
7641
  @type instance: L{objects.Instance}
7642
  @param instance: the instance which owns the device
7643
  @type device: L{objects.Disk}
7644
  @param device: the device to create
7645
  @param info: the extra 'metadata' we should attach to the device
7646
      (this will be represented as a LVM tag)
7647
  @type force_open: boolean
7648
  @param force_open: this parameter will be passes to the
7649
      L{backend.BlockdevCreate} function where it specifies
7650
      whether we run on primary or not, and it affects both
7651
      the child assembly and the device own Open() execution
7652

7653
  """
7654
  lu.cfg.SetDiskID(device, node)
7655
  result = lu.rpc.call_blockdev_create(node, device, device.size,
7656
                                       instance.name, force_open, info)
7657
  result.Raise("Can't create block device %s on"
7658
               " node %s for instance %s" % (device, node, instance.name))
7659
  if device.physical_id is None:
7660
    device.physical_id = result.payload
7661

    
7662

    
7663
def _GenerateUniqueNames(lu, exts):
7664
  """Generate a suitable LV name.
7665

7666
  This will generate a logical volume name for the given instance.
7667

7668
  """
7669
  results = []
7670
  for val in exts:
7671
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
7672
    results.append("%s%s" % (new_id, val))
7673
  return results
7674

    
7675

    
7676
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
7677
                         iv_name, p_minor, s_minor):
7678
  """Generate a drbd8 device complete with its children.
7679

7680
  """
7681
  assert len(vgnames) == len(names) == 2
7682
  port = lu.cfg.AllocatePort()
7683
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
7684
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
7685
                          logical_id=(vgnames[0], names[0]))
7686
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
7687
                          logical_id=(vgnames[1], names[1]))
7688
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
7689
                          logical_id=(primary, secondary, port,
7690
                                      p_minor, s_minor,
7691
                                      shared_secret),
7692
                          children=[dev_data, dev_meta],
7693
                          iv_name=iv_name)
7694
  return drbd_dev
7695

    
7696

    
7697
def _GenerateDiskTemplate(lu, template_name,
7698
                          instance_name, primary_node,
7699
                          secondary_nodes, disk_info,
7700
                          file_storage_dir, file_driver,
7701
                          base_index, feedback_fn):
7702
  """Generate the entire disk layout for a given template type.
7703

7704
  """
7705
  #TODO: compute space requirements
7706

    
7707
  vgname = lu.cfg.GetVGName()
7708
  disk_count = len(disk_info)
7709
  disks = []
7710
  if template_name == constants.DT_DISKLESS:
7711
    pass
7712
  elif template_name == constants.DT_PLAIN:
7713
    if len(secondary_nodes) != 0:
7714
      raise errors.ProgrammerError("Wrong template configuration")
7715

    
7716
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
7717
                                      for i in range(disk_count)])
7718
    for idx, disk in enumerate(disk_info):
7719
      disk_index = idx + base_index
7720
      vg = disk.get(constants.IDISK_VG, vgname)
7721
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
7722
      disk_dev = objects.Disk(dev_type=constants.LD_LV,
7723
                              size=disk[constants.IDISK_SIZE],
7724
                              logical_id=(vg, names[idx]),
7725
                              iv_name="disk/%d" % disk_index,
7726
                              mode=disk[constants.IDISK_MODE])
7727
      disks.append(disk_dev)
7728
  elif template_name == constants.DT_DRBD8:
7729
    if len(secondary_nodes) != 1:
7730
      raise errors.ProgrammerError("Wrong template configuration")
7731
    remote_node = secondary_nodes[0]
7732
    minors = lu.cfg.AllocateDRBDMinor(
7733
      [primary_node, remote_node] * len(disk_info), instance_name)
7734

    
7735
    names = []
7736
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
7737
                                               for i in range(disk_count)]):
7738
      names.append(lv_prefix + "_data")
7739
      names.append(lv_prefix + "_meta")
7740
    for idx, disk in enumerate(disk_info):
7741
      disk_index = idx + base_index
7742
      data_vg = disk.get(constants.IDISK_VG, vgname)
7743
      meta_vg = disk.get(constants.IDISK_METAVG, data_vg)
7744
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
7745
                                      disk[constants.IDISK_SIZE],
7746
                                      [data_vg, meta_vg],
7747
                                      names[idx * 2:idx * 2 + 2],
7748
                                      "disk/%d" % disk_index,
7749
                                      minors[idx * 2], minors[idx * 2 + 1])
7750
      disk_dev.mode = disk[constants.IDISK_MODE]
7751
      disks.append(disk_dev)
7752
  elif template_name == constants.DT_FILE:
7753
    if len(secondary_nodes) != 0:
7754
      raise errors.ProgrammerError("Wrong template configuration")
7755

    
7756
    opcodes.RequireFileStorage()
7757

    
7758
    for idx, disk in enumerate(disk_info):
7759
      disk_index = idx + base_index
7760
      disk_dev = objects.Disk(dev_type=constants.LD_FILE,
7761
                              size=disk[constants.IDISK_SIZE],
7762
                              iv_name="disk/%d" % disk_index,
7763
                              logical_id=(file_driver,
7764
                                          "%s/disk%d" % (file_storage_dir,
7765
                                                         disk_index)),
7766
                              mode=disk[constants.IDISK_MODE])
7767
      disks.append(disk_dev)
7768
  elif template_name == constants.DT_SHARED_FILE:
7769
    if len(secondary_nodes) != 0:
7770
      raise errors.ProgrammerError("Wrong template configuration")
7771

    
7772
    opcodes.RequireSharedFileStorage()
7773

    
7774
    for idx, disk in enumerate(disk_info):
7775
      disk_index = idx + base_index
7776
      disk_dev = objects.Disk(dev_type=constants.LD_FILE,
7777
                              size=disk[constants.IDISK_SIZE],
7778
                              iv_name="disk/%d" % disk_index,
7779
                              logical_id=(file_driver,
7780
                                          "%s/disk%d" % (file_storage_dir,
7781
                                                         disk_index)),
7782
                              mode=disk[constants.IDISK_MODE])
7783
      disks.append(disk_dev)
7784
  elif template_name == constants.DT_BLOCK:
7785
    if len(secondary_nodes) != 0:
7786
      raise errors.ProgrammerError("Wrong template configuration")
7787

    
7788
    for idx, disk in enumerate(disk_info):
7789
      disk_index = idx + base_index
7790
      disk_dev = objects.Disk(dev_type=constants.LD_BLOCKDEV,
7791
                              size=disk[constants.IDISK_SIZE],
7792
                              logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
7793
                                          disk[constants.IDISK_ADOPT]),
7794
                              iv_name="disk/%d" % disk_index,
7795
                              mode=disk[constants.IDISK_MODE])
7796
      disks.append(disk_dev)
7797

    
7798
  else:
7799
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
7800
  return disks
7801

    
7802

    
7803
def _GetInstanceInfoText(instance):
7804
  """Compute that text that should be added to the disk's metadata.
7805

7806
  """
7807
  return "originstname+%s" % instance.name
7808

    
7809

    
7810
def _CalcEta(time_taken, written, total_size):
7811
  """Calculates the ETA based on size written and total size.
7812

7813
  @param time_taken: The time taken so far
7814
  @param written: amount written so far
7815
  @param total_size: The total size of data to be written
7816
  @return: The remaining time in seconds
7817

7818
  """
7819
  avg_time = time_taken / float(written)
7820
  return (total_size - written) * avg_time
7821

    
7822

    
7823
def _WipeDisks(lu, instance):
7824
  """Wipes instance disks.
7825

7826
  @type lu: L{LogicalUnit}
7827
  @param lu: the logical unit on whose behalf we execute
7828
  @type instance: L{objects.Instance}
7829
  @param instance: the instance whose disks we should create
7830
  @return: the success of the wipe
7831

7832
  """
7833
  node = instance.primary_node
7834

    
7835
  for device in instance.disks:
7836
    lu.cfg.SetDiskID(device, node)
7837

    
7838
  logging.info("Pause sync of instance %s disks", instance.name)
7839
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
7840

    
7841
  for idx, success in enumerate(result.payload):
7842
    if not success:
7843
      logging.warn("pause-sync of instance %s for disks %d failed",
7844
                   instance.name, idx)
7845

    
7846
  try:
7847
    for idx, device in enumerate(instance.disks):
7848
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
7849
      # MAX_WIPE_CHUNK at max
7850
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
7851
                            constants.MIN_WIPE_CHUNK_PERCENT)
7852
      # we _must_ make this an int, otherwise rounding errors will
7853
      # occur
7854
      wipe_chunk_size = int(wipe_chunk_size)
7855

    
7856
      lu.LogInfo("* Wiping disk %d", idx)
7857
      logging.info("Wiping disk %d for instance %s, node %s using"
7858
                   " chunk size %s", idx, instance.name, node, wipe_chunk_size)
7859

    
7860
      offset = 0
7861
      size = device.size
7862
      last_output = 0
7863
      start_time = time.time()
7864

    
7865
      while offset < size:
7866
        wipe_size = min(wipe_chunk_size, size - offset)
7867
        logging.debug("Wiping disk %d, offset %s, chunk %s",
7868
                      idx, offset, wipe_size)
7869
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
7870
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
7871
                     (idx, offset, wipe_size))
7872
        now = time.time()
7873
        offset += wipe_size
7874
        if now - last_output >= 60:
7875
          eta = _CalcEta(now - start_time, offset, size)
7876
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
7877
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
7878
          last_output = now
7879
  finally:
7880
    logging.info("Resume sync of instance %s disks", instance.name)
7881

    
7882
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
7883

    
7884
    for idx, success in enumerate(result.payload):
7885
      if not success:
7886
        lu.LogWarning("Resume sync of disk %d failed, please have a"
7887
                      " look at the status and troubleshoot the issue", idx)
7888
        logging.warn("resume-sync of instance %s for disks %d failed",
7889
                     instance.name, idx)
7890

    
7891

    
7892
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
7893
  """Create all disks for an instance.
7894

7895
  This abstracts away some work from AddInstance.
7896

7897
  @type lu: L{LogicalUnit}
7898
  @param lu: the logical unit on whose behalf we execute
7899
  @type instance: L{objects.Instance}
7900
  @param instance: the instance whose disks we should create
7901
  @type to_skip: list
7902
  @param to_skip: list of indices to skip
7903
  @type target_node: string
7904
  @param target_node: if passed, overrides the target node for creation
7905
  @rtype: boolean
7906
  @return: the success of the creation
7907

7908
  """
7909
  info = _GetInstanceInfoText(instance)
7910
  if target_node is None:
7911
    pnode = instance.primary_node
7912
    all_nodes = instance.all_nodes
7913
  else:
7914
    pnode = target_node
7915
    all_nodes = [pnode]
7916

    
7917
  if instance.disk_template in constants.DTS_FILEBASED:
7918
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
7919
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
7920

    
7921
    result.Raise("Failed to create directory '%s' on"
7922
                 " node %s" % (file_storage_dir, pnode))
7923

    
7924
  # Note: this needs to be kept in sync with adding of disks in
7925
  # LUInstanceSetParams
7926
  for idx, device in enumerate(instance.disks):
7927
    if to_skip and idx in to_skip:
7928
      continue
7929
    logging.info("Creating volume %s for instance %s",
7930
                 device.iv_name, instance.name)
7931
    #HARDCODE
7932
    for node in all_nodes:
7933
      f_create = node == pnode
7934
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
7935

    
7936

    
7937
def _RemoveDisks(lu, instance, target_node=None):
7938
  """Remove all disks for an instance.
7939

7940
  This abstracts away some work from `AddInstance()` and
7941
  `RemoveInstance()`. Note that in case some of the devices couldn't
7942
  be removed, the removal will continue with the other ones (compare
7943
  with `_CreateDisks()`).
7944

7945
  @type lu: L{LogicalUnit}
7946
  @param lu: the logical unit on whose behalf we execute
7947
  @type instance: L{objects.Instance}
7948
  @param instance: the instance whose disks we should remove
7949
  @type target_node: string
7950
  @param target_node: used to override the node on which to remove the disks
7951
  @rtype: boolean
7952
  @return: the success of the removal
7953

7954
  """
7955
  logging.info("Removing block devices for instance %s", instance.name)
7956

    
7957
  all_result = True
7958
  for device in instance.disks:
7959
    if target_node:
7960
      edata = [(target_node, device)]
7961
    else:
7962
      edata = device.ComputeNodeTree(instance.primary_node)
7963
    for node, disk in edata:
7964
      lu.cfg.SetDiskID(disk, node)
7965
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
7966
      if msg:
7967
        lu.LogWarning("Could not remove block device %s on node %s,"
7968
                      " continuing anyway: %s", device.iv_name, node, msg)
7969
        all_result = False
7970

    
7971
    # if this is a DRBD disk, return its port to the pool
7972
    if device.dev_type in constants.LDS_DRBD:
7973
      tcp_port = device.logical_id[2]
7974
      lu.cfg.AddTcpUdpPort(tcp_port)
7975

    
7976
  if instance.disk_template == constants.DT_FILE:
7977
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
7978
    if target_node:
7979
      tgt = target_node
7980
    else:
7981
      tgt = instance.primary_node
7982
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
7983
    if result.fail_msg:
7984
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
7985
                    file_storage_dir, instance.primary_node, result.fail_msg)
7986
      all_result = False
7987

    
7988
  return all_result
7989

    
7990

    
7991
def _ComputeDiskSizePerVG(disk_template, disks):
7992
  """Compute disk size requirements in the volume group
7993

7994
  """
7995
  def _compute(disks, payload):
7996
    """Universal algorithm.
7997

7998
    """
7999
    vgs = {}
8000
    for disk in disks:
8001
      vgs[disk[constants.IDISK_VG]] = \
8002
        vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload
8003

    
8004
    return vgs
8005

    
8006
  # Required free disk space as a function of disk and swap space
8007
  req_size_dict = {
8008
    constants.DT_DISKLESS: {},
8009
    constants.DT_PLAIN: _compute(disks, 0),
8010
    # 128 MB are added for drbd metadata for each disk
8011
    constants.DT_DRBD8: _compute(disks, 128),
8012
    constants.DT_FILE: {},
8013
    constants.DT_SHARED_FILE: {},
8014
  }
8015

    
8016
  if disk_template not in req_size_dict:
8017
    raise errors.ProgrammerError("Disk template '%s' size requirement"
8018
                                 " is unknown" % disk_template)
8019

    
8020
  return req_size_dict[disk_template]
8021

    
8022

    
8023
def _ComputeDiskSize(disk_template, disks):
8024
  """Compute disk size requirements in the volume group
8025

8026
  """
8027
  # Required free disk space as a function of disk and swap space
8028
  req_size_dict = {
8029
    constants.DT_DISKLESS: None,
8030
    constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
8031
    # 128 MB are added for drbd metadata for each disk
8032
    constants.DT_DRBD8: sum(d[constants.IDISK_SIZE] + 128 for d in disks),
8033
    constants.DT_FILE: None,
8034
    constants.DT_SHARED_FILE: 0,
8035
    constants.DT_BLOCK: 0,
8036
  }
8037

    
8038
  if disk_template not in req_size_dict:
8039
    raise errors.ProgrammerError("Disk template '%s' size requirement"
8040
                                 " is unknown" % disk_template)
8041

    
8042
  return req_size_dict[disk_template]
8043

    
8044

    
8045
def _FilterVmNodes(lu, nodenames):
8046
  """Filters out non-vm_capable nodes from a list.
8047

8048
  @type lu: L{LogicalUnit}
8049
  @param lu: the logical unit for which we check
8050
  @type nodenames: list
8051
  @param nodenames: the list of nodes on which we should check
8052
  @rtype: list
8053
  @return: the list of vm-capable nodes
8054

8055
  """
8056
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
8057
  return [name for name in nodenames if name not in vm_nodes]
8058

    
8059

    
8060
def _CheckHVParams(lu, nodenames, hvname, hvparams):
8061
  """Hypervisor parameter validation.
8062

8063
  This function abstract the hypervisor parameter validation to be
8064
  used in both instance create and instance modify.
8065

8066
  @type lu: L{LogicalUnit}
8067
  @param lu: the logical unit for which we check
8068
  @type nodenames: list
8069
  @param nodenames: the list of nodes on which we should check
8070
  @type hvname: string
8071
  @param hvname: the name of the hypervisor we should use
8072
  @type hvparams: dict
8073
  @param hvparams: the parameters which we need to check
8074
  @raise errors.OpPrereqError: if the parameters are not valid
8075

8076
  """
8077
  nodenames = _FilterVmNodes(lu, nodenames)
8078
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
8079
                                                  hvname,
8080
                                                  hvparams)
8081
  for node in nodenames:
8082
    info = hvinfo[node]
8083
    if info.offline:
8084
      continue
8085
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
8086

    
8087

    
8088
def _CheckOSParams(lu, required, nodenames, osname, osparams):
8089
  """OS parameters validation.
8090

8091
  @type lu: L{LogicalUnit}
8092
  @param lu: the logical unit for which we check
8093
  @type required: boolean
8094
  @param required: whether the validation should fail if the OS is not
8095
      found
8096
  @type nodenames: list
8097
  @param nodenames: the list of nodes on which we should check
8098
  @type osname: string
8099
  @param osname: the name of the hypervisor we should use
8100
  @type osparams: dict
8101
  @param osparams: the parameters which we need to check
8102
  @raise errors.OpPrereqError: if the parameters are not valid
8103

8104
  """
8105
  nodenames = _FilterVmNodes(lu, nodenames)
8106
  result = lu.rpc.call_os_validate(required, nodenames, osname,
8107
                                   [constants.OS_VALIDATE_PARAMETERS],
8108
                                   osparams)
8109
  for node, nres in result.items():
8110
    # we don't check for offline cases since this should be run only
8111
    # against the master node and/or an instance's nodes
8112
    nres.Raise("OS Parameters validation failed on node %s" % node)
8113
    if not nres.payload:
8114
      lu.LogInfo("OS %s not found on node %s, validation skipped",
8115
                 osname, node)
8116

    
8117

    
8118
class LUInstanceCreate(LogicalUnit):
8119
  """Create an instance.
8120

8121
  """
8122
  HPATH = "instance-add"
8123
  HTYPE = constants.HTYPE_INSTANCE
8124
  REQ_BGL = False
8125

    
8126
  def CheckArguments(self):
8127
    """Check arguments.
8128

8129
    """
8130
    # do not require name_check to ease forward/backward compatibility
8131
    # for tools
8132
    if self.op.no_install and self.op.start:
8133
      self.LogInfo("No-installation mode selected, disabling startup")
8134
      self.op.start = False
8135
    # validate/normalize the instance name
8136
    self.op.instance_name = \
8137
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
8138

    
8139
    if self.op.ip_check and not self.op.name_check:
8140
      # TODO: make the ip check more flexible and not depend on the name check
8141
      raise errors.OpPrereqError("Cannot do IP address check without a name"
8142
                                 " check", errors.ECODE_INVAL)
8143

    
8144
    # check nics' parameter names
8145
    for nic in self.op.nics:
8146
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
8147

    
8148
    # check disks. parameter names and consistent adopt/no-adopt strategy
8149
    has_adopt = has_no_adopt = False
8150
    for disk in self.op.disks:
8151
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
8152
      if constants.IDISK_ADOPT in disk:
8153
        has_adopt = True
8154
      else:
8155
        has_no_adopt = True
8156
    if has_adopt and has_no_adopt:
8157
      raise errors.OpPrereqError("Either all disks are adopted or none is",
8158
                                 errors.ECODE_INVAL)
8159
    if has_adopt:
8160
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
8161
        raise errors.OpPrereqError("Disk adoption is not supported for the"
8162
                                   " '%s' disk template" %
8163
                                   self.op.disk_template,
8164
                                   errors.ECODE_INVAL)
8165
      if self.op.iallocator is not None:
8166
        raise errors.OpPrereqError("Disk adoption not allowed with an"
8167
                                   " iallocator script", errors.ECODE_INVAL)
8168
      if self.op.mode == constants.INSTANCE_IMPORT:
8169
        raise errors.OpPrereqError("Disk adoption not allowed for"
8170
                                   " instance import", errors.ECODE_INVAL)
8171
    else:
8172
      if self.op.disk_template in constants.DTS_MUST_ADOPT:
8173
        raise errors.OpPrereqError("Disk template %s requires disk adoption,"
8174
                                   " but no 'adopt' parameter given" %
8175
                                   self.op.disk_template,
8176
                                   errors.ECODE_INVAL)
8177

    
8178
    self.adopt_disks = has_adopt
8179

    
8180
    # instance name verification
8181
    if self.op.name_check:
8182
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
8183
      self.op.instance_name = self.hostname1.name
8184
      # used in CheckPrereq for ip ping check
8185
      self.check_ip = self.hostname1.ip
8186
    else:
8187
      self.check_ip = None
8188

    
8189
    # file storage checks
8190
    if (self.op.file_driver and
8191
        not self.op.file_driver in constants.FILE_DRIVER):
8192
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
8193
                                 self.op.file_driver, errors.ECODE_INVAL)
8194

    
8195
    if self.op.disk_template == constants.DT_FILE:
8196
      opcodes.RequireFileStorage()
8197
    elif self.op.disk_template == constants.DT_SHARED_FILE:
8198
      opcodes.RequireSharedFileStorage()
8199

    
8200
    ### Node/iallocator related checks
8201
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
8202

    
8203
    if self.op.pnode is not None:
8204
      if self.op.disk_template in constants.DTS_INT_MIRROR:
8205
        if self.op.snode is None:
8206
          raise errors.OpPrereqError("The networked disk templates need"
8207
                                     " a mirror node", errors.ECODE_INVAL)
8208
      elif self.op.snode:
8209
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
8210
                        " template")
8211
        self.op.snode = None
8212

    
8213
    self._cds = _GetClusterDomainSecret()
8214

    
8215
    if self.op.mode == constants.INSTANCE_IMPORT:
8216
      # On import force_variant must be True, because if we forced it at
8217
      # initial install, our only chance when importing it back is that it
8218
      # works again!
8219
      self.op.force_variant = True
8220

    
8221
      if self.op.no_install:
8222
        self.LogInfo("No-installation mode has no effect during import")
8223

    
8224
    elif self.op.mode == constants.INSTANCE_CREATE:
8225
      if self.op.os_type is None:
8226
        raise errors.OpPrereqError("No guest OS specified",
8227
                                   errors.ECODE_INVAL)
8228
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
8229
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
8230
                                   " installation" % self.op.os_type,
8231
                                   errors.ECODE_STATE)
8232
      if self.op.disk_template is None:
8233
        raise errors.OpPrereqError("No disk template specified",
8234
                                   errors.ECODE_INVAL)
8235

    
8236
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
8237
      # Check handshake to ensure both clusters have the same domain secret
8238
      src_handshake = self.op.source_handshake
8239
      if not src_handshake:
8240
        raise errors.OpPrereqError("Missing source handshake",
8241
                                   errors.ECODE_INVAL)
8242

    
8243
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
8244
                                                           src_handshake)
8245
      if errmsg:
8246
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
8247
                                   errors.ECODE_INVAL)
8248

    
8249
      # Load and check source CA
8250
      self.source_x509_ca_pem = self.op.source_x509_ca
8251
      if not self.source_x509_ca_pem:
8252
        raise errors.OpPrereqError("Missing source X509 CA",
8253
                                   errors.ECODE_INVAL)
8254

    
8255
      try:
8256
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
8257
                                                    self._cds)
8258
      except OpenSSL.crypto.Error, err:
8259
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
8260
                                   (err, ), errors.ECODE_INVAL)
8261

    
8262
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
8263
      if errcode is not None:
8264
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
8265
                                   errors.ECODE_INVAL)
8266

    
8267
      self.source_x509_ca = cert
8268

    
8269
      src_instance_name = self.op.source_instance_name
8270
      if not src_instance_name:
8271
        raise errors.OpPrereqError("Missing source instance name",
8272
                                   errors.ECODE_INVAL)
8273

    
8274
      self.source_instance_name = \
8275
          netutils.GetHostname(name=src_instance_name).name
8276

    
8277
    else:
8278
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
8279
                                 self.op.mode, errors.ECODE_INVAL)
8280

    
8281
  def ExpandNames(self):
8282
    """ExpandNames for CreateInstance.
8283

8284
    Figure out the right locks for instance creation.
8285

8286
    """
8287
    self.needed_locks = {}
8288

    
8289
    instance_name = self.op.instance_name
8290
    # this is just a preventive check, but someone might still add this
8291
    # instance in the meantime, and creation will fail at lock-add time
8292
    if instance_name in self.cfg.GetInstanceList():
8293
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
8294
                                 instance_name, errors.ECODE_EXISTS)
8295

    
8296
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
8297

    
8298
    if self.op.iallocator:
8299
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8300
    else:
8301
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
8302
      nodelist = [self.op.pnode]
8303
      if self.op.snode is not None:
8304
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
8305
        nodelist.append(self.op.snode)
8306
      self.needed_locks[locking.LEVEL_NODE] = nodelist
8307

    
8308
    # in case of import lock the source node too
8309
    if self.op.mode == constants.INSTANCE_IMPORT:
8310
      src_node = self.op.src_node
8311
      src_path = self.op.src_path
8312

    
8313
      if src_path is None:
8314
        self.op.src_path = src_path = self.op.instance_name
8315

    
8316
      if src_node is None:
8317
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8318
        self.op.src_node = None
8319
        if os.path.isabs(src_path):
8320
          raise errors.OpPrereqError("Importing an instance from a path"
8321
                                     " requires a source node option",
8322
                                     errors.ECODE_INVAL)
8323
      else:
8324
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
8325
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
8326
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
8327
        if not os.path.isabs(src_path):
8328
          self.op.src_path = src_path = \
8329
            utils.PathJoin(constants.EXPORT_DIR, src_path)
8330

    
8331
  def _RunAllocator(self):
8332
    """Run the allocator based on input opcode.
8333

8334
    """
8335
    nics = [n.ToDict() for n in self.nics]
8336
    ial = IAllocator(self.cfg, self.rpc,
8337
                     mode=constants.IALLOCATOR_MODE_ALLOC,
8338
                     name=self.op.instance_name,
8339
                     disk_template=self.op.disk_template,
8340
                     tags=self.op.tags,
8341
                     os=self.op.os_type,
8342
                     vcpus=self.be_full[constants.BE_VCPUS],
8343
                     memory=self.be_full[constants.BE_MEMORY],
8344
                     disks=self.disks,
8345
                     nics=nics,
8346
                     hypervisor=self.op.hypervisor,
8347
                     )
8348

    
8349
    ial.Run(self.op.iallocator)
8350

    
8351
    if not ial.success:
8352
      raise errors.OpPrereqError("Can't compute nodes using"
8353
                                 " iallocator '%s': %s" %
8354
                                 (self.op.iallocator, ial.info),
8355
                                 errors.ECODE_NORES)
8356
    if len(ial.result) != ial.required_nodes:
8357
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8358
                                 " of nodes (%s), required %s" %
8359
                                 (self.op.iallocator, len(ial.result),
8360
                                  ial.required_nodes), errors.ECODE_FAULT)
8361
    self.op.pnode = ial.result[0]
8362
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
8363
                 self.op.instance_name, self.op.iallocator,
8364
                 utils.CommaJoin(ial.result))
8365
    if ial.required_nodes == 2:
8366
      self.op.snode = ial.result[1]
8367

    
8368
  def BuildHooksEnv(self):
8369
    """Build hooks env.
8370

8371
    This runs on master, primary and secondary nodes of the instance.
8372

8373
    """
8374
    env = {
8375
      "ADD_MODE": self.op.mode,
8376
      }
8377
    if self.op.mode == constants.INSTANCE_IMPORT:
8378
      env["SRC_NODE"] = self.op.src_node
8379
      env["SRC_PATH"] = self.op.src_path
8380
      env["SRC_IMAGES"] = self.src_images
8381

    
8382
    env.update(_BuildInstanceHookEnv(
8383
      name=self.op.instance_name,
8384
      primary_node=self.op.pnode,
8385
      secondary_nodes=self.secondaries,
8386
      status=self.op.start,
8387
      os_type=self.op.os_type,
8388
      memory=self.be_full[constants.BE_MEMORY],
8389
      vcpus=self.be_full[constants.BE_VCPUS],
8390
      nics=_NICListToTuple(self, self.nics),
8391
      disk_template=self.op.disk_template,
8392
      disks=[(d[constants.IDISK_SIZE], d[constants.IDISK_MODE])
8393
             for d in self.disks],
8394
      bep=self.be_full,
8395
      hvp=self.hv_full,
8396
      hypervisor_name=self.op.hypervisor,
8397
      tags=self.op.tags,
8398
    ))
8399

    
8400
    return env
8401

    
8402
  def BuildHooksNodes(self):
8403
    """Build hooks nodes.
8404

8405
    """
8406
    nl = [self.cfg.GetMasterNode(), self.op.pnode] + self.secondaries
8407
    return nl, nl
8408

    
8409
  def _ReadExportInfo(self):
8410
    """Reads the export information from disk.
8411

8412
    It will override the opcode source node and path with the actual
8413
    information, if these two were not specified before.
8414

8415
    @return: the export information
8416

8417
    """
8418
    assert self.op.mode == constants.INSTANCE_IMPORT
8419

    
8420
    src_node = self.op.src_node
8421
    src_path = self.op.src_path
8422

    
8423
    if src_node is None:
8424
      locked_nodes = self.owned_locks(locking.LEVEL_NODE)
8425
      exp_list = self.rpc.call_export_list(locked_nodes)
8426
      found = False
8427
      for node in exp_list:
8428
        if exp_list[node].fail_msg:
8429
          continue
8430
        if src_path in exp_list[node].payload:
8431
          found = True
8432
          self.op.src_node = src_node = node
8433
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
8434
                                                       src_path)
8435
          break
8436
      if not found:
8437
        raise errors.OpPrereqError("No export found for relative path %s" %
8438
                                    src_path, errors.ECODE_INVAL)
8439

    
8440
    _CheckNodeOnline(self, src_node)
8441
    result = self.rpc.call_export_info(src_node, src_path)
8442
    result.Raise("No export or invalid export found in dir %s" % src_path)
8443

    
8444
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
8445
    if not export_info.has_section(constants.INISECT_EXP):
8446
      raise errors.ProgrammerError("Corrupted export config",
8447
                                   errors.ECODE_ENVIRON)
8448

    
8449
    ei_version = export_info.get(constants.INISECT_EXP, "version")
8450
    if (int(ei_version) != constants.EXPORT_VERSION):
8451
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
8452
                                 (ei_version, constants.EXPORT_VERSION),
8453
                                 errors.ECODE_ENVIRON)
8454
    return export_info
8455

    
8456
  def _ReadExportParams(self, einfo):
8457
    """Use export parameters as defaults.
8458

8459
    In case the opcode doesn't specify (as in override) some instance
8460
    parameters, then try to use them from the export information, if
8461
    that declares them.
8462

8463
    """
8464
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
8465

    
8466
    if self.op.disk_template is None:
8467
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
8468
        self.op.disk_template = einfo.get(constants.INISECT_INS,
8469
                                          "disk_template")
8470
      else:
8471
        raise errors.OpPrereqError("No disk template specified and the export"
8472
                                   " is missing the disk_template information",
8473
                                   errors.ECODE_INVAL)
8474

    
8475
    if not self.op.disks:
8476
      if einfo.has_option(constants.INISECT_INS, "disk_count"):
8477
        disks = []
8478
        # TODO: import the disk iv_name too
8479
        for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
8480
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
8481
          disks.append({constants.IDISK_SIZE: disk_sz})
8482
        self.op.disks = disks
8483
      else:
8484
        raise errors.OpPrereqError("No disk info specified and the export"
8485
                                   " is missing the disk information",
8486
                                   errors.ECODE_INVAL)
8487

    
8488
    if (not self.op.nics and
8489
        einfo.has_option(constants.INISECT_INS, "nic_count")):
8490
      nics = []
8491
      for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
8492
        ndict = {}
8493
        for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
8494
          v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
8495
          ndict[name] = v
8496
        nics.append(ndict)
8497
      self.op.nics = nics
8498

    
8499
    if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
8500
      self.op.tags = einfo.get(constants.INISECT_INS, "tags").split()
8501

    
8502
    if (self.op.hypervisor is None and
8503
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
8504
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
8505

    
8506
    if einfo.has_section(constants.INISECT_HYP):
8507
      # use the export parameters but do not override the ones
8508
      # specified by the user
8509
      for name, value in einfo.items(constants.INISECT_HYP):
8510
        if name not in self.op.hvparams:
8511
          self.op.hvparams[name] = value
8512

    
8513
    if einfo.has_section(constants.INISECT_BEP):
8514
      # use the parameters, without overriding
8515
      for name, value in einfo.items(constants.INISECT_BEP):
8516
        if name not in self.op.beparams:
8517
          self.op.beparams[name] = value
8518
    else:
8519
      # try to read the parameters old style, from the main section
8520
      for name in constants.BES_PARAMETERS:
8521
        if (name not in self.op.beparams and
8522
            einfo.has_option(constants.INISECT_INS, name)):
8523
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
8524

    
8525
    if einfo.has_section(constants.INISECT_OSP):
8526
      # use the parameters, without overriding
8527
      for name, value in einfo.items(constants.INISECT_OSP):
8528
        if name not in self.op.osparams:
8529
          self.op.osparams[name] = value
8530

    
8531
  def _RevertToDefaults(self, cluster):
8532
    """Revert the instance parameters to the default values.
8533

8534
    """
8535
    # hvparams
8536
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
8537
    for name in self.op.hvparams.keys():
8538
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
8539
        del self.op.hvparams[name]
8540
    # beparams
8541
    be_defs = cluster.SimpleFillBE({})
8542
    for name in self.op.beparams.keys():
8543
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
8544
        del self.op.beparams[name]
8545
    # nic params
8546
    nic_defs = cluster.SimpleFillNIC({})
8547
    for nic in self.op.nics:
8548
      for name in constants.NICS_PARAMETERS:
8549
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
8550
          del nic[name]
8551
    # osparams
8552
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
8553
    for name in self.op.osparams.keys():
8554
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
8555
        del self.op.osparams[name]
8556

    
8557
  def _CalculateFileStorageDir(self):
8558
    """Calculate final instance file storage dir.
8559

8560
    """
8561
    # file storage dir calculation/check
8562
    self.instance_file_storage_dir = None
8563
    if self.op.disk_template in constants.DTS_FILEBASED:
8564
      # build the full file storage dir path
8565
      joinargs = []
8566

    
8567
      if self.op.disk_template == constants.DT_SHARED_FILE:
8568
        get_fsd_fn = self.cfg.GetSharedFileStorageDir
8569
      else:
8570
        get_fsd_fn = self.cfg.GetFileStorageDir
8571

    
8572
      cfg_storagedir = get_fsd_fn()
8573
      if not cfg_storagedir:
8574
        raise errors.OpPrereqError("Cluster file storage dir not defined")
8575
      joinargs.append(cfg_storagedir)
8576

    
8577
      if self.op.file_storage_dir is not None:
8578
        joinargs.append(self.op.file_storage_dir)
8579

    
8580
      joinargs.append(self.op.instance_name)
8581

    
8582
      # pylint: disable=W0142
8583
      self.instance_file_storage_dir = utils.PathJoin(*joinargs)
8584

    
8585
  def CheckPrereq(self):
8586
    """Check prerequisites.
8587

8588
    """
8589
    self._CalculateFileStorageDir()
8590

    
8591
    if self.op.mode == constants.INSTANCE_IMPORT:
8592
      export_info = self._ReadExportInfo()
8593
      self._ReadExportParams(export_info)
8594

    
8595
    if (not self.cfg.GetVGName() and
8596
        self.op.disk_template not in constants.DTS_NOT_LVM):
8597
      raise errors.OpPrereqError("Cluster does not support lvm-based"
8598
                                 " instances", errors.ECODE_STATE)
8599

    
8600
    if self.op.hypervisor is None:
8601
      self.op.hypervisor = self.cfg.GetHypervisorType()
8602

    
8603
    cluster = self.cfg.GetClusterInfo()
8604
    enabled_hvs = cluster.enabled_hypervisors
8605
    if self.op.hypervisor not in enabled_hvs:
8606
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
8607
                                 " cluster (%s)" % (self.op.hypervisor,
8608
                                  ",".join(enabled_hvs)),
8609
                                 errors.ECODE_STATE)
8610

    
8611
    # Check tag validity
8612
    for tag in self.op.tags:
8613
      objects.TaggableObject.ValidateTag(tag)
8614

    
8615
    # check hypervisor parameter syntax (locally)
8616
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
8617
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
8618
                                      self.op.hvparams)
8619
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
8620
    hv_type.CheckParameterSyntax(filled_hvp)
8621
    self.hv_full = filled_hvp
8622
    # check that we don't specify global parameters on an instance
8623
    _CheckGlobalHvParams(self.op.hvparams)
8624

    
8625
    # fill and remember the beparams dict
8626
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
8627
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
8628

    
8629
    # build os parameters
8630
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
8631

    
8632
    # now that hvp/bep are in final format, let's reset to defaults,
8633
    # if told to do so
8634
    if self.op.identify_defaults:
8635
      self._RevertToDefaults(cluster)
8636

    
8637
    # NIC buildup
8638
    self.nics = []
8639
    for idx, nic in enumerate(self.op.nics):
8640
      nic_mode_req = nic.get(constants.INIC_MODE, None)
8641
      nic_mode = nic_mode_req
8642
      if nic_mode is None:
8643
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
8644

    
8645
      # in routed mode, for the first nic, the default ip is 'auto'
8646
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
8647
        default_ip_mode = constants.VALUE_AUTO
8648
      else:
8649
        default_ip_mode = constants.VALUE_NONE
8650

    
8651
      # ip validity checks
8652
      ip = nic.get(constants.INIC_IP, default_ip_mode)
8653
      if ip is None or ip.lower() == constants.VALUE_NONE:
8654
        nic_ip = None
8655
      elif ip.lower() == constants.VALUE_AUTO:
8656
        if not self.op.name_check:
8657
          raise errors.OpPrereqError("IP address set to auto but name checks"
8658
                                     " have been skipped",
8659
                                     errors.ECODE_INVAL)
8660
        nic_ip = self.hostname1.ip
8661
      else:
8662
        if not netutils.IPAddress.IsValid(ip):
8663
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
8664
                                     errors.ECODE_INVAL)
8665
        nic_ip = ip
8666

    
8667
      # TODO: check the ip address for uniqueness
8668
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
8669
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
8670
                                   errors.ECODE_INVAL)
8671

    
8672
      # MAC address verification
8673
      mac = nic.get(constants.INIC_MAC, constants.VALUE_AUTO)
8674
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
8675
        mac = utils.NormalizeAndValidateMac(mac)
8676

    
8677
        try:
8678
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
8679
        except errors.ReservationError:
8680
          raise errors.OpPrereqError("MAC address %s already in use"
8681
                                     " in cluster" % mac,
8682
                                     errors.ECODE_NOTUNIQUE)
8683

    
8684
      #  Build nic parameters
8685
      link = nic.get(constants.INIC_LINK, None)
8686
      nicparams = {}
8687
      if nic_mode_req:
8688
        nicparams[constants.NIC_MODE] = nic_mode_req
8689
      if link:
8690
        nicparams[constants.NIC_LINK] = link
8691

    
8692
      check_params = cluster.SimpleFillNIC(nicparams)
8693
      objects.NIC.CheckParameterSyntax(check_params)
8694
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
8695

    
8696
    # disk checks/pre-build
8697
    default_vg = self.cfg.GetVGName()
8698
    self.disks = []
8699
    for disk in self.op.disks:
8700
      mode = disk.get(constants.IDISK_MODE, constants.DISK_RDWR)
8701
      if mode not in constants.DISK_ACCESS_SET:
8702
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
8703
                                   mode, errors.ECODE_INVAL)
8704
      size = disk.get(constants.IDISK_SIZE, None)
8705
      if size is None:
8706
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
8707
      try:
8708
        size = int(size)
8709
      except (TypeError, ValueError):
8710
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
8711
                                   errors.ECODE_INVAL)
8712

    
8713
      data_vg = disk.get(constants.IDISK_VG, default_vg)
8714
      new_disk = {
8715
        constants.IDISK_SIZE: size,
8716
        constants.IDISK_MODE: mode,
8717
        constants.IDISK_VG: data_vg,
8718
        constants.IDISK_METAVG: disk.get(constants.IDISK_METAVG, data_vg),
8719
        }
8720
      if constants.IDISK_ADOPT in disk:
8721
        new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
8722
      self.disks.append(new_disk)
8723

    
8724
    if self.op.mode == constants.INSTANCE_IMPORT:
8725

    
8726
      # Check that the new instance doesn't have less disks than the export
8727
      instance_disks = len(self.disks)
8728
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
8729
      if instance_disks < export_disks:
8730
        raise errors.OpPrereqError("Not enough disks to import."
8731
                                   " (instance: %d, export: %d)" %
8732
                                   (instance_disks, export_disks),
8733
                                   errors.ECODE_INVAL)
8734

    
8735
      disk_images = []
8736
      for idx in range(export_disks):
8737
        option = "disk%d_dump" % idx
8738
        if export_info.has_option(constants.INISECT_INS, option):
8739
          # FIXME: are the old os-es, disk sizes, etc. useful?
8740
          export_name = export_info.get(constants.INISECT_INS, option)
8741
          image = utils.PathJoin(self.op.src_path, export_name)
8742
          disk_images.append(image)
8743
        else:
8744
          disk_images.append(False)
8745

    
8746
      self.src_images = disk_images
8747

    
8748
      old_name = export_info.get(constants.INISECT_INS, "name")
8749
      try:
8750
        exp_nic_count = export_info.getint(constants.INISECT_INS, "nic_count")
8751
      except (TypeError, ValueError), err:
8752
        raise errors.OpPrereqError("Invalid export file, nic_count is not"
8753
                                   " an integer: %s" % str(err),
8754
                                   errors.ECODE_STATE)
8755
      if self.op.instance_name == old_name:
8756
        for idx, nic in enumerate(self.nics):
8757
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
8758
            nic_mac_ini = "nic%d_mac" % idx
8759
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
8760

    
8761
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
8762

    
8763
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
8764
    if self.op.ip_check:
8765
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
8766
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
8767
                                   (self.check_ip, self.op.instance_name),
8768
                                   errors.ECODE_NOTUNIQUE)
8769

    
8770
    #### mac address generation
8771
    # By generating here the mac address both the allocator and the hooks get
8772
    # the real final mac address rather than the 'auto' or 'generate' value.
8773
    # There is a race condition between the generation and the instance object
8774
    # creation, which means that we know the mac is valid now, but we're not
8775
    # sure it will be when we actually add the instance. If things go bad
8776
    # adding the instance will abort because of a duplicate mac, and the
8777
    # creation job will fail.
8778
    for nic in self.nics:
8779
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
8780
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
8781

    
8782
    #### allocator run
8783

    
8784
    if self.op.iallocator is not None:
8785
      self._RunAllocator()
8786

    
8787
    #### node related checks
8788

    
8789
    # check primary node
8790
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
8791
    assert self.pnode is not None, \
8792
      "Cannot retrieve locked node %s" % self.op.pnode
8793
    if pnode.offline:
8794
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
8795
                                 pnode.name, errors.ECODE_STATE)
8796
    if pnode.drained:
8797
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
8798
                                 pnode.name, errors.ECODE_STATE)
8799
    if not pnode.vm_capable:
8800
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
8801
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
8802

    
8803
    self.secondaries = []
8804

    
8805
    # mirror node verification
8806
    if self.op.disk_template in constants.DTS_INT_MIRROR:
8807
      if self.op.snode == pnode.name:
8808
        raise errors.OpPrereqError("The secondary node cannot be the"
8809
                                   " primary node", errors.ECODE_INVAL)
8810
      _CheckNodeOnline(self, self.op.snode)
8811
      _CheckNodeNotDrained(self, self.op.snode)
8812
      _CheckNodeVmCapable(self, self.op.snode)
8813
      self.secondaries.append(self.op.snode)
8814

    
8815
    nodenames = [pnode.name] + self.secondaries
8816

    
8817
    if not self.adopt_disks:
8818
      # Check lv size requirements, if not adopting
8819
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
8820
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
8821

    
8822
    elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
8823
      all_lvs = set(["%s/%s" % (disk[constants.IDISK_VG],
8824
                                disk[constants.IDISK_ADOPT])
8825
                     for disk in self.disks])
8826
      if len(all_lvs) != len(self.disks):
8827
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
8828
                                   errors.ECODE_INVAL)
8829
      for lv_name in all_lvs:
8830
        try:
8831
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
8832
          # to ReserveLV uses the same syntax
8833
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
8834
        except errors.ReservationError:
8835
          raise errors.OpPrereqError("LV named %s used by another instance" %
8836
                                     lv_name, errors.ECODE_NOTUNIQUE)
8837

    
8838
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
8839
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
8840

    
8841
      node_lvs = self.rpc.call_lv_list([pnode.name],
8842
                                       vg_names.payload.keys())[pnode.name]
8843
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
8844
      node_lvs = node_lvs.payload
8845

    
8846
      delta = all_lvs.difference(node_lvs.keys())
8847
      if delta:
8848
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
8849
                                   utils.CommaJoin(delta),
8850
                                   errors.ECODE_INVAL)
8851
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
8852
      if online_lvs:
8853
        raise errors.OpPrereqError("Online logical volumes found, cannot"
8854
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
8855
                                   errors.ECODE_STATE)
8856
      # update the size of disk based on what is found
8857
      for dsk in self.disks:
8858
        dsk[constants.IDISK_SIZE] = \
8859
          int(float(node_lvs["%s/%s" % (dsk[constants.IDISK_VG],
8860
                                        dsk[constants.IDISK_ADOPT])][0]))
8861

    
8862
    elif self.op.disk_template == constants.DT_BLOCK:
8863
      # Normalize and de-duplicate device paths
8864
      all_disks = set([os.path.abspath(disk[constants.IDISK_ADOPT])
8865
                       for disk in self.disks])
8866
      if len(all_disks) != len(self.disks):
8867
        raise errors.OpPrereqError("Duplicate disk names given for adoption",
8868
                                   errors.ECODE_INVAL)
8869
      baddisks = [d for d in all_disks
8870
                  if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
8871
      if baddisks:
8872
        raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
8873
                                   " cannot be adopted" %
8874
                                   (", ".join(baddisks),
8875
                                    constants.ADOPTABLE_BLOCKDEV_ROOT),
8876
                                   errors.ECODE_INVAL)
8877

    
8878
      node_disks = self.rpc.call_bdev_sizes([pnode.name],
8879
                                            list(all_disks))[pnode.name]
8880
      node_disks.Raise("Cannot get block device information from node %s" %
8881
                       pnode.name)
8882
      node_disks = node_disks.payload
8883
      delta = all_disks.difference(node_disks.keys())
8884
      if delta:
8885
        raise errors.OpPrereqError("Missing block device(s): %s" %
8886
                                   utils.CommaJoin(delta),
8887
                                   errors.ECODE_INVAL)
8888
      for dsk in self.disks:
8889
        dsk[constants.IDISK_SIZE] = \
8890
          int(float(node_disks[dsk[constants.IDISK_ADOPT]]))
8891

    
8892
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
8893

    
8894
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
8895
    # check OS parameters (remotely)
8896
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
8897

    
8898
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
8899

    
8900
    # memory check on primary node
8901
    if self.op.start:
8902
      _CheckNodeFreeMemory(self, self.pnode.name,
8903
                           "creating instance %s" % self.op.instance_name,
8904
                           self.be_full[constants.BE_MEMORY],
8905
                           self.op.hypervisor)
8906

    
8907
    self.dry_run_result = list(nodenames)
8908

    
8909
  def Exec(self, feedback_fn):
8910
    """Create and add the instance to the cluster.
8911

8912
    """
8913
    instance = self.op.instance_name
8914
    pnode_name = self.pnode.name
8915

    
8916
    ht_kind = self.op.hypervisor
8917
    if ht_kind in constants.HTS_REQ_PORT:
8918
      network_port = self.cfg.AllocatePort()
8919
    else:
8920
      network_port = None
8921

    
8922
    disks = _GenerateDiskTemplate(self,
8923
                                  self.op.disk_template,
8924
                                  instance, pnode_name,
8925
                                  self.secondaries,
8926
                                  self.disks,
8927
                                  self.instance_file_storage_dir,
8928
                                  self.op.file_driver,
8929
                                  0,
8930
                                  feedback_fn)
8931

    
8932
    iobj = objects.Instance(name=instance, os=self.op.os_type,
8933
                            primary_node=pnode_name,
8934
                            nics=self.nics, disks=disks,
8935
                            disk_template=self.op.disk_template,
8936
                            admin_up=False,
8937
                            network_port=network_port,
8938
                            beparams=self.op.beparams,
8939
                            hvparams=self.op.hvparams,
8940
                            hypervisor=self.op.hypervisor,
8941
                            osparams=self.op.osparams,
8942
                            )
8943

    
8944
    if self.op.tags:
8945
      for tag in self.op.tags:
8946
        iobj.AddTag(tag)
8947

    
8948
    if self.adopt_disks:
8949
      if self.op.disk_template == constants.DT_PLAIN:
8950
        # rename LVs to the newly-generated names; we need to construct
8951
        # 'fake' LV disks with the old data, plus the new unique_id
8952
        tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
8953
        rename_to = []
8954
        for t_dsk, a_dsk in zip(tmp_disks, self.disks):
8955
          rename_to.append(t_dsk.logical_id)
8956
          t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
8957
          self.cfg.SetDiskID(t_dsk, pnode_name)
8958
        result = self.rpc.call_blockdev_rename(pnode_name,
8959
                                               zip(tmp_disks, rename_to))
8960
        result.Raise("Failed to rename adoped LVs")
8961
    else:
8962
      feedback_fn("* creating instance disks...")
8963
      try:
8964
        _CreateDisks(self, iobj)
8965
      except errors.OpExecError:
8966
        self.LogWarning("Device creation failed, reverting...")
8967
        try:
8968
          _RemoveDisks(self, iobj)
8969
        finally:
8970
          self.cfg.ReleaseDRBDMinors(instance)
8971
          raise
8972

    
8973
    feedback_fn("adding instance %s to cluster config" % instance)
8974

    
8975
    self.cfg.AddInstance(iobj, self.proc.GetECId())
8976

    
8977
    # Declare that we don't want to remove the instance lock anymore, as we've
8978
    # added the instance to the config
8979
    del self.remove_locks[locking.LEVEL_INSTANCE]
8980

    
8981
    if self.op.mode == constants.INSTANCE_IMPORT:
8982
      # Release unused nodes
8983
      _ReleaseLocks(self, locking.LEVEL_NODE, keep=[self.op.src_node])
8984
    else:
8985
      # Release all nodes
8986
      _ReleaseLocks(self, locking.LEVEL_NODE)
8987

    
8988
    disk_abort = False
8989
    if not self.adopt_disks and self.cfg.GetClusterInfo().prealloc_wipe_disks:
8990
      feedback_fn("* wiping instance disks...")
8991
      try:
8992
        _WipeDisks(self, iobj)
8993
      except errors.OpExecError, err:
8994
        logging.exception("Wiping disks failed")
8995
        self.LogWarning("Wiping instance disks failed (%s)", err)
8996
        disk_abort = True
8997

    
8998
    if disk_abort:
8999
      # Something is already wrong with the disks, don't do anything else
9000
      pass
9001
    elif self.op.wait_for_sync:
9002
      disk_abort = not _WaitForSync(self, iobj)
9003
    elif iobj.disk_template in constants.DTS_INT_MIRROR:
9004
      # make sure the disks are not degraded (still sync-ing is ok)
9005
      feedback_fn("* checking mirrors status")
9006
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
9007
    else:
9008
      disk_abort = False
9009

    
9010
    if disk_abort:
9011
      _RemoveDisks(self, iobj)
9012
      self.cfg.RemoveInstance(iobj.name)
9013
      # Make sure the instance lock gets removed
9014
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
9015
      raise errors.OpExecError("There are some degraded disks for"
9016
                               " this instance")
9017

    
9018
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
9019
      if self.op.mode == constants.INSTANCE_CREATE:
9020
        if not self.op.no_install:
9021
          pause_sync = (iobj.disk_template in constants.DTS_INT_MIRROR and
9022
                        not self.op.wait_for_sync)
9023
          if pause_sync:
9024
            feedback_fn("* pausing disk sync to install instance OS")
9025
            result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
9026
                                                              iobj.disks, True)
9027
            for idx, success in enumerate(result.payload):
9028
              if not success:
9029
                logging.warn("pause-sync of instance %s for disk %d failed",
9030
                             instance, idx)
9031

    
9032
          feedback_fn("* running the instance OS create scripts...")
9033
          # FIXME: pass debug option from opcode to backend
9034
          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
9035
                                                 self.op.debug_level)
9036
          if pause_sync:
9037
            feedback_fn("* resuming disk sync")
9038
            result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
9039
                                                              iobj.disks, False)
9040
            for idx, success in enumerate(result.payload):
9041
              if not success:
9042
                logging.warn("resume-sync of instance %s for disk %d failed",
9043
                             instance, idx)
9044

    
9045
          result.Raise("Could not add os for instance %s"
9046
                       " on node %s" % (instance, pnode_name))
9047

    
9048
      elif self.op.mode == constants.INSTANCE_IMPORT:
9049
        feedback_fn("* running the instance OS import scripts...")
9050

    
9051
        transfers = []
9052

    
9053
        for idx, image in enumerate(self.src_images):
9054
          if not image:
9055
            continue
9056

    
9057
          # FIXME: pass debug option from opcode to backend
9058
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
9059
                                             constants.IEIO_FILE, (image, ),
9060
                                             constants.IEIO_SCRIPT,
9061
                                             (iobj.disks[idx], idx),
9062
                                             None)
9063
          transfers.append(dt)
9064

    
9065
        import_result = \
9066
          masterd.instance.TransferInstanceData(self, feedback_fn,
9067
                                                self.op.src_node, pnode_name,
9068
                                                self.pnode.secondary_ip,
9069
                                                iobj, transfers)
9070
        if not compat.all(import_result):
9071
          self.LogWarning("Some disks for instance %s on node %s were not"
9072
                          " imported successfully" % (instance, pnode_name))
9073

    
9074
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
9075
        feedback_fn("* preparing remote import...")
9076
        # The source cluster will stop the instance before attempting to make a
9077
        # connection. In some cases stopping an instance can take a long time,
9078
        # hence the shutdown timeout is added to the connection timeout.
9079
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
9080
                           self.op.source_shutdown_timeout)
9081
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
9082

    
9083
        assert iobj.primary_node == self.pnode.name
9084
        disk_results = \
9085
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
9086
                                        self.source_x509_ca,
9087
                                        self._cds, timeouts)
9088
        if not compat.all(disk_results):
9089
          # TODO: Should the instance still be started, even if some disks
9090
          # failed to import (valid for local imports, too)?
9091
          self.LogWarning("Some disks for instance %s on node %s were not"
9092
                          " imported successfully" % (instance, pnode_name))
9093

    
9094
        # Run rename script on newly imported instance
9095
        assert iobj.name == instance
9096
        feedback_fn("Running rename script for %s" % instance)
9097
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
9098
                                                   self.source_instance_name,
9099
                                                   self.op.debug_level)
9100
        if result.fail_msg:
9101
          self.LogWarning("Failed to run rename script for %s on node"
9102
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
9103

    
9104
      else:
9105
        # also checked in the prereq part
9106
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
9107
                                     % self.op.mode)
9108

    
9109
    if self.op.start:
9110
      iobj.admin_up = True
9111
      self.cfg.Update(iobj, feedback_fn)
9112
      logging.info("Starting instance %s on node %s", instance, pnode_name)
9113
      feedback_fn("* starting instance...")
9114
      result = self.rpc.call_instance_start(pnode_name, iobj,
9115
                                            None, None, False)
9116
      result.Raise("Could not start instance")
9117

    
9118
    return list(iobj.all_nodes)
9119

    
9120

    
9121
class LUInstanceConsole(NoHooksLU):
9122
  """Connect to an instance's console.
9123

9124
  This is somewhat special in that it returns the command line that
9125
  you need to run on the master node in order to connect to the
9126
  console.
9127

9128
  """
9129
  REQ_BGL = False
9130

    
9131
  def ExpandNames(self):
9132
    self._ExpandAndLockInstance()
9133

    
9134
  def CheckPrereq(self):
9135
    """Check prerequisites.
9136

9137
    This checks that the instance is in the cluster.
9138

9139
    """
9140
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9141
    assert self.instance is not None, \
9142
      "Cannot retrieve locked instance %s" % self.op.instance_name
9143
    _CheckNodeOnline(self, self.instance.primary_node)
9144

    
9145
  def Exec(self, feedback_fn):
9146
    """Connect to the console of an instance
9147

9148
    """
9149
    instance = self.instance
9150
    node = instance.primary_node
9151

    
9152
    node_insts = self.rpc.call_instance_list([node],
9153
                                             [instance.hypervisor])[node]
9154
    node_insts.Raise("Can't get node information from %s" % node)
9155

    
9156
    if instance.name not in node_insts.payload:
9157
      if instance.admin_up:
9158
        state = constants.INSTST_ERRORDOWN
9159
      else:
9160
        state = constants.INSTST_ADMINDOWN
9161
      raise errors.OpExecError("Instance %s is not running (state %s)" %
9162
                               (instance.name, state))
9163

    
9164
    logging.debug("Connecting to console of %s on %s", instance.name, node)
9165

    
9166
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
9167

    
9168

    
9169
def _GetInstanceConsole(cluster, instance):
9170
  """Returns console information for an instance.
9171

9172
  @type cluster: L{objects.Cluster}
9173
  @type instance: L{objects.Instance}
9174
  @rtype: dict
9175

9176
  """
9177
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
9178
  # beparams and hvparams are passed separately, to avoid editing the
9179
  # instance and then saving the defaults in the instance itself.
9180
  hvparams = cluster.FillHV(instance)
9181
  beparams = cluster.FillBE(instance)
9182
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
9183

    
9184
  assert console.instance == instance.name
9185
  assert console.Validate()
9186

    
9187
  return console.ToDict()
9188

    
9189

    
9190
class LUInstanceReplaceDisks(LogicalUnit):
9191
  """Replace the disks of an instance.
9192

9193
  """
9194
  HPATH = "mirrors-replace"
9195
  HTYPE = constants.HTYPE_INSTANCE
9196
  REQ_BGL = False
9197

    
9198
  def CheckArguments(self):
9199
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
9200
                                  self.op.iallocator)
9201

    
9202
  def ExpandNames(self):
9203
    self._ExpandAndLockInstance()
9204

    
9205
    assert locking.LEVEL_NODE not in self.needed_locks
9206
    assert locking.LEVEL_NODEGROUP not in self.needed_locks
9207

    
9208
    assert self.op.iallocator is None or self.op.remote_node is None, \
9209
      "Conflicting options"
9210

    
9211
    if self.op.remote_node is not None:
9212
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9213

    
9214
      # Warning: do not remove the locking of the new secondary here
9215
      # unless DRBD8.AddChildren is changed to work in parallel;
9216
      # currently it doesn't since parallel invocations of
9217
      # FindUnusedMinor will conflict
9218
      self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
9219
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
9220
    else:
9221
      self.needed_locks[locking.LEVEL_NODE] = []
9222
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9223

    
9224
      if self.op.iallocator is not None:
9225
        # iallocator will select a new node in the same group
9226
        self.needed_locks[locking.LEVEL_NODEGROUP] = []
9227

    
9228
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
9229
                                   self.op.iallocator, self.op.remote_node,
9230
                                   self.op.disks, False, self.op.early_release)
9231

    
9232
    self.tasklets = [self.replacer]
9233

    
9234
  def DeclareLocks(self, level):
9235
    if level == locking.LEVEL_NODEGROUP:
9236
      assert self.op.remote_node is None
9237
      assert self.op.iallocator is not None
9238
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
9239

    
9240
      self.share_locks[locking.LEVEL_NODEGROUP] = 1
9241
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
9242
        self.cfg.GetInstanceNodeGroups(self.op.instance_name)
9243

    
9244
    elif level == locking.LEVEL_NODE:
9245
      if self.op.iallocator is not None:
9246
        assert self.op.remote_node is None
9247
        assert not self.needed_locks[locking.LEVEL_NODE]
9248

    
9249
        # Lock member nodes of all locked groups
9250
        self.needed_locks[locking.LEVEL_NODE] = [node_name
9251
          for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
9252
          for node_name in self.cfg.GetNodeGroup(group_uuid).members]
9253
      else:
9254
        self._LockInstancesNodes()
9255

    
9256
  def BuildHooksEnv(self):
9257
    """Build hooks env.
9258

9259
    This runs on the master, the primary and all the secondaries.
9260

9261
    """
9262
    instance = self.replacer.instance
9263
    env = {
9264
      "MODE": self.op.mode,
9265
      "NEW_SECONDARY": self.op.remote_node,
9266
      "OLD_SECONDARY": instance.secondary_nodes[0],
9267
      }
9268
    env.update(_BuildInstanceHookEnvByObject(self, instance))
9269
    return env
9270

    
9271
  def BuildHooksNodes(self):
9272
    """Build hooks nodes.
9273

9274
    """
9275
    instance = self.replacer.instance
9276
    nl = [
9277
      self.cfg.GetMasterNode(),
9278
      instance.primary_node,
9279
      ]
9280
    if self.op.remote_node is not None:
9281
      nl.append(self.op.remote_node)
9282
    return nl, nl
9283

    
9284
  def CheckPrereq(self):
9285
    """Check prerequisites.
9286

9287
    """
9288
    assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
9289
            self.op.iallocator is None)
9290

    
9291
    owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
9292
    if owned_groups:
9293
      _CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
9294

    
9295
    return LogicalUnit.CheckPrereq(self)
9296

    
9297

    
9298
class TLReplaceDisks(Tasklet):
9299
  """Replaces disks for an instance.
9300

9301
  Note: Locking is not within the scope of this class.
9302

9303
  """
9304
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
9305
               disks, delay_iallocator, early_release):
9306
    """Initializes this class.
9307

9308
    """
9309
    Tasklet.__init__(self, lu)
9310

    
9311
    # Parameters
9312
    self.instance_name = instance_name
9313
    self.mode = mode
9314
    self.iallocator_name = iallocator_name
9315
    self.remote_node = remote_node
9316
    self.disks = disks
9317
    self.delay_iallocator = delay_iallocator
9318
    self.early_release = early_release
9319

    
9320
    # Runtime data
9321
    self.instance = None
9322
    self.new_node = None
9323
    self.target_node = None
9324
    self.other_node = None
9325
    self.remote_node_info = None
9326
    self.node_secondary_ip = None
9327

    
9328
  @staticmethod
9329
  def CheckArguments(mode, remote_node, iallocator):
9330
    """Helper function for users of this class.
9331

9332
    """
9333
    # check for valid parameter combination
9334
    if mode == constants.REPLACE_DISK_CHG:
9335
      if remote_node is None and iallocator is None:
9336
        raise errors.OpPrereqError("When changing the secondary either an"
9337
                                   " iallocator script must be used or the"
9338
                                   " new node given", errors.ECODE_INVAL)
9339

    
9340
      if remote_node is not None and iallocator is not None:
9341
        raise errors.OpPrereqError("Give either the iallocator or the new"
9342
                                   " secondary, not both", errors.ECODE_INVAL)
9343

    
9344
    elif remote_node is not None or iallocator is not None:
9345
      # Not replacing the secondary
9346
      raise errors.OpPrereqError("The iallocator and new node options can"
9347
                                 " only be used when changing the"
9348
                                 " secondary node", errors.ECODE_INVAL)
9349

    
9350
  @staticmethod
9351
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
9352
    """Compute a new secondary node using an IAllocator.
9353

9354
    """
9355
    ial = IAllocator(lu.cfg, lu.rpc,
9356
                     mode=constants.IALLOCATOR_MODE_RELOC,
9357
                     name=instance_name,
9358
                     relocate_from=list(relocate_from))
9359

    
9360
    ial.Run(iallocator_name)
9361

    
9362
    if not ial.success:
9363
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
9364
                                 " %s" % (iallocator_name, ial.info),
9365
                                 errors.ECODE_NORES)
9366

    
9367
    if len(ial.result) != ial.required_nodes:
9368
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
9369
                                 " of nodes (%s), required %s" %
9370
                                 (iallocator_name,
9371
                                  len(ial.result), ial.required_nodes),
9372
                                 errors.ECODE_FAULT)
9373

    
9374
    remote_node_name = ial.result[0]
9375

    
9376
    lu.LogInfo("Selected new secondary for instance '%s': %s",
9377
               instance_name, remote_node_name)
9378

    
9379
    return remote_node_name
9380

    
9381
  def _FindFaultyDisks(self, node_name):
9382
    """Wrapper for L{_FindFaultyInstanceDisks}.
9383

9384
    """
9385
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
9386
                                    node_name, True)
9387

    
9388
  def _CheckDisksActivated(self, instance):
9389
    """Checks if the instance disks are activated.
9390

9391
    @param instance: The instance to check disks
9392
    @return: True if they are activated, False otherwise
9393

9394
    """
9395
    nodes = instance.all_nodes
9396

    
9397
    for idx, dev in enumerate(instance.disks):
9398
      for node in nodes:
9399
        self.lu.LogInfo("Checking disk/%d on %s", idx, node)
9400
        self.cfg.SetDiskID(dev, node)
9401

    
9402
        result = self.rpc.call_blockdev_find(node, dev)
9403

    
9404
        if result.offline:
9405
          continue
9406
        elif result.fail_msg or not result.payload:
9407
          return False
9408

    
9409
    return True
9410

    
9411
  def CheckPrereq(self):
9412
    """Check prerequisites.
9413

9414
    This checks that the instance is in the cluster.
9415

9416
    """
9417
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
9418
    assert instance is not None, \
9419
      "Cannot retrieve locked instance %s" % self.instance_name
9420

    
9421
    if instance.disk_template != constants.DT_DRBD8:
9422
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
9423
                                 " instances", errors.ECODE_INVAL)
9424

    
9425
    if len(instance.secondary_nodes) != 1:
9426
      raise errors.OpPrereqError("The instance has a strange layout,"
9427
                                 " expected one secondary but found %d" %
9428
                                 len(instance.secondary_nodes),
9429
                                 errors.ECODE_FAULT)
9430

    
9431
    if not self.delay_iallocator:
9432
      self._CheckPrereq2()
9433

    
9434
  def _CheckPrereq2(self):
9435
    """Check prerequisites, second part.
9436

9437
    This function should always be part of CheckPrereq. It was separated and is
9438
    now called from Exec because during node evacuation iallocator was only
9439
    called with an unmodified cluster model, not taking planned changes into
9440
    account.
9441

9442
    """
9443
    instance = self.instance
9444
    secondary_node = instance.secondary_nodes[0]
9445

    
9446
    if self.iallocator_name is None:
9447
      remote_node = self.remote_node
9448
    else:
9449
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
9450
                                       instance.name, instance.secondary_nodes)
9451

    
9452
    if remote_node is None:
9453
      self.remote_node_info = None
9454
    else:
9455
      assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
9456
             "Remote node '%s' is not locked" % remote_node
9457

    
9458
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
9459
      assert self.remote_node_info is not None, \
9460
        "Cannot retrieve locked node %s" % remote_node
9461

    
9462
    if remote_node == self.instance.primary_node:
9463
      raise errors.OpPrereqError("The specified node is the primary node of"
9464
                                 " the instance", errors.ECODE_INVAL)
9465

    
9466
    if remote_node == secondary_node:
9467
      raise errors.OpPrereqError("The specified node is already the"
9468
                                 " secondary node of the instance",
9469
                                 errors.ECODE_INVAL)
9470

    
9471
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
9472
                                    constants.REPLACE_DISK_CHG):
9473
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
9474
                                 errors.ECODE_INVAL)
9475

    
9476
    if self.mode == constants.REPLACE_DISK_AUTO:
9477
      if not self._CheckDisksActivated(instance):
9478
        raise errors.OpPrereqError("Please run activate-disks on instance %s"
9479
                                   " first" % self.instance_name,
9480
                                   errors.ECODE_STATE)
9481
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
9482
      faulty_secondary = self._FindFaultyDisks(secondary_node)
9483

    
9484
      if faulty_primary and faulty_secondary:
9485
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
9486
                                   " one node and can not be repaired"
9487
                                   " automatically" % self.instance_name,
9488
                                   errors.ECODE_STATE)
9489

    
9490
      if faulty_primary:
9491
        self.disks = faulty_primary
9492
        self.target_node = instance.primary_node
9493
        self.other_node = secondary_node
9494
        check_nodes = [self.target_node, self.other_node]
9495
      elif faulty_secondary:
9496
        self.disks = faulty_secondary
9497
        self.target_node = secondary_node
9498
        self.other_node = instance.primary_node
9499
        check_nodes = [self.target_node, self.other_node]
9500
      else:
9501
        self.disks = []
9502
        check_nodes = []
9503

    
9504
    else:
9505
      # Non-automatic modes
9506
      if self.mode == constants.REPLACE_DISK_PRI:
9507
        self.target_node = instance.primary_node
9508
        self.other_node = secondary_node
9509
        check_nodes = [self.target_node, self.other_node]
9510

    
9511
      elif self.mode == constants.REPLACE_DISK_SEC:
9512
        self.target_node = secondary_node
9513
        self.other_node = instance.primary_node
9514
        check_nodes = [self.target_node, self.other_node]
9515

    
9516
      elif self.mode == constants.REPLACE_DISK_CHG:
9517
        self.new_node = remote_node
9518
        self.other_node = instance.primary_node
9519
        self.target_node = secondary_node
9520
        check_nodes = [self.new_node, self.other_node]
9521

    
9522
        _CheckNodeNotDrained(self.lu, remote_node)
9523
        _CheckNodeVmCapable(self.lu, remote_node)
9524

    
9525
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
9526
        assert old_node_info is not None
9527
        if old_node_info.offline and not self.early_release:
9528
          # doesn't make sense to delay the release
9529
          self.early_release = True
9530
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
9531
                          " early-release mode", secondary_node)
9532

    
9533
      else:
9534
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
9535
                                     self.mode)
9536

    
9537
      # If not specified all disks should be replaced
9538
      if not self.disks:
9539
        self.disks = range(len(self.instance.disks))
9540

    
9541
    for node in check_nodes:
9542
      _CheckNodeOnline(self.lu, node)
9543

    
9544
    touched_nodes = frozenset(node_name for node_name in [self.new_node,
9545
                                                          self.other_node,
9546
                                                          self.target_node]
9547
                              if node_name is not None)
9548

    
9549
    # Release unneeded node locks
9550
    _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
9551

    
9552
    # Release any owned node group
9553
    if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
9554
      _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
9555

    
9556
    # Check whether disks are valid
9557
    for disk_idx in self.disks:
9558
      instance.FindDisk(disk_idx)
9559

    
9560
    # Get secondary node IP addresses
9561
    self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
9562
                                  in self.cfg.GetMultiNodeInfo(touched_nodes))
9563

    
9564
  def Exec(self, feedback_fn):
9565
    """Execute disk replacement.
9566

9567
    This dispatches the disk replacement to the appropriate handler.
9568

9569
    """
9570
    if self.delay_iallocator:
9571
      self._CheckPrereq2()
9572

    
9573
    if __debug__:
9574
      # Verify owned locks before starting operation
9575
      owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
9576
      assert set(owned_nodes) == set(self.node_secondary_ip), \
9577
          ("Incorrect node locks, owning %s, expected %s" %
9578
           (owned_nodes, self.node_secondary_ip.keys()))
9579

    
9580
      owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
9581
      assert list(owned_instances) == [self.instance_name], \
9582
          "Instance '%s' not locked" % self.instance_name
9583

    
9584
      assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
9585
          "Should not own any node group lock at this point"
9586

    
9587
    if not self.disks:
9588
      feedback_fn("No disks need replacement")
9589
      return
9590

    
9591
    feedback_fn("Replacing disk(s) %s for %s" %
9592
                (utils.CommaJoin(self.disks), self.instance.name))
9593

    
9594
    activate_disks = (not self.instance.admin_up)
9595

    
9596
    # Activate the instance disks if we're replacing them on a down instance
9597
    if activate_disks:
9598
      _StartInstanceDisks(self.lu, self.instance, True)
9599

    
9600
    try:
9601
      # Should we replace the secondary node?
9602
      if self.new_node is not None:
9603
        fn = self._ExecDrbd8Secondary
9604
      else:
9605
        fn = self._ExecDrbd8DiskOnly
9606

    
9607
      result = fn(feedback_fn)
9608
    finally:
9609
      # Deactivate the instance disks if we're replacing them on a
9610
      # down instance
9611
      if activate_disks:
9612
        _SafeShutdownInstanceDisks(self.lu, self.instance)
9613

    
9614
    if __debug__:
9615
      # Verify owned locks
9616
      owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
9617
      nodes = frozenset(self.node_secondary_ip)
9618
      assert ((self.early_release and not owned_nodes) or
9619
              (not self.early_release and not (set(owned_nodes) - nodes))), \
9620
        ("Not owning the correct locks, early_release=%s, owned=%r,"
9621
         " nodes=%r" % (self.early_release, owned_nodes, nodes))
9622

    
9623
    return result
9624

    
9625
  def _CheckVolumeGroup(self, nodes):
9626
    self.lu.LogInfo("Checking volume groups")
9627

    
9628
    vgname = self.cfg.GetVGName()
9629

    
9630
    # Make sure volume group exists on all involved nodes
9631
    results = self.rpc.call_vg_list(nodes)
9632
    if not results:
9633
      raise errors.OpExecError("Can't list volume groups on the nodes")
9634

    
9635
    for node in nodes:
9636
      res = results[node]
9637
      res.Raise("Error checking node %s" % node)
9638
      if vgname not in res.payload:
9639
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
9640
                                 (vgname, node))
9641

    
9642
  def _CheckDisksExistence(self, nodes):
9643
    # Check disk existence
9644
    for idx, dev in enumerate(self.instance.disks):
9645
      if idx not in self.disks:
9646
        continue
9647

    
9648
      for node in nodes:
9649
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
9650
        self.cfg.SetDiskID(dev, node)
9651

    
9652
        result = self.rpc.call_blockdev_find(node, dev)
9653

    
9654
        msg = result.fail_msg
9655
        if msg or not result.payload:
9656
          if not msg:
9657
            msg = "disk not found"
9658
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
9659
                                   (idx, node, msg))
9660

    
9661
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
9662
    for idx, dev in enumerate(self.instance.disks):
9663
      if idx not in self.disks:
9664
        continue
9665

    
9666
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
9667
                      (idx, node_name))
9668

    
9669
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
9670
                                   ldisk=ldisk):
9671
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
9672
                                 " replace disks for instance %s" %
9673
                                 (node_name, self.instance.name))
9674

    
9675
  def _CreateNewStorage(self, node_name):
9676
    """Create new storage on the primary or secondary node.
9677

9678
    This is only used for same-node replaces, not for changing the
9679
    secondary node, hence we don't want to modify the existing disk.
9680

9681
    """
9682
    iv_names = {}
9683

    
9684
    for idx, dev in enumerate(self.instance.disks):
9685
      if idx not in self.disks:
9686
        continue
9687

    
9688
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
9689

    
9690
      self.cfg.SetDiskID(dev, node_name)
9691

    
9692
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
9693
      names = _GenerateUniqueNames(self.lu, lv_names)
9694

    
9695
      vg_data = dev.children[0].logical_id[0]
9696
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
9697
                             logical_id=(vg_data, names[0]))
9698
      vg_meta = dev.children[1].logical_id[0]
9699
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
9700
                             logical_id=(vg_meta, names[1]))
9701

    
9702
      new_lvs = [lv_data, lv_meta]
9703
      old_lvs = [child.Copy() for child in dev.children]
9704
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
9705

    
9706
      # we pass force_create=True to force the LVM creation
9707
      for new_lv in new_lvs:
9708
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
9709
                        _GetInstanceInfoText(self.instance), False)
9710

    
9711
    return iv_names
9712

    
9713
  def _CheckDevices(self, node_name, iv_names):
9714
    for name, (dev, _, _) in iv_names.iteritems():
9715
      self.cfg.SetDiskID(dev, node_name)
9716

    
9717
      result = self.rpc.call_blockdev_find(node_name, dev)
9718

    
9719
      msg = result.fail_msg
9720
      if msg or not result.payload:
9721
        if not msg:
9722
          msg = "disk not found"
9723
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
9724
                                 (name, msg))
9725

    
9726
      if result.payload.is_degraded:
9727
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
9728

    
9729
  def _RemoveOldStorage(self, node_name, iv_names):
9730
    for name, (_, old_lvs, _) in iv_names.iteritems():
9731
      self.lu.LogInfo("Remove logical volumes for %s" % name)
9732

    
9733
      for lv in old_lvs:
9734
        self.cfg.SetDiskID(lv, node_name)
9735

    
9736
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
9737
        if msg:
9738
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
9739
                             hint="remove unused LVs manually")
9740

    
9741
  def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
9742
    """Replace a disk on the primary or secondary for DRBD 8.
9743

9744
    The algorithm for replace is quite complicated:
9745

9746
      1. for each disk to be replaced:
9747

9748
        1. create new LVs on the target node with unique names
9749
        1. detach old LVs from the drbd device
9750
        1. rename old LVs to name_replaced.<time_t>
9751
        1. rename new LVs to old LVs
9752
        1. attach the new LVs (with the old names now) to the drbd device
9753

9754
      1. wait for sync across all devices
9755

9756
      1. for each modified disk:
9757

9758
        1. remove old LVs (which have the name name_replaces.<time_t>)
9759

9760
    Failures are not very well handled.
9761

9762
    """
9763
    steps_total = 6
9764

    
9765
    # Step: check device activation
9766
    self.lu.LogStep(1, steps_total, "Check device existence")
9767
    self._CheckDisksExistence([self.other_node, self.target_node])
9768
    self._CheckVolumeGroup([self.target_node, self.other_node])
9769

    
9770
    # Step: check other node consistency
9771
    self.lu.LogStep(2, steps_total, "Check peer consistency")
9772
    self._CheckDisksConsistency(self.other_node,
9773
                                self.other_node == self.instance.primary_node,
9774
                                False)
9775

    
9776
    # Step: create new storage
9777
    self.lu.LogStep(3, steps_total, "Allocate new storage")
9778
    iv_names = self._CreateNewStorage(self.target_node)
9779

    
9780
    # Step: for each lv, detach+rename*2+attach
9781
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
9782
    for dev, old_lvs, new_lvs in iv_names.itervalues():
9783
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
9784

    
9785
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
9786
                                                     old_lvs)
9787
      result.Raise("Can't detach drbd from local storage on node"
9788
                   " %s for device %s" % (self.target_node, dev.iv_name))
9789
      #dev.children = []
9790
      #cfg.Update(instance)
9791

    
9792
      # ok, we created the new LVs, so now we know we have the needed
9793
      # storage; as such, we proceed on the target node to rename
9794
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
9795
      # using the assumption that logical_id == physical_id (which in
9796
      # turn is the unique_id on that node)
9797

    
9798
      # FIXME(iustin): use a better name for the replaced LVs
9799
      temp_suffix = int(time.time())
9800
      ren_fn = lambda d, suff: (d.physical_id[0],
9801
                                d.physical_id[1] + "_replaced-%s" % suff)
9802

    
9803
      # Build the rename list based on what LVs exist on the node
9804
      rename_old_to_new = []
9805
      for to_ren in old_lvs:
9806
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
9807
        if not result.fail_msg and result.payload:
9808
          # device exists
9809
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
9810

    
9811
      self.lu.LogInfo("Renaming the old LVs on the target node")
9812
      result = self.rpc.call_blockdev_rename(self.target_node,
9813
                                             rename_old_to_new)
9814
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
9815

    
9816
      # Now we rename the new LVs to the old LVs
9817
      self.lu.LogInfo("Renaming the new LVs on the target node")
9818
      rename_new_to_old = [(new, old.physical_id)
9819
                           for old, new in zip(old_lvs, new_lvs)]
9820
      result = self.rpc.call_blockdev_rename(self.target_node,
9821
                                             rename_new_to_old)
9822
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
9823

    
9824
      # Intermediate steps of in memory modifications
9825
      for old, new in zip(old_lvs, new_lvs):
9826
        new.logical_id = old.logical_id
9827
        self.cfg.SetDiskID(new, self.target_node)
9828

    
9829
      # We need to modify old_lvs so that removal later removes the
9830
      # right LVs, not the newly added ones; note that old_lvs is a
9831
      # copy here
9832
      for disk in old_lvs:
9833
        disk.logical_id = ren_fn(disk, temp_suffix)
9834
        self.cfg.SetDiskID(disk, self.target_node)
9835

    
9836
      # Now that the new lvs have the old name, we can add them to the device
9837
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
9838
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
9839
                                                  new_lvs)
9840
      msg = result.fail_msg
9841
      if msg:
9842
        for new_lv in new_lvs:
9843
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
9844
                                               new_lv).fail_msg
9845
          if msg2:
9846
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
9847
                               hint=("cleanup manually the unused logical"
9848
                                     "volumes"))
9849
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
9850

    
9851
    cstep = 5
9852
    if self.early_release:
9853
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
9854
      cstep += 1
9855
      self._RemoveOldStorage(self.target_node, iv_names)
9856
      # WARNING: we release both node locks here, do not do other RPCs
9857
      # than WaitForSync to the primary node
9858
      _ReleaseLocks(self.lu, locking.LEVEL_NODE,
9859
                    names=[self.target_node, self.other_node])
9860

    
9861
    # Wait for sync
9862
    # This can fail as the old devices are degraded and _WaitForSync
9863
    # does a combined result over all disks, so we don't check its return value
9864
    self.lu.LogStep(cstep, steps_total, "Sync devices")
9865
    cstep += 1
9866
    _WaitForSync(self.lu, self.instance)
9867

    
9868
    # Check all devices manually
9869
    self._CheckDevices(self.instance.primary_node, iv_names)
9870

    
9871
    # Step: remove old storage
9872
    if not self.early_release:
9873
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
9874
      cstep += 1
9875
      self._RemoveOldStorage(self.target_node, iv_names)
9876

    
9877
  def _ExecDrbd8Secondary(self, feedback_fn):
9878
    """Replace the secondary node for DRBD 8.
9879

9880
    The algorithm for replace is quite complicated:
9881
      - for all disks of the instance:
9882
        - create new LVs on the new node with same names
9883
        - shutdown the drbd device on the old secondary
9884
        - disconnect the drbd network on the primary
9885
        - create the drbd device on the new secondary
9886
        - network attach the drbd on the primary, using an artifice:
9887
          the drbd code for Attach() will connect to the network if it
9888
          finds a device which is connected to the good local disks but
9889
          not network enabled
9890
      - wait for sync across all devices
9891
      - remove all disks from the old secondary
9892

9893
    Failures are not very well handled.
9894

9895
    """
9896
    steps_total = 6
9897

    
9898
    pnode = self.instance.primary_node
9899

    
9900
    # Step: check device activation
9901
    self.lu.LogStep(1, steps_total, "Check device existence")
9902
    self._CheckDisksExistence([self.instance.primary_node])
9903
    self._CheckVolumeGroup([self.instance.primary_node])
9904

    
9905
    # Step: check other node consistency
9906
    self.lu.LogStep(2, steps_total, "Check peer consistency")
9907
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
9908

    
9909
    # Step: create new storage
9910
    self.lu.LogStep(3, steps_total, "Allocate new storage")
9911
    for idx, dev in enumerate(self.instance.disks):
9912
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
9913
                      (self.new_node, idx))
9914
      # we pass force_create=True to force LVM creation
9915
      for new_lv in dev.children:
9916
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
9917
                        _GetInstanceInfoText(self.instance), False)
9918

    
9919
    # Step 4: dbrd minors and drbd setups changes
9920
    # after this, we must manually remove the drbd minors on both the
9921
    # error and the success paths
9922
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
9923
    minors = self.cfg.AllocateDRBDMinor([self.new_node
9924
                                         for dev in self.instance.disks],
9925
                                        self.instance.name)
9926
    logging.debug("Allocated minors %r", minors)
9927

    
9928
    iv_names = {}
9929
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
9930
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
9931
                      (self.new_node, idx))
9932
      # create new devices on new_node; note that we create two IDs:
9933
      # one without port, so the drbd will be activated without
9934
      # networking information on the new node at this stage, and one
9935
      # with network, for the latter activation in step 4
9936
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
9937
      if self.instance.primary_node == o_node1:
9938
        p_minor = o_minor1
9939
      else:
9940
        assert self.instance.primary_node == o_node2, "Three-node instance?"
9941
        p_minor = o_minor2
9942

    
9943
      new_alone_id = (self.instance.primary_node, self.new_node, None,
9944
                      p_minor, new_minor, o_secret)
9945
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
9946
                    p_minor, new_minor, o_secret)
9947

    
9948
      iv_names[idx] = (dev, dev.children, new_net_id)
9949
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
9950
                    new_net_id)
9951
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
9952
                              logical_id=new_alone_id,
9953
                              children=dev.children,
9954
                              size=dev.size)
9955
      try:
9956
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
9957
                              _GetInstanceInfoText(self.instance), False)
9958
      except errors.GenericError:
9959
        self.cfg.ReleaseDRBDMinors(self.instance.name)
9960
        raise
9961

    
9962
    # We have new devices, shutdown the drbd on the old secondary
9963
    for idx, dev in enumerate(self.instance.disks):
9964
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
9965
      self.cfg.SetDiskID(dev, self.target_node)
9966
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
9967
      if msg:
9968
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
9969
                           "node: %s" % (idx, msg),
9970
                           hint=("Please cleanup this device manually as"
9971
                                 " soon as possible"))
9972

    
9973
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
9974
    result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
9975
                                               self.instance.disks)[pnode]
9976

    
9977
    msg = result.fail_msg
9978
    if msg:
9979
      # detaches didn't succeed (unlikely)
9980
      self.cfg.ReleaseDRBDMinors(self.instance.name)
9981
      raise errors.OpExecError("Can't detach the disks from the network on"
9982
                               " old node: %s" % (msg,))
9983

    
9984
    # if we managed to detach at least one, we update all the disks of
9985
    # the instance to point to the new secondary
9986
    self.lu.LogInfo("Updating instance configuration")
9987
    for dev, _, new_logical_id in iv_names.itervalues():
9988
      dev.logical_id = new_logical_id
9989
      self.cfg.SetDiskID(dev, self.instance.primary_node)
9990

    
9991
    self.cfg.Update(self.instance, feedback_fn)
9992

    
9993
    # and now perform the drbd attach
9994
    self.lu.LogInfo("Attaching primary drbds to new secondary"
9995
                    " (standalone => connected)")
9996
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
9997
                                            self.new_node],
9998
                                           self.node_secondary_ip,
9999
                                           self.instance.disks,
10000
                                           self.instance.name,
10001
                                           False)
10002
    for to_node, to_result in result.items():
10003
      msg = to_result.fail_msg
10004
      if msg:
10005
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
10006
                           to_node, msg,
10007
                           hint=("please do a gnt-instance info to see the"
10008
                                 " status of disks"))
10009
    cstep = 5
10010
    if self.early_release:
10011
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
10012
      cstep += 1
10013
      self._RemoveOldStorage(self.target_node, iv_names)
10014
      # WARNING: we release all node locks here, do not do other RPCs
10015
      # than WaitForSync to the primary node
10016
      _ReleaseLocks(self.lu, locking.LEVEL_NODE,
10017
                    names=[self.instance.primary_node,
10018
                           self.target_node,
10019
                           self.new_node])
10020

    
10021
    # Wait for sync
10022
    # This can fail as the old devices are degraded and _WaitForSync
10023
    # does a combined result over all disks, so we don't check its return value
10024
    self.lu.LogStep(cstep, steps_total, "Sync devices")
10025
    cstep += 1
10026
    _WaitForSync(self.lu, self.instance)
10027

    
10028
    # Check all devices manually
10029
    self._CheckDevices(self.instance.primary_node, iv_names)
10030

    
10031
    # Step: remove old storage
10032
    if not self.early_release:
10033
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
10034
      self._RemoveOldStorage(self.target_node, iv_names)
10035

    
10036

    
10037
class LURepairNodeStorage(NoHooksLU):
10038
  """Repairs the volume group on a node.
10039

10040
  """
10041
  REQ_BGL = False
10042

    
10043
  def CheckArguments(self):
10044
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
10045

    
10046
    storage_type = self.op.storage_type
10047

    
10048
    if (constants.SO_FIX_CONSISTENCY not in
10049
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
10050
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
10051
                                 " repaired" % storage_type,
10052
                                 errors.ECODE_INVAL)
10053

    
10054
  def ExpandNames(self):
10055
    self.needed_locks = {
10056
      locking.LEVEL_NODE: [self.op.node_name],
10057
      }
10058

    
10059
  def _CheckFaultyDisks(self, instance, node_name):
10060
    """Ensure faulty disks abort the opcode or at least warn."""
10061
    try:
10062
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
10063
                                  node_name, True):
10064
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
10065
                                   " node '%s'" % (instance.name, node_name),
10066
                                   errors.ECODE_STATE)
10067
    except errors.OpPrereqError, err:
10068
      if self.op.ignore_consistency:
10069
        self.proc.LogWarning(str(err.args[0]))
10070
      else:
10071
        raise
10072

    
10073
  def CheckPrereq(self):
10074
    """Check prerequisites.
10075

10076
    """
10077
    # Check whether any instance on this node has faulty disks
10078
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
10079
      if not inst.admin_up:
10080
        continue
10081
      check_nodes = set(inst.all_nodes)
10082
      check_nodes.discard(self.op.node_name)
10083
      for inst_node_name in check_nodes:
10084
        self._CheckFaultyDisks(inst, inst_node_name)
10085

    
10086
  def Exec(self, feedback_fn):
10087
    feedback_fn("Repairing storage unit '%s' on %s ..." %
10088
                (self.op.name, self.op.node_name))
10089

    
10090
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
10091
    result = self.rpc.call_storage_execute(self.op.node_name,
10092
                                           self.op.storage_type, st_args,
10093
                                           self.op.name,
10094
                                           constants.SO_FIX_CONSISTENCY)
10095
    result.Raise("Failed to repair storage unit '%s' on %s" %
10096
                 (self.op.name, self.op.node_name))
10097

    
10098

    
10099
class LUNodeEvacuate(NoHooksLU):
10100
  """Evacuates instances off a list of nodes.
10101

10102
  """
10103
  REQ_BGL = False
10104

    
10105
  def CheckArguments(self):
10106
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
10107

    
10108
  def ExpandNames(self):
10109
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
10110

    
10111
    if self.op.remote_node is not None:
10112
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
10113
      assert self.op.remote_node
10114

    
10115
      if self.op.remote_node == self.op.node_name:
10116
        raise errors.OpPrereqError("Can not use evacuated node as a new"
10117
                                   " secondary node", errors.ECODE_INVAL)
10118

    
10119
      if self.op.mode != constants.IALLOCATOR_NEVAC_SEC:
10120
        raise errors.OpPrereqError("Without the use of an iallocator only"
10121
                                   " secondary instances can be evacuated",
10122
                                   errors.ECODE_INVAL)
10123

    
10124
    # Declare locks
10125
    self.share_locks = _ShareAll()
10126
    self.needed_locks = {
10127
      locking.LEVEL_INSTANCE: [],
10128
      locking.LEVEL_NODEGROUP: [],
10129
      locking.LEVEL_NODE: [],
10130
      }
10131

    
10132
    # Determine nodes (via group) optimistically, needs verification once locks
10133
    # have been acquired
10134
    self.lock_nodes = self._DetermineNodes()
10135

    
10136
  def _DetermineNodes(self):
10137
    """Gets the list of nodes to operate on.
10138

10139
    """
10140
    if self.op.remote_node is None:
10141
      # Iallocator will choose any node(s) in the same group
10142
      group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
10143
    else:
10144
      group_nodes = frozenset([self.op.remote_node])
10145

    
10146
    # Determine nodes to be locked
10147
    return set([self.op.node_name]) | group_nodes
10148

    
10149
  def _DetermineInstances(self):
10150
    """Builds list of instances to operate on.
10151

10152
    """
10153
    assert self.op.mode in constants.IALLOCATOR_NEVAC_MODES
10154

    
10155
    if self.op.mode == constants.IALLOCATOR_NEVAC_PRI:
10156
      # Primary instances only
10157
      inst_fn = _GetNodePrimaryInstances
10158
      assert self.op.remote_node is None, \
10159
        "Evacuating primary instances requires iallocator"
10160
    elif self.op.mode == constants.IALLOCATOR_NEVAC_SEC:
10161
      # Secondary instances only
10162
      inst_fn = _GetNodeSecondaryInstances
10163
    else:
10164
      # All instances
10165
      assert self.op.mode == constants.IALLOCATOR_NEVAC_ALL
10166
      inst_fn = _GetNodeInstances
10167
      # TODO: In 2.6, change the iallocator interface to take an evacuation mode
10168
      # per instance
10169
      raise errors.OpPrereqError("Due to an issue with the iallocator"
10170
                                 " interface it is not possible to evacuate"
10171
                                 " all instances at once; specify explicitly"
10172
                                 " whether to evacuate primary or secondary"
10173
                                 " instances",
10174
                                 errors.ECODE_INVAL)
10175

    
10176
    return inst_fn(self.cfg, self.op.node_name)
10177

    
10178
  def DeclareLocks(self, level):
10179
    if level == locking.LEVEL_INSTANCE:
10180
      # Lock instances optimistically, needs verification once node and group
10181
      # locks have been acquired
10182
      self.needed_locks[locking.LEVEL_INSTANCE] = \
10183
        set(i.name for i in self._DetermineInstances())
10184

    
10185
    elif level == locking.LEVEL_NODEGROUP:
10186
      # Lock node groups for all potential target nodes optimistically, needs
10187
      # verification once nodes have been acquired
10188
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
10189
        self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
10190

    
10191
    elif level == locking.LEVEL_NODE:
10192
      self.needed_locks[locking.LEVEL_NODE] = self.lock_nodes
10193

    
10194
  def CheckPrereq(self):
10195
    # Verify locks
10196
    owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
10197
    owned_nodes = self.owned_locks(locking.LEVEL_NODE)
10198
    owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
10199

    
10200
    need_nodes = self._DetermineNodes()
10201

    
10202
    if not owned_nodes.issuperset(need_nodes):
10203
      raise errors.OpPrereqError("Nodes in same group as '%s' changed since"
10204
                                 " locks were acquired, current nodes are"
10205
                                 " are '%s', used to be '%s'; retry the"
10206
                                 " operation" %
10207
                                 (self.op.node_name,
10208
                                  utils.CommaJoin(need_nodes),
10209
                                  utils.CommaJoin(owned_nodes)),
10210
                                 errors.ECODE_STATE)
10211

    
10212
    wanted_groups = self.cfg.GetNodeGroupsFromNodes(owned_nodes)
10213
    if owned_groups != wanted_groups:
10214
      raise errors.OpExecError("Node groups changed since locks were acquired,"
10215
                               " current groups are '%s', used to be '%s';"
10216
                               " retry the operation" %
10217
                               (utils.CommaJoin(wanted_groups),
10218
                                utils.CommaJoin(owned_groups)))
10219

    
10220
    # Determine affected instances
10221
    self.instances = self._DetermineInstances()
10222
    self.instance_names = [i.name for i in self.instances]
10223

    
10224
    if set(self.instance_names) != owned_instances:
10225
      raise errors.OpExecError("Instances on node '%s' changed since locks"
10226
                               " were acquired, current instances are '%s',"
10227
                               " used to be '%s'; retry the operation" %
10228
                               (self.op.node_name,
10229
                                utils.CommaJoin(self.instance_names),
10230
                                utils.CommaJoin(owned_instances)))
10231

    
10232
    if self.instance_names:
10233
      self.LogInfo("Evacuating instances from node '%s': %s",
10234
                   self.op.node_name,
10235
                   utils.CommaJoin(utils.NiceSort(self.instance_names)))
10236
    else:
10237
      self.LogInfo("No instances to evacuate from node '%s'",
10238
                   self.op.node_name)
10239

    
10240
    if self.op.remote_node is not None:
10241
      for i in self.instances:
10242
        if i.primary_node == self.op.remote_node:
10243
          raise errors.OpPrereqError("Node %s is the primary node of"
10244
                                     " instance %s, cannot use it as"
10245
                                     " secondary" %
10246
                                     (self.op.remote_node, i.name),
10247
                                     errors.ECODE_INVAL)
10248

    
10249
  def Exec(self, feedback_fn):
10250
    assert (self.op.iallocator is not None) ^ (self.op.remote_node is not None)
10251

    
10252
    if not self.instance_names:
10253
      # No instances to evacuate
10254
      jobs = []
10255

    
10256
    elif self.op.iallocator is not None:
10257
      # TODO: Implement relocation to other group
10258
      ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_NODE_EVAC,
10259
                       evac_mode=self.op.mode,
10260
                       instances=list(self.instance_names))
10261

    
10262
      ial.Run(self.op.iallocator)
10263

    
10264
      if not ial.success:
10265
        raise errors.OpPrereqError("Can't compute node evacuation using"
10266
                                   " iallocator '%s': %s" %
10267
                                   (self.op.iallocator, ial.info),
10268
                                   errors.ECODE_NORES)
10269

    
10270
      jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, True)
10271

    
10272
    elif self.op.remote_node is not None:
10273
      assert self.op.mode == constants.IALLOCATOR_NEVAC_SEC
10274
      jobs = [
10275
        [opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
10276
                                        remote_node=self.op.remote_node,
10277
                                        disks=[],
10278
                                        mode=constants.REPLACE_DISK_CHG,
10279
                                        early_release=self.op.early_release)]
10280
        for instance_name in self.instance_names
10281
        ]
10282

    
10283
    else:
10284
      raise errors.ProgrammerError("No iallocator or remote node")
10285

    
10286
    return ResultWithJobs(jobs)
10287

    
10288

    
10289
def _SetOpEarlyRelease(early_release, op):
10290
  """Sets C{early_release} flag on opcodes if available.
10291

10292
  """
10293
  try:
10294
    op.early_release = early_release
10295
  except AttributeError:
10296
    assert not isinstance(op, opcodes.OpInstanceReplaceDisks)
10297

    
10298
  return op
10299

    
10300

    
10301
def _NodeEvacDest(use_nodes, group, nodes):
10302
  """Returns group or nodes depending on caller's choice.
10303

10304
  """
10305
  if use_nodes:
10306
    return utils.CommaJoin(nodes)
10307
  else:
10308
    return group
10309

    
10310

    
10311
def _LoadNodeEvacResult(lu, alloc_result, early_release, use_nodes):
10312
  """Unpacks the result of change-group and node-evacuate iallocator requests.
10313

10314
  Iallocator modes L{constants.IALLOCATOR_MODE_NODE_EVAC} and
10315
  L{constants.IALLOCATOR_MODE_CHG_GROUP}.
10316

10317
  @type lu: L{LogicalUnit}
10318
  @param lu: Logical unit instance
10319
  @type alloc_result: tuple/list
10320
  @param alloc_result: Result from iallocator
10321
  @type early_release: bool
10322
  @param early_release: Whether to release locks early if possible
10323
  @type use_nodes: bool
10324
  @param use_nodes: Whether to display node names instead of groups
10325

10326
  """
10327
  (moved, failed, jobs) = alloc_result
10328

    
10329
  if failed:
10330
    failreason = utils.CommaJoin("%s (%s)" % (name, reason)
10331
                                 for (name, reason) in failed)
10332
    lu.LogWarning("Unable to evacuate instances %s", failreason)
10333
    raise errors.OpExecError("Unable to evacuate instances %s" % failreason)
10334

    
10335
  if moved:
10336
    lu.LogInfo("Instances to be moved: %s",
10337
               utils.CommaJoin("%s (to %s)" %
10338
                               (name, _NodeEvacDest(use_nodes, group, nodes))
10339
                               for (name, group, nodes) in moved))
10340

    
10341
  return [map(compat.partial(_SetOpEarlyRelease, early_release),
10342
              map(opcodes.OpCode.LoadOpCode, ops))
10343
          for ops in jobs]
10344

    
10345

    
10346
class LUInstanceGrowDisk(LogicalUnit):
10347
  """Grow a disk of an instance.
10348

10349
  """
10350
  HPATH = "disk-grow"
10351
  HTYPE = constants.HTYPE_INSTANCE
10352
  REQ_BGL = False
10353

    
10354
  def ExpandNames(self):
10355
    self._ExpandAndLockInstance()
10356
    self.needed_locks[locking.LEVEL_NODE] = []
10357
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10358

    
10359
  def DeclareLocks(self, level):
10360
    if level == locking.LEVEL_NODE:
10361
      self._LockInstancesNodes()
10362

    
10363
  def BuildHooksEnv(self):
10364
    """Build hooks env.
10365

10366
    This runs on the master, the primary and all the secondaries.
10367

10368
    """
10369
    env = {
10370
      "DISK": self.op.disk,
10371
      "AMOUNT": self.op.amount,
10372
      }
10373
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
10374
    return env
10375

    
10376
  def BuildHooksNodes(self):
10377
    """Build hooks nodes.
10378

10379
    """
10380
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
10381
    return (nl, nl)
10382

    
10383
  def CheckPrereq(self):
10384
    """Check prerequisites.
10385

10386
    This checks that the instance is in the cluster.
10387

10388
    """
10389
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
10390
    assert instance is not None, \
10391
      "Cannot retrieve locked instance %s" % self.op.instance_name
10392
    nodenames = list(instance.all_nodes)
10393
    for node in nodenames:
10394
      _CheckNodeOnline(self, node)
10395

    
10396
    self.instance = instance
10397

    
10398
    if instance.disk_template not in constants.DTS_GROWABLE:
10399
      raise errors.OpPrereqError("Instance's disk layout does not support"
10400
                                 " growing", errors.ECODE_INVAL)
10401

    
10402
    self.disk = instance.FindDisk(self.op.disk)
10403

    
10404
    if instance.disk_template not in (constants.DT_FILE,
10405
                                      constants.DT_SHARED_FILE):
10406
      # TODO: check the free disk space for file, when that feature will be
10407
      # supported
10408
      _CheckNodesFreeDiskPerVG(self, nodenames,
10409
                               self.disk.ComputeGrowth(self.op.amount))
10410

    
10411
  def Exec(self, feedback_fn):
10412
    """Execute disk grow.
10413

10414
    """
10415
    instance = self.instance
10416
    disk = self.disk
10417

    
10418
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
10419
    if not disks_ok:
10420
      raise errors.OpExecError("Cannot activate block device to grow")
10421

    
10422
    # First run all grow ops in dry-run mode
10423
    for node in instance.all_nodes:
10424
      self.cfg.SetDiskID(disk, node)
10425
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, True)
10426
      result.Raise("Grow request failed to node %s" % node)
10427

    
10428
    # We know that (as far as we can test) operations across different
10429
    # nodes will succeed, time to run it for real
10430
    for node in instance.all_nodes:
10431
      self.cfg.SetDiskID(disk, node)
10432
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, False)
10433
      result.Raise("Grow request failed to node %s" % node)
10434

    
10435
      # TODO: Rewrite code to work properly
10436
      # DRBD goes into sync mode for a short amount of time after executing the
10437
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
10438
      # calling "resize" in sync mode fails. Sleeping for a short amount of
10439
      # time is a work-around.
10440
      time.sleep(5)
10441

    
10442
    disk.RecordGrow(self.op.amount)
10443
    self.cfg.Update(instance, feedback_fn)
10444
    if self.op.wait_for_sync:
10445
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
10446
      if disk_abort:
10447
        self.proc.LogWarning("Disk sync-ing has not returned a good"
10448
                             " status; please check the instance")
10449
      if not instance.admin_up:
10450
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
10451
    elif not instance.admin_up:
10452
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
10453
                           " not supposed to be running because no wait for"
10454
                           " sync mode was requested")
10455

    
10456

    
10457
class LUInstanceQueryData(NoHooksLU):
10458
  """Query runtime instance data.
10459

10460
  """
10461
  REQ_BGL = False
10462

    
10463
  def ExpandNames(self):
10464
    self.needed_locks = {}
10465

    
10466
    # Use locking if requested or when non-static information is wanted
10467
    if not (self.op.static or self.op.use_locking):
10468
      self.LogWarning("Non-static data requested, locks need to be acquired")
10469
      self.op.use_locking = True
10470

    
10471
    if self.op.instances or not self.op.use_locking:
10472
      # Expand instance names right here
10473
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
10474
    else:
10475
      # Will use acquired locks
10476
      self.wanted_names = None
10477

    
10478
    if self.op.use_locking:
10479
      self.share_locks = _ShareAll()
10480

    
10481
      if self.wanted_names is None:
10482
        self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
10483
      else:
10484
        self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
10485

    
10486
      self.needed_locks[locking.LEVEL_NODE] = []
10487
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10488

    
10489
  def DeclareLocks(self, level):
10490
    if self.op.use_locking and level == locking.LEVEL_NODE:
10491
      self._LockInstancesNodes()
10492

    
10493
  def CheckPrereq(self):
10494
    """Check prerequisites.
10495

10496
    This only checks the optional instance list against the existing names.
10497

10498
    """
10499
    if self.wanted_names is None:
10500
      assert self.op.use_locking, "Locking was not used"
10501
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
10502

    
10503
    self.wanted_instances = \
10504
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
10505

    
10506
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
10507
    """Returns the status of a block device
10508

10509
    """
10510
    if self.op.static or not node:
10511
      return None
10512

    
10513
    self.cfg.SetDiskID(dev, node)
10514

    
10515
    result = self.rpc.call_blockdev_find(node, dev)
10516
    if result.offline:
10517
      return None
10518

    
10519
    result.Raise("Can't compute disk status for %s" % instance_name)
10520

    
10521
    status = result.payload
10522
    if status is None:
10523
      return None
10524

    
10525
    return (status.dev_path, status.major, status.minor,
10526
            status.sync_percent, status.estimated_time,
10527
            status.is_degraded, status.ldisk_status)
10528

    
10529
  def _ComputeDiskStatus(self, instance, snode, dev):
10530
    """Compute block device status.
10531

10532
    """
10533
    if dev.dev_type in constants.LDS_DRBD:
10534
      # we change the snode then (otherwise we use the one passed in)
10535
      if dev.logical_id[0] == instance.primary_node:
10536
        snode = dev.logical_id[1]
10537
      else:
10538
        snode = dev.logical_id[0]
10539

    
10540
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
10541
                                              instance.name, dev)
10542
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
10543

    
10544
    if dev.children:
10545
      dev_children = map(compat.partial(self._ComputeDiskStatus,
10546
                                        instance, snode),
10547
                         dev.children)
10548
    else:
10549
      dev_children = []
10550

    
10551
    return {
10552
      "iv_name": dev.iv_name,
10553
      "dev_type": dev.dev_type,
10554
      "logical_id": dev.logical_id,
10555
      "physical_id": dev.physical_id,
10556
      "pstatus": dev_pstatus,
10557
      "sstatus": dev_sstatus,
10558
      "children": dev_children,
10559
      "mode": dev.mode,
10560
      "size": dev.size,
10561
      }
10562

    
10563
  def Exec(self, feedback_fn):
10564
    """Gather and return data"""
10565
    result = {}
10566

    
10567
    cluster = self.cfg.GetClusterInfo()
10568

    
10569
    pri_nodes = self.cfg.GetMultiNodeInfo(i.primary_node
10570
                                          for i in self.wanted_instances)
10571
    for instance, (_, pnode) in zip(self.wanted_instances, pri_nodes):
10572
      if self.op.static or pnode.offline:
10573
        remote_state = None
10574
        if pnode.offline:
10575
          self.LogWarning("Primary node %s is marked offline, returning static"
10576
                          " information only for instance %s" %
10577
                          (pnode.name, instance.name))
10578
      else:
10579
        remote_info = self.rpc.call_instance_info(instance.primary_node,
10580
                                                  instance.name,
10581
                                                  instance.hypervisor)
10582
        remote_info.Raise("Error checking node %s" % instance.primary_node)
10583
        remote_info = remote_info.payload
10584
        if remote_info and "state" in remote_info:
10585
          remote_state = "up"
10586
        else:
10587
          remote_state = "down"
10588

    
10589
      if instance.admin_up:
10590
        config_state = "up"
10591
      else:
10592
        config_state = "down"
10593

    
10594
      disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
10595
                  instance.disks)
10596

    
10597
      result[instance.name] = {
10598
        "name": instance.name,
10599
        "config_state": config_state,
10600
        "run_state": remote_state,
10601
        "pnode": instance.primary_node,
10602
        "snodes": instance.secondary_nodes,
10603
        "os": instance.os,
10604
        # this happens to be the same format used for hooks
10605
        "nics": _NICListToTuple(self, instance.nics),
10606
        "disk_template": instance.disk_template,
10607
        "disks": disks,
10608
        "hypervisor": instance.hypervisor,
10609
        "network_port": instance.network_port,
10610
        "hv_instance": instance.hvparams,
10611
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
10612
        "be_instance": instance.beparams,
10613
        "be_actual": cluster.FillBE(instance),
10614
        "os_instance": instance.osparams,
10615
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
10616
        "serial_no": instance.serial_no,
10617
        "mtime": instance.mtime,
10618
        "ctime": instance.ctime,
10619
        "uuid": instance.uuid,
10620
        }
10621

    
10622
    return result
10623

    
10624

    
10625
class LUInstanceSetParams(LogicalUnit):
10626
  """Modifies an instances's parameters.
10627

10628
  """
10629
  HPATH = "instance-modify"
10630
  HTYPE = constants.HTYPE_INSTANCE
10631
  REQ_BGL = False
10632

    
10633
  def CheckArguments(self):
10634
    if not (self.op.nics or self.op.disks or self.op.disk_template or
10635
            self.op.hvparams or self.op.beparams or self.op.os_name):
10636
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
10637

    
10638
    if self.op.hvparams:
10639
      _CheckGlobalHvParams(self.op.hvparams)
10640

    
10641
    # Disk validation
10642
    disk_addremove = 0
10643
    for disk_op, disk_dict in self.op.disks:
10644
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
10645
      if disk_op == constants.DDM_REMOVE:
10646
        disk_addremove += 1
10647
        continue
10648
      elif disk_op == constants.DDM_ADD:
10649
        disk_addremove += 1
10650
      else:
10651
        if not isinstance(disk_op, int):
10652
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
10653
        if not isinstance(disk_dict, dict):
10654
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
10655
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
10656

    
10657
      if disk_op == constants.DDM_ADD:
10658
        mode = disk_dict.setdefault(constants.IDISK_MODE, constants.DISK_RDWR)
10659
        if mode not in constants.DISK_ACCESS_SET:
10660
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
10661
                                     errors.ECODE_INVAL)
10662
        size = disk_dict.get(constants.IDISK_SIZE, None)
10663
        if size is None:
10664
          raise errors.OpPrereqError("Required disk parameter size missing",
10665
                                     errors.ECODE_INVAL)
10666
        try:
10667
          size = int(size)
10668
        except (TypeError, ValueError), err:
10669
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
10670
                                     str(err), errors.ECODE_INVAL)
10671
        disk_dict[constants.IDISK_SIZE] = size
10672
      else:
10673
        # modification of disk
10674
        if constants.IDISK_SIZE in disk_dict:
10675
          raise errors.OpPrereqError("Disk size change not possible, use"
10676
                                     " grow-disk", errors.ECODE_INVAL)
10677

    
10678
    if disk_addremove > 1:
10679
      raise errors.OpPrereqError("Only one disk add or remove operation"
10680
                                 " supported at a time", errors.ECODE_INVAL)
10681

    
10682
    if self.op.disks and self.op.disk_template is not None:
10683
      raise errors.OpPrereqError("Disk template conversion and other disk"
10684
                                 " changes not supported at the same time",
10685
                                 errors.ECODE_INVAL)
10686

    
10687
    if (self.op.disk_template and
10688
        self.op.disk_template in constants.DTS_INT_MIRROR and
10689
        self.op.remote_node is None):
10690
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
10691
                                 " one requires specifying a secondary node",
10692
                                 errors.ECODE_INVAL)
10693

    
10694
    # NIC validation
10695
    nic_addremove = 0
10696
    for nic_op, nic_dict in self.op.nics:
10697
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
10698
      if nic_op == constants.DDM_REMOVE:
10699
        nic_addremove += 1
10700
        continue
10701
      elif nic_op == constants.DDM_ADD:
10702
        nic_addremove += 1
10703
      else:
10704
        if not isinstance(nic_op, int):
10705
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
10706
        if not isinstance(nic_dict, dict):
10707
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
10708
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
10709

    
10710
      # nic_dict should be a dict
10711
      nic_ip = nic_dict.get(constants.INIC_IP, None)
10712
      if nic_ip is not None:
10713
        if nic_ip.lower() == constants.VALUE_NONE:
10714
          nic_dict[constants.INIC_IP] = None
10715
        else:
10716
          if not netutils.IPAddress.IsValid(nic_ip):
10717
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
10718
                                       errors.ECODE_INVAL)
10719

    
10720
      nic_bridge = nic_dict.get("bridge", None)
10721
      nic_link = nic_dict.get(constants.INIC_LINK, None)
10722
      if nic_bridge and nic_link:
10723
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
10724
                                   " at the same time", errors.ECODE_INVAL)
10725
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
10726
        nic_dict["bridge"] = None
10727
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
10728
        nic_dict[constants.INIC_LINK] = None
10729

    
10730
      if nic_op == constants.DDM_ADD:
10731
        nic_mac = nic_dict.get(constants.INIC_MAC, None)
10732
        if nic_mac is None:
10733
          nic_dict[constants.INIC_MAC] = constants.VALUE_AUTO
10734

    
10735
      if constants.INIC_MAC in nic_dict:
10736
        nic_mac = nic_dict[constants.INIC_MAC]
10737
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
10738
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
10739

    
10740
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
10741
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
10742
                                     " modifying an existing nic",
10743
                                     errors.ECODE_INVAL)
10744

    
10745
    if nic_addremove > 1:
10746
      raise errors.OpPrereqError("Only one NIC add or remove operation"
10747
                                 " supported at a time", errors.ECODE_INVAL)
10748

    
10749
  def ExpandNames(self):
10750
    self._ExpandAndLockInstance()
10751
    self.needed_locks[locking.LEVEL_NODE] = []
10752
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10753

    
10754
  def DeclareLocks(self, level):
10755
    if level == locking.LEVEL_NODE:
10756
      self._LockInstancesNodes()
10757
      if self.op.disk_template and self.op.remote_node:
10758
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
10759
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
10760

    
10761
  def BuildHooksEnv(self):
10762
    """Build hooks env.
10763

10764
    This runs on the master, primary and secondaries.
10765

10766
    """
10767
    args = dict()
10768
    if constants.BE_MEMORY in self.be_new:
10769
      args["memory"] = self.be_new[constants.BE_MEMORY]
10770
    if constants.BE_VCPUS in self.be_new:
10771
      args["vcpus"] = self.be_new[constants.BE_VCPUS]
10772
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
10773
    # information at all.
10774
    if self.op.nics:
10775
      args["nics"] = []
10776
      nic_override = dict(self.op.nics)
10777
      for idx, nic in enumerate(self.instance.nics):
10778
        if idx in nic_override:
10779
          this_nic_override = nic_override[idx]
10780
        else:
10781
          this_nic_override = {}
10782
        if constants.INIC_IP in this_nic_override:
10783
          ip = this_nic_override[constants.INIC_IP]
10784
        else:
10785
          ip = nic.ip
10786
        if constants.INIC_MAC in this_nic_override:
10787
          mac = this_nic_override[constants.INIC_MAC]
10788
        else:
10789
          mac = nic.mac
10790
        if idx in self.nic_pnew:
10791
          nicparams = self.nic_pnew[idx]
10792
        else:
10793
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
10794
        mode = nicparams[constants.NIC_MODE]
10795
        link = nicparams[constants.NIC_LINK]
10796
        args["nics"].append((ip, mac, mode, link))
10797
      if constants.DDM_ADD in nic_override:
10798
        ip = nic_override[constants.DDM_ADD].get(constants.INIC_IP, None)
10799
        mac = nic_override[constants.DDM_ADD][constants.INIC_MAC]
10800
        nicparams = self.nic_pnew[constants.DDM_ADD]
10801
        mode = nicparams[constants.NIC_MODE]
10802
        link = nicparams[constants.NIC_LINK]
10803
        args["nics"].append((ip, mac, mode, link))
10804
      elif constants.DDM_REMOVE in nic_override:
10805
        del args["nics"][-1]
10806

    
10807
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
10808
    if self.op.disk_template:
10809
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
10810

    
10811
    return env
10812

    
10813
  def BuildHooksNodes(self):
10814
    """Build hooks nodes.
10815

10816
    """
10817
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
10818
    return (nl, nl)
10819

    
10820
  def CheckPrereq(self):
10821
    """Check prerequisites.
10822

10823
    This only checks the instance list against the existing names.
10824

10825
    """
10826
    # checking the new params on the primary/secondary nodes
10827

    
10828
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
10829
    cluster = self.cluster = self.cfg.GetClusterInfo()
10830
    assert self.instance is not None, \
10831
      "Cannot retrieve locked instance %s" % self.op.instance_name
10832
    pnode = instance.primary_node
10833
    nodelist = list(instance.all_nodes)
10834

    
10835
    # OS change
10836
    if self.op.os_name and not self.op.force:
10837
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
10838
                      self.op.force_variant)
10839
      instance_os = self.op.os_name
10840
    else:
10841
      instance_os = instance.os
10842

    
10843
    if self.op.disk_template:
10844
      if instance.disk_template == self.op.disk_template:
10845
        raise errors.OpPrereqError("Instance already has disk template %s" %
10846
                                   instance.disk_template, errors.ECODE_INVAL)
10847

    
10848
      if (instance.disk_template,
10849
          self.op.disk_template) not in self._DISK_CONVERSIONS:
10850
        raise errors.OpPrereqError("Unsupported disk template conversion from"
10851
                                   " %s to %s" % (instance.disk_template,
10852
                                                  self.op.disk_template),
10853
                                   errors.ECODE_INVAL)
10854
      _CheckInstanceDown(self, instance, "cannot change disk template")
10855
      if self.op.disk_template in constants.DTS_INT_MIRROR:
10856
        if self.op.remote_node == pnode:
10857
          raise errors.OpPrereqError("Given new secondary node %s is the same"
10858
                                     " as the primary node of the instance" %
10859
                                     self.op.remote_node, errors.ECODE_STATE)
10860
        _CheckNodeOnline(self, self.op.remote_node)
10861
        _CheckNodeNotDrained(self, self.op.remote_node)
10862
        # FIXME: here we assume that the old instance type is DT_PLAIN
10863
        assert instance.disk_template == constants.DT_PLAIN
10864
        disks = [{constants.IDISK_SIZE: d.size,
10865
                  constants.IDISK_VG: d.logical_id[0]}
10866
                 for d in instance.disks]
10867
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
10868
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
10869

    
10870
    # hvparams processing
10871
    if self.op.hvparams:
10872
      hv_type = instance.hypervisor
10873
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
10874
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
10875
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
10876

    
10877
      # local check
10878
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
10879
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
10880
      self.hv_new = hv_new # the new actual values
10881
      self.hv_inst = i_hvdict # the new dict (without defaults)
10882
    else:
10883
      self.hv_new = self.hv_inst = {}
10884

    
10885
    # beparams processing
10886
    if self.op.beparams:
10887
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
10888
                                   use_none=True)
10889
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
10890
      be_new = cluster.SimpleFillBE(i_bedict)
10891
      self.be_new = be_new # the new actual values
10892
      self.be_inst = i_bedict # the new dict (without defaults)
10893
    else:
10894
      self.be_new = self.be_inst = {}
10895
    be_old = cluster.FillBE(instance)
10896

    
10897
    # osparams processing
10898
    if self.op.osparams:
10899
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
10900
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
10901
      self.os_inst = i_osdict # the new dict (without defaults)
10902
    else:
10903
      self.os_inst = {}
10904

    
10905
    self.warn = []
10906

    
10907
    if (constants.BE_MEMORY in self.op.beparams and not self.op.force and
10908
        be_new[constants.BE_MEMORY] > be_old[constants.BE_MEMORY]):
10909
      mem_check_list = [pnode]
10910
      if be_new[constants.BE_AUTO_BALANCE]:
10911
        # either we changed auto_balance to yes or it was from before
10912
        mem_check_list.extend(instance.secondary_nodes)
10913
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
10914
                                                  instance.hypervisor)
10915
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
10916
                                         instance.hypervisor)
10917
      pninfo = nodeinfo[pnode]
10918
      msg = pninfo.fail_msg
10919
      if msg:
10920
        # Assume the primary node is unreachable and go ahead
10921
        self.warn.append("Can't get info from primary node %s: %s" %
10922
                         (pnode, msg))
10923
      elif not isinstance(pninfo.payload.get("memory_free", None), int):
10924
        self.warn.append("Node data from primary node %s doesn't contain"
10925
                         " free memory information" % pnode)
10926
      elif instance_info.fail_msg:
10927
        self.warn.append("Can't get instance runtime information: %s" %
10928
                        instance_info.fail_msg)
10929
      else:
10930
        if instance_info.payload:
10931
          current_mem = int(instance_info.payload["memory"])
10932
        else:
10933
          # Assume instance not running
10934
          # (there is a slight race condition here, but it's not very probable,
10935
          # and we have no other way to check)
10936
          current_mem = 0
10937
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
10938
                    pninfo.payload["memory_free"])
10939
        if miss_mem > 0:
10940
          raise errors.OpPrereqError("This change will prevent the instance"
10941
                                     " from starting, due to %d MB of memory"
10942
                                     " missing on its primary node" % miss_mem,
10943
                                     errors.ECODE_NORES)
10944

    
10945
      if be_new[constants.BE_AUTO_BALANCE]:
10946
        for node, nres in nodeinfo.items():
10947
          if node not in instance.secondary_nodes:
10948
            continue
10949
          nres.Raise("Can't get info from secondary node %s" % node,
10950
                     prereq=True, ecode=errors.ECODE_STATE)
10951
          if not isinstance(nres.payload.get("memory_free", None), int):
10952
            raise errors.OpPrereqError("Secondary node %s didn't return free"
10953
                                       " memory information" % node,
10954
                                       errors.ECODE_STATE)
10955
          elif be_new[constants.BE_MEMORY] > nres.payload["memory_free"]:
10956
            raise errors.OpPrereqError("This change will prevent the instance"
10957
                                       " from failover to its secondary node"
10958
                                       " %s, due to not enough memory" % node,
10959
                                       errors.ECODE_STATE)
10960

    
10961
    # NIC processing
10962
    self.nic_pnew = {}
10963
    self.nic_pinst = {}
10964
    for nic_op, nic_dict in self.op.nics:
10965
      if nic_op == constants.DDM_REMOVE:
10966
        if not instance.nics:
10967
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
10968
                                     errors.ECODE_INVAL)
10969
        continue
10970
      if nic_op != constants.DDM_ADD:
10971
        # an existing nic
10972
        if not instance.nics:
10973
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
10974
                                     " no NICs" % nic_op,
10975
                                     errors.ECODE_INVAL)
10976
        if nic_op < 0 or nic_op >= len(instance.nics):
10977
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
10978
                                     " are 0 to %d" %
10979
                                     (nic_op, len(instance.nics) - 1),
10980
                                     errors.ECODE_INVAL)
10981
        old_nic_params = instance.nics[nic_op].nicparams
10982
        old_nic_ip = instance.nics[nic_op].ip
10983
      else:
10984
        old_nic_params = {}
10985
        old_nic_ip = None
10986

    
10987
      update_params_dict = dict([(key, nic_dict[key])
10988
                                 for key in constants.NICS_PARAMETERS
10989
                                 if key in nic_dict])
10990

    
10991
      if "bridge" in nic_dict:
10992
        update_params_dict[constants.NIC_LINK] = nic_dict["bridge"]
10993

    
10994
      new_nic_params = _GetUpdatedParams(old_nic_params,
10995
                                         update_params_dict)
10996
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
10997
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
10998
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
10999
      self.nic_pinst[nic_op] = new_nic_params
11000
      self.nic_pnew[nic_op] = new_filled_nic_params
11001
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
11002

    
11003
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
11004
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
11005
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
11006
        if msg:
11007
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
11008
          if self.op.force:
11009
            self.warn.append(msg)
11010
          else:
11011
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
11012
      if new_nic_mode == constants.NIC_MODE_ROUTED:
11013
        if constants.INIC_IP in nic_dict:
11014
          nic_ip = nic_dict[constants.INIC_IP]
11015
        else:
11016
          nic_ip = old_nic_ip
11017
        if nic_ip is None:
11018
          raise errors.OpPrereqError("Cannot set the nic ip to None"
11019
                                     " on a routed nic", errors.ECODE_INVAL)
11020
      if constants.INIC_MAC in nic_dict:
11021
        nic_mac = nic_dict[constants.INIC_MAC]
11022
        if nic_mac is None:
11023
          raise errors.OpPrereqError("Cannot set the nic mac to None",
11024
                                     errors.ECODE_INVAL)
11025
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
11026
          # otherwise generate the mac
11027
          nic_dict[constants.INIC_MAC] = \
11028
            self.cfg.GenerateMAC(self.proc.GetECId())
11029
        else:
11030
          # or validate/reserve the current one
11031
          try:
11032
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
11033
          except errors.ReservationError:
11034
            raise errors.OpPrereqError("MAC address %s already in use"
11035
                                       " in cluster" % nic_mac,
11036
                                       errors.ECODE_NOTUNIQUE)
11037

    
11038
    # DISK processing
11039
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
11040
      raise errors.OpPrereqError("Disk operations not supported for"
11041
                                 " diskless instances",
11042
                                 errors.ECODE_INVAL)
11043
    for disk_op, _ in self.op.disks:
11044
      if disk_op == constants.DDM_REMOVE:
11045
        if len(instance.disks) == 1:
11046
          raise errors.OpPrereqError("Cannot remove the last disk of"
11047
                                     " an instance", errors.ECODE_INVAL)
11048
        _CheckInstanceDown(self, instance, "cannot remove disks")
11049

    
11050
      if (disk_op == constants.DDM_ADD and
11051
          len(instance.disks) >= constants.MAX_DISKS):
11052
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
11053
                                   " add more" % constants.MAX_DISKS,
11054
                                   errors.ECODE_STATE)
11055
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
11056
        # an existing disk
11057
        if disk_op < 0 or disk_op >= len(instance.disks):
11058
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
11059
                                     " are 0 to %d" %
11060
                                     (disk_op, len(instance.disks)),
11061
                                     errors.ECODE_INVAL)
11062

    
11063
    return
11064

    
11065
  def _ConvertPlainToDrbd(self, feedback_fn):
11066
    """Converts an instance from plain to drbd.
11067

11068
    """
11069
    feedback_fn("Converting template to drbd")
11070
    instance = self.instance
11071
    pnode = instance.primary_node
11072
    snode = self.op.remote_node
11073

    
11074
    # create a fake disk info for _GenerateDiskTemplate
11075
    disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
11076
                  constants.IDISK_VG: d.logical_id[0]}
11077
                 for d in instance.disks]
11078
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
11079
                                      instance.name, pnode, [snode],
11080
                                      disk_info, None, None, 0, feedback_fn)
11081
    info = _GetInstanceInfoText(instance)
11082
    feedback_fn("Creating aditional volumes...")
11083
    # first, create the missing data and meta devices
11084
    for disk in new_disks:
11085
      # unfortunately this is... not too nice
11086
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
11087
                            info, True)
11088
      for child in disk.children:
11089
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
11090
    # at this stage, all new LVs have been created, we can rename the
11091
    # old ones
11092
    feedback_fn("Renaming original volumes...")
11093
    rename_list = [(o, n.children[0].logical_id)
11094
                   for (o, n) in zip(instance.disks, new_disks)]
11095
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
11096
    result.Raise("Failed to rename original LVs")
11097

    
11098
    feedback_fn("Initializing DRBD devices...")
11099
    # all child devices are in place, we can now create the DRBD devices
11100
    for disk in new_disks:
11101
      for node in [pnode, snode]:
11102
        f_create = node == pnode
11103
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
11104

    
11105
    # at this point, the instance has been modified
11106
    instance.disk_template = constants.DT_DRBD8
11107
    instance.disks = new_disks
11108
    self.cfg.Update(instance, feedback_fn)
11109

    
11110
    # disks are created, waiting for sync
11111
    disk_abort = not _WaitForSync(self, instance,
11112
                                  oneshot=not self.op.wait_for_sync)
11113
    if disk_abort:
11114
      raise errors.OpExecError("There are some degraded disks for"
11115
                               " this instance, please cleanup manually")
11116

    
11117
  def _ConvertDrbdToPlain(self, feedback_fn):
11118
    """Converts an instance from drbd to plain.
11119

11120
    """
11121
    instance = self.instance
11122
    assert len(instance.secondary_nodes) == 1
11123
    pnode = instance.primary_node
11124
    snode = instance.secondary_nodes[0]
11125
    feedback_fn("Converting template to plain")
11126

    
11127
    old_disks = instance.disks
11128
    new_disks = [d.children[0] for d in old_disks]
11129

    
11130
    # copy over size and mode
11131
    for parent, child in zip(old_disks, new_disks):
11132
      child.size = parent.size
11133
      child.mode = parent.mode
11134

    
11135
    # this is a DRBD disk, return its port to the pool
11136
    # NOTE: this must be done right before the call to cfg.Update!
11137
    for disk in old_disks:
11138
      tcp_port = disk.logical_id[2]
11139
      self.cfg.AddTcpUdpPort(tcp_port)
11140

    
11141
    # update instance structure
11142
    instance.disks = new_disks
11143
    instance.disk_template = constants.DT_PLAIN
11144
    self.cfg.Update(instance, feedback_fn)
11145

    
11146
    feedback_fn("Removing volumes on the secondary node...")
11147
    for disk in old_disks:
11148
      self.cfg.SetDiskID(disk, snode)
11149
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
11150
      if msg:
11151
        self.LogWarning("Could not remove block device %s on node %s,"
11152
                        " continuing anyway: %s", disk.iv_name, snode, msg)
11153

    
11154
    feedback_fn("Removing unneeded volumes on the primary node...")
11155
    for idx, disk in enumerate(old_disks):
11156
      meta = disk.children[1]
11157
      self.cfg.SetDiskID(meta, pnode)
11158
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
11159
      if msg:
11160
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
11161
                        " continuing anyway: %s", idx, pnode, msg)
11162

    
11163
  def Exec(self, feedback_fn):
11164
    """Modifies an instance.
11165

11166
    All parameters take effect only at the next restart of the instance.
11167

11168
    """
11169
    # Process here the warnings from CheckPrereq, as we don't have a
11170
    # feedback_fn there.
11171
    for warn in self.warn:
11172
      feedback_fn("WARNING: %s" % warn)
11173

    
11174
    result = []
11175
    instance = self.instance
11176
    # disk changes
11177
    for disk_op, disk_dict in self.op.disks:
11178
      if disk_op == constants.DDM_REMOVE:
11179
        # remove the last disk
11180
        device = instance.disks.pop()
11181
        device_idx = len(instance.disks)
11182
        for node, disk in device.ComputeNodeTree(instance.primary_node):
11183
          self.cfg.SetDiskID(disk, node)
11184
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
11185
          if msg:
11186
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
11187
                            " continuing anyway", device_idx, node, msg)
11188
        result.append(("disk/%d" % device_idx, "remove"))
11189

    
11190
        # if this is a DRBD disk, return its port to the pool
11191
        if device.dev_type in constants.LDS_DRBD:
11192
          tcp_port = device.logical_id[2]
11193
          self.cfg.AddTcpUdpPort(tcp_port)
11194
      elif disk_op == constants.DDM_ADD:
11195
        # add a new disk
11196
        if instance.disk_template in (constants.DT_FILE,
11197
                                        constants.DT_SHARED_FILE):
11198
          file_driver, file_path = instance.disks[0].logical_id
11199
          file_path = os.path.dirname(file_path)
11200
        else:
11201
          file_driver = file_path = None
11202
        disk_idx_base = len(instance.disks)
11203
        new_disk = _GenerateDiskTemplate(self,
11204
                                         instance.disk_template,
11205
                                         instance.name, instance.primary_node,
11206
                                         instance.secondary_nodes,
11207
                                         [disk_dict],
11208
                                         file_path,
11209
                                         file_driver,
11210
                                         disk_idx_base, feedback_fn)[0]
11211
        instance.disks.append(new_disk)
11212
        info = _GetInstanceInfoText(instance)
11213

    
11214
        logging.info("Creating volume %s for instance %s",
11215
                     new_disk.iv_name, instance.name)
11216
        # Note: this needs to be kept in sync with _CreateDisks
11217
        #HARDCODE
11218
        for node in instance.all_nodes:
11219
          f_create = node == instance.primary_node
11220
          try:
11221
            _CreateBlockDev(self, node, instance, new_disk,
11222
                            f_create, info, f_create)
11223
          except errors.OpExecError, err:
11224
            self.LogWarning("Failed to create volume %s (%s) on"
11225
                            " node %s: %s",
11226
                            new_disk.iv_name, new_disk, node, err)
11227
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
11228
                       (new_disk.size, new_disk.mode)))
11229
      else:
11230
        # change a given disk
11231
        instance.disks[disk_op].mode = disk_dict[constants.IDISK_MODE]
11232
        result.append(("disk.mode/%d" % disk_op,
11233
                       disk_dict[constants.IDISK_MODE]))
11234

    
11235
    if self.op.disk_template:
11236
      r_shut = _ShutdownInstanceDisks(self, instance)
11237
      if not r_shut:
11238
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
11239
                                 " proceed with disk template conversion")
11240
      mode = (instance.disk_template, self.op.disk_template)
11241
      try:
11242
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
11243
      except:
11244
        self.cfg.ReleaseDRBDMinors(instance.name)
11245
        raise
11246
      result.append(("disk_template", self.op.disk_template))
11247

    
11248
    # NIC changes
11249
    for nic_op, nic_dict in self.op.nics:
11250
      if nic_op == constants.DDM_REMOVE:
11251
        # remove the last nic
11252
        del instance.nics[-1]
11253
        result.append(("nic.%d" % len(instance.nics), "remove"))
11254
      elif nic_op == constants.DDM_ADD:
11255
        # mac and bridge should be set, by now
11256
        mac = nic_dict[constants.INIC_MAC]
11257
        ip = nic_dict.get(constants.INIC_IP, None)
11258
        nicparams = self.nic_pinst[constants.DDM_ADD]
11259
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
11260
        instance.nics.append(new_nic)
11261
        result.append(("nic.%d" % (len(instance.nics) - 1),
11262
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
11263
                       (new_nic.mac, new_nic.ip,
11264
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
11265
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
11266
                       )))
11267
      else:
11268
        for key in (constants.INIC_MAC, constants.INIC_IP):
11269
          if key in nic_dict:
11270
            setattr(instance.nics[nic_op], key, nic_dict[key])
11271
        if nic_op in self.nic_pinst:
11272
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
11273
        for key, val in nic_dict.iteritems():
11274
          result.append(("nic.%s/%d" % (key, nic_op), val))
11275

    
11276
    # hvparams changes
11277
    if self.op.hvparams:
11278
      instance.hvparams = self.hv_inst
11279
      for key, val in self.op.hvparams.iteritems():
11280
        result.append(("hv/%s" % key, val))
11281

    
11282
    # beparams changes
11283
    if self.op.beparams:
11284
      instance.beparams = self.be_inst
11285
      for key, val in self.op.beparams.iteritems():
11286
        result.append(("be/%s" % key, val))
11287

    
11288
    # OS change
11289
    if self.op.os_name:
11290
      instance.os = self.op.os_name
11291

    
11292
    # osparams changes
11293
    if self.op.osparams:
11294
      instance.osparams = self.os_inst
11295
      for key, val in self.op.osparams.iteritems():
11296
        result.append(("os/%s" % key, val))
11297

    
11298
    self.cfg.Update(instance, feedback_fn)
11299

    
11300
    return result
11301

    
11302
  _DISK_CONVERSIONS = {
11303
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
11304
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
11305
    }
11306

    
11307

    
11308
class LUInstanceChangeGroup(LogicalUnit):
11309
  HPATH = "instance-change-group"
11310
  HTYPE = constants.HTYPE_INSTANCE
11311
  REQ_BGL = False
11312

    
11313
  def ExpandNames(self):
11314
    self.share_locks = _ShareAll()
11315
    self.needed_locks = {
11316
      locking.LEVEL_NODEGROUP: [],
11317
      locking.LEVEL_NODE: [],
11318
      }
11319

    
11320
    self._ExpandAndLockInstance()
11321

    
11322
    if self.op.target_groups:
11323
      self.req_target_uuids = map(self.cfg.LookupNodeGroup,
11324
                                  self.op.target_groups)
11325
    else:
11326
      self.req_target_uuids = None
11327

    
11328
    self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
11329

    
11330
  def DeclareLocks(self, level):
11331
    if level == locking.LEVEL_NODEGROUP:
11332
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
11333

    
11334
      if self.req_target_uuids:
11335
        lock_groups = set(self.req_target_uuids)
11336

    
11337
        # Lock all groups used by instance optimistically; this requires going
11338
        # via the node before it's locked, requiring verification later on
11339
        instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
11340
        lock_groups.update(instance_groups)
11341
      else:
11342
        # No target groups, need to lock all of them
11343
        lock_groups = locking.ALL_SET
11344

    
11345
      self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
11346

    
11347
    elif level == locking.LEVEL_NODE:
11348
      if self.req_target_uuids:
11349
        # Lock all nodes used by instances
11350
        self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
11351
        self._LockInstancesNodes()
11352

    
11353
        # Lock all nodes in all potential target groups
11354
        lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
11355
                       self.cfg.GetInstanceNodeGroups(self.op.instance_name))
11356
        member_nodes = [node_name
11357
                        for group in lock_groups
11358
                        for node_name in self.cfg.GetNodeGroup(group).members]
11359
        self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
11360
      else:
11361
        # Lock all nodes as all groups are potential targets
11362
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11363

    
11364
  def CheckPrereq(self):
11365
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
11366
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
11367
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
11368

    
11369
    assert (self.req_target_uuids is None or
11370
            owned_groups.issuperset(self.req_target_uuids))
11371
    assert owned_instances == set([self.op.instance_name])
11372

    
11373
    # Get instance information
11374
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
11375

    
11376
    # Check if node groups for locked instance are still correct
11377
    assert owned_nodes.issuperset(self.instance.all_nodes), \
11378
      ("Instance %s's nodes changed while we kept the lock" %
11379
       self.op.instance_name)
11380

    
11381
    inst_groups = _CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
11382
                                           owned_groups)
11383

    
11384
    if self.req_target_uuids:
11385
      # User requested specific target groups
11386
      self.target_uuids = frozenset(self.req_target_uuids)
11387
    else:
11388
      # All groups except those used by the instance are potential targets
11389
      self.target_uuids = owned_groups - inst_groups
11390

    
11391
    conflicting_groups = self.target_uuids & inst_groups
11392
    if conflicting_groups:
11393
      raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
11394
                                 " used by the instance '%s'" %
11395
                                 (utils.CommaJoin(conflicting_groups),
11396
                                  self.op.instance_name),
11397
                                 errors.ECODE_INVAL)
11398

    
11399
    if not self.target_uuids:
11400
      raise errors.OpPrereqError("There are no possible target groups",
11401
                                 errors.ECODE_INVAL)
11402

    
11403
  def BuildHooksEnv(self):
11404
    """Build hooks env.
11405

11406
    """
11407
    assert self.target_uuids
11408

    
11409
    env = {
11410
      "TARGET_GROUPS": " ".join(self.target_uuids),
11411
      }
11412

    
11413
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
11414

    
11415
    return env
11416

    
11417
  def BuildHooksNodes(self):
11418
    """Build hooks nodes.
11419

11420
    """
11421
    mn = self.cfg.GetMasterNode()
11422
    return ([mn], [mn])
11423

    
11424
  def Exec(self, feedback_fn):
11425
    instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
11426

    
11427
    assert instances == [self.op.instance_name], "Instance not locked"
11428

    
11429
    ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
11430
                     instances=instances, target_groups=list(self.target_uuids))
11431

    
11432
    ial.Run(self.op.iallocator)
11433

    
11434
    if not ial.success:
11435
      raise errors.OpPrereqError("Can't compute solution for changing group of"
11436
                                 " instance '%s' using iallocator '%s': %s" %
11437
                                 (self.op.instance_name, self.op.iallocator,
11438
                                  ial.info),
11439
                                 errors.ECODE_NORES)
11440

    
11441
    jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
11442

    
11443
    self.LogInfo("Iallocator returned %s job(s) for changing group of"
11444
                 " instance '%s'", len(jobs), self.op.instance_name)
11445

    
11446
    return ResultWithJobs(jobs)
11447

    
11448

    
11449
class LUBackupQuery(NoHooksLU):
11450
  """Query the exports list
11451

11452
  """
11453
  REQ_BGL = False
11454

    
11455
  def ExpandNames(self):
11456
    self.needed_locks = {}
11457
    self.share_locks[locking.LEVEL_NODE] = 1
11458
    if not self.op.nodes:
11459
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11460
    else:
11461
      self.needed_locks[locking.LEVEL_NODE] = \
11462
        _GetWantedNodes(self, self.op.nodes)
11463

    
11464
  def Exec(self, feedback_fn):
11465
    """Compute the list of all the exported system images.
11466

11467
    @rtype: dict
11468
    @return: a dictionary with the structure node->(export-list)
11469
        where export-list is a list of the instances exported on
11470
        that node.
11471

11472
    """
11473
    self.nodes = self.owned_locks(locking.LEVEL_NODE)
11474
    rpcresult = self.rpc.call_export_list(self.nodes)
11475
    result = {}
11476
    for node in rpcresult:
11477
      if rpcresult[node].fail_msg:
11478
        result[node] = False
11479
      else:
11480
        result[node] = rpcresult[node].payload
11481

    
11482
    return result
11483

    
11484

    
11485
class LUBackupPrepare(NoHooksLU):
11486
  """Prepares an instance for an export and returns useful information.
11487

11488
  """
11489
  REQ_BGL = False
11490

    
11491
  def ExpandNames(self):
11492
    self._ExpandAndLockInstance()
11493

    
11494
  def CheckPrereq(self):
11495
    """Check prerequisites.
11496

11497
    """
11498
    instance_name = self.op.instance_name
11499

    
11500
    self.instance = self.cfg.GetInstanceInfo(instance_name)
11501
    assert self.instance is not None, \
11502
          "Cannot retrieve locked instance %s" % self.op.instance_name
11503
    _CheckNodeOnline(self, self.instance.primary_node)
11504

    
11505
    self._cds = _GetClusterDomainSecret()
11506

    
11507
  def Exec(self, feedback_fn):
11508
    """Prepares an instance for an export.
11509

11510
    """
11511
    instance = self.instance
11512

    
11513
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
11514
      salt = utils.GenerateSecret(8)
11515

    
11516
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
11517
      result = self.rpc.call_x509_cert_create(instance.primary_node,
11518
                                              constants.RIE_CERT_VALIDITY)
11519
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
11520

    
11521
      (name, cert_pem) = result.payload
11522

    
11523
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
11524
                                             cert_pem)
11525

    
11526
      return {
11527
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
11528
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
11529
                          salt),
11530
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
11531
        }
11532

    
11533
    return None
11534

    
11535

    
11536
class LUBackupExport(LogicalUnit):
11537
  """Export an instance to an image in the cluster.
11538

11539
  """
11540
  HPATH = "instance-export"
11541
  HTYPE = constants.HTYPE_INSTANCE
11542
  REQ_BGL = False
11543

    
11544
  def CheckArguments(self):
11545
    """Check the arguments.
11546

11547
    """
11548
    self.x509_key_name = self.op.x509_key_name
11549
    self.dest_x509_ca_pem = self.op.destination_x509_ca
11550

    
11551
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
11552
      if not self.x509_key_name:
11553
        raise errors.OpPrereqError("Missing X509 key name for encryption",
11554
                                   errors.ECODE_INVAL)
11555

    
11556
      if not self.dest_x509_ca_pem:
11557
        raise errors.OpPrereqError("Missing destination X509 CA",
11558
                                   errors.ECODE_INVAL)
11559

    
11560
  def ExpandNames(self):
11561
    self._ExpandAndLockInstance()
11562

    
11563
    # Lock all nodes for local exports
11564
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11565
      # FIXME: lock only instance primary and destination node
11566
      #
11567
      # Sad but true, for now we have do lock all nodes, as we don't know where
11568
      # the previous export might be, and in this LU we search for it and
11569
      # remove it from its current node. In the future we could fix this by:
11570
      #  - making a tasklet to search (share-lock all), then create the
11571
      #    new one, then one to remove, after
11572
      #  - removing the removal operation altogether
11573
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11574

    
11575
  def DeclareLocks(self, level):
11576
    """Last minute lock declaration."""
11577
    # All nodes are locked anyway, so nothing to do here.
11578

    
11579
  def BuildHooksEnv(self):
11580
    """Build hooks env.
11581

11582
    This will run on the master, primary node and target node.
11583

11584
    """
11585
    env = {
11586
      "EXPORT_MODE": self.op.mode,
11587
      "EXPORT_NODE": self.op.target_node,
11588
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
11589
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
11590
      # TODO: Generic function for boolean env variables
11591
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
11592
      }
11593

    
11594
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
11595

    
11596
    return env
11597

    
11598
  def BuildHooksNodes(self):
11599
    """Build hooks nodes.
11600

11601
    """
11602
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
11603

    
11604
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11605
      nl.append(self.op.target_node)
11606

    
11607
    return (nl, nl)
11608

    
11609
  def CheckPrereq(self):
11610
    """Check prerequisites.
11611

11612
    This checks that the instance and node names are valid.
11613

11614
    """
11615
    instance_name = self.op.instance_name
11616

    
11617
    self.instance = self.cfg.GetInstanceInfo(instance_name)
11618
    assert self.instance is not None, \
11619
          "Cannot retrieve locked instance %s" % self.op.instance_name
11620
    _CheckNodeOnline(self, self.instance.primary_node)
11621

    
11622
    if (self.op.remove_instance and self.instance.admin_up and
11623
        not self.op.shutdown):
11624
      raise errors.OpPrereqError("Can not remove instance without shutting it"
11625
                                 " down before")
11626

    
11627
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11628
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
11629
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
11630
      assert self.dst_node is not None
11631

    
11632
      _CheckNodeOnline(self, self.dst_node.name)
11633
      _CheckNodeNotDrained(self, self.dst_node.name)
11634

    
11635
      self._cds = None
11636
      self.dest_disk_info = None
11637
      self.dest_x509_ca = None
11638

    
11639
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
11640
      self.dst_node = None
11641

    
11642
      if len(self.op.target_node) != len(self.instance.disks):
11643
        raise errors.OpPrereqError(("Received destination information for %s"
11644
                                    " disks, but instance %s has %s disks") %
11645
                                   (len(self.op.target_node), instance_name,
11646
                                    len(self.instance.disks)),
11647
                                   errors.ECODE_INVAL)
11648

    
11649
      cds = _GetClusterDomainSecret()
11650

    
11651
      # Check X509 key name
11652
      try:
11653
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
11654
      except (TypeError, ValueError), err:
11655
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
11656

    
11657
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
11658
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
11659
                                   errors.ECODE_INVAL)
11660

    
11661
      # Load and verify CA
11662
      try:
11663
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
11664
      except OpenSSL.crypto.Error, err:
11665
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
11666
                                   (err, ), errors.ECODE_INVAL)
11667

    
11668
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
11669
      if errcode is not None:
11670
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
11671
                                   (msg, ), errors.ECODE_INVAL)
11672

    
11673
      self.dest_x509_ca = cert
11674

    
11675
      # Verify target information
11676
      disk_info = []
11677
      for idx, disk_data in enumerate(self.op.target_node):
11678
        try:
11679
          (host, port, magic) = \
11680
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
11681
        except errors.GenericError, err:
11682
          raise errors.OpPrereqError("Target info for disk %s: %s" %
11683
                                     (idx, err), errors.ECODE_INVAL)
11684

    
11685
        disk_info.append((host, port, magic))
11686

    
11687
      assert len(disk_info) == len(self.op.target_node)
11688
      self.dest_disk_info = disk_info
11689

    
11690
    else:
11691
      raise errors.ProgrammerError("Unhandled export mode %r" %
11692
                                   self.op.mode)
11693

    
11694
    # instance disk type verification
11695
    # TODO: Implement export support for file-based disks
11696
    for disk in self.instance.disks:
11697
      if disk.dev_type == constants.LD_FILE:
11698
        raise errors.OpPrereqError("Export not supported for instances with"
11699
                                   " file-based disks", errors.ECODE_INVAL)
11700

    
11701
  def _CleanupExports(self, feedback_fn):
11702
    """Removes exports of current instance from all other nodes.
11703

11704
    If an instance in a cluster with nodes A..D was exported to node C, its
11705
    exports will be removed from the nodes A, B and D.
11706

11707
    """
11708
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
11709

    
11710
    nodelist = self.cfg.GetNodeList()
11711
    nodelist.remove(self.dst_node.name)
11712

    
11713
    # on one-node clusters nodelist will be empty after the removal
11714
    # if we proceed the backup would be removed because OpBackupQuery
11715
    # substitutes an empty list with the full cluster node list.
11716
    iname = self.instance.name
11717
    if nodelist:
11718
      feedback_fn("Removing old exports for instance %s" % iname)
11719
      exportlist = self.rpc.call_export_list(nodelist)
11720
      for node in exportlist:
11721
        if exportlist[node].fail_msg:
11722
          continue
11723
        if iname in exportlist[node].payload:
11724
          msg = self.rpc.call_export_remove(node, iname).fail_msg
11725
          if msg:
11726
            self.LogWarning("Could not remove older export for instance %s"
11727
                            " on node %s: %s", iname, node, msg)
11728

    
11729
  def Exec(self, feedback_fn):
11730
    """Export an instance to an image in the cluster.
11731

11732
    """
11733
    assert self.op.mode in constants.EXPORT_MODES
11734

    
11735
    instance = self.instance
11736
    src_node = instance.primary_node
11737

    
11738
    if self.op.shutdown:
11739
      # shutdown the instance, but not the disks
11740
      feedback_fn("Shutting down instance %s" % instance.name)
11741
      result = self.rpc.call_instance_shutdown(src_node, instance,
11742
                                               self.op.shutdown_timeout)
11743
      # TODO: Maybe ignore failures if ignore_remove_failures is set
11744
      result.Raise("Could not shutdown instance %s on"
11745
                   " node %s" % (instance.name, src_node))
11746

    
11747
    # set the disks ID correctly since call_instance_start needs the
11748
    # correct drbd minor to create the symlinks
11749
    for disk in instance.disks:
11750
      self.cfg.SetDiskID(disk, src_node)
11751

    
11752
    activate_disks = (not instance.admin_up)
11753

    
11754
    if activate_disks:
11755
      # Activate the instance disks if we'exporting a stopped instance
11756
      feedback_fn("Activating disks for %s" % instance.name)
11757
      _StartInstanceDisks(self, instance, None)
11758

    
11759
    try:
11760
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
11761
                                                     instance)
11762

    
11763
      helper.CreateSnapshots()
11764
      try:
11765
        if (self.op.shutdown and instance.admin_up and
11766
            not self.op.remove_instance):
11767
          assert not activate_disks
11768
          feedback_fn("Starting instance %s" % instance.name)
11769
          result = self.rpc.call_instance_start(src_node, instance,
11770
                                                None, None, False)
11771
          msg = result.fail_msg
11772
          if msg:
11773
            feedback_fn("Failed to start instance: %s" % msg)
11774
            _ShutdownInstanceDisks(self, instance)
11775
            raise errors.OpExecError("Could not start instance: %s" % msg)
11776

    
11777
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
11778
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
11779
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
11780
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
11781
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
11782

    
11783
          (key_name, _, _) = self.x509_key_name
11784

    
11785
          dest_ca_pem = \
11786
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
11787
                                            self.dest_x509_ca)
11788

    
11789
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
11790
                                                     key_name, dest_ca_pem,
11791
                                                     timeouts)
11792
      finally:
11793
        helper.Cleanup()
11794

    
11795
      # Check for backwards compatibility
11796
      assert len(dresults) == len(instance.disks)
11797
      assert compat.all(isinstance(i, bool) for i in dresults), \
11798
             "Not all results are boolean: %r" % dresults
11799

    
11800
    finally:
11801
      if activate_disks:
11802
        feedback_fn("Deactivating disks for %s" % instance.name)
11803
        _ShutdownInstanceDisks(self, instance)
11804

    
11805
    if not (compat.all(dresults) and fin_resu):
11806
      failures = []
11807
      if not fin_resu:
11808
        failures.append("export finalization")
11809
      if not compat.all(dresults):
11810
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
11811
                               if not dsk)
11812
        failures.append("disk export: disk(s) %s" % fdsk)
11813

    
11814
      raise errors.OpExecError("Export failed, errors in %s" %
11815
                               utils.CommaJoin(failures))
11816

    
11817
    # At this point, the export was successful, we can cleanup/finish
11818

    
11819
    # Remove instance if requested
11820
    if self.op.remove_instance:
11821
      feedback_fn("Removing instance %s" % instance.name)
11822
      _RemoveInstance(self, feedback_fn, instance,
11823
                      self.op.ignore_remove_failures)
11824

    
11825
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11826
      self._CleanupExports(feedback_fn)
11827

    
11828
    return fin_resu, dresults
11829

    
11830

    
11831
class LUBackupRemove(NoHooksLU):
11832
  """Remove exports related to the named instance.
11833

11834
  """
11835
  REQ_BGL = False
11836

    
11837
  def ExpandNames(self):
11838
    self.needed_locks = {}
11839
    # We need all nodes to be locked in order for RemoveExport to work, but we
11840
    # don't need to lock the instance itself, as nothing will happen to it (and
11841
    # we can remove exports also for a removed instance)
11842
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11843

    
11844
  def Exec(self, feedback_fn):
11845
    """Remove any export.
11846

11847
    """
11848
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
11849
    # If the instance was not found we'll try with the name that was passed in.
11850
    # This will only work if it was an FQDN, though.
11851
    fqdn_warn = False
11852
    if not instance_name:
11853
      fqdn_warn = True
11854
      instance_name = self.op.instance_name
11855

    
11856
    locked_nodes = self.owned_locks(locking.LEVEL_NODE)
11857
    exportlist = self.rpc.call_export_list(locked_nodes)
11858
    found = False
11859
    for node in exportlist:
11860
      msg = exportlist[node].fail_msg
11861
      if msg:
11862
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
11863
        continue
11864
      if instance_name in exportlist[node].payload:
11865
        found = True
11866
        result = self.rpc.call_export_remove(node, instance_name)
11867
        msg = result.fail_msg
11868
        if msg:
11869
          logging.error("Could not remove export for instance %s"
11870
                        " on node %s: %s", instance_name, node, msg)
11871

    
11872
    if fqdn_warn and not found:
11873
      feedback_fn("Export not found. If trying to remove an export belonging"
11874
                  " to a deleted instance please use its Fully Qualified"
11875
                  " Domain Name.")
11876

    
11877

    
11878
class LUGroupAdd(LogicalUnit):
11879
  """Logical unit for creating node groups.
11880

11881
  """
11882
  HPATH = "group-add"
11883
  HTYPE = constants.HTYPE_GROUP
11884
  REQ_BGL = False
11885

    
11886
  def ExpandNames(self):
11887
    # We need the new group's UUID here so that we can create and acquire the
11888
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
11889
    # that it should not check whether the UUID exists in the configuration.
11890
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
11891
    self.needed_locks = {}
11892
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
11893

    
11894
  def CheckPrereq(self):
11895
    """Check prerequisites.
11896

11897
    This checks that the given group name is not an existing node group
11898
    already.
11899

11900
    """
11901
    try:
11902
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11903
    except errors.OpPrereqError:
11904
      pass
11905
    else:
11906
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
11907
                                 " node group (UUID: %s)" %
11908
                                 (self.op.group_name, existing_uuid),
11909
                                 errors.ECODE_EXISTS)
11910

    
11911
    if self.op.ndparams:
11912
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
11913

    
11914
  def BuildHooksEnv(self):
11915
    """Build hooks env.
11916

11917
    """
11918
    return {
11919
      "GROUP_NAME": self.op.group_name,
11920
      }
11921

    
11922
  def BuildHooksNodes(self):
11923
    """Build hooks nodes.
11924

11925
    """
11926
    mn = self.cfg.GetMasterNode()
11927
    return ([mn], [mn])
11928

    
11929
  def Exec(self, feedback_fn):
11930
    """Add the node group to the cluster.
11931

11932
    """
11933
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
11934
                                  uuid=self.group_uuid,
11935
                                  alloc_policy=self.op.alloc_policy,
11936
                                  ndparams=self.op.ndparams)
11937

    
11938
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
11939
    del self.remove_locks[locking.LEVEL_NODEGROUP]
11940

    
11941

    
11942
class LUGroupAssignNodes(NoHooksLU):
11943
  """Logical unit for assigning nodes to groups.
11944

11945
  """
11946
  REQ_BGL = False
11947

    
11948
  def ExpandNames(self):
11949
    # These raise errors.OpPrereqError on their own:
11950
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11951
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
11952

    
11953
    # We want to lock all the affected nodes and groups. We have readily
11954
    # available the list of nodes, and the *destination* group. To gather the
11955
    # list of "source" groups, we need to fetch node information later on.
11956
    self.needed_locks = {
11957
      locking.LEVEL_NODEGROUP: set([self.group_uuid]),
11958
      locking.LEVEL_NODE: self.op.nodes,
11959
      }
11960

    
11961
  def DeclareLocks(self, level):
11962
    if level == locking.LEVEL_NODEGROUP:
11963
      assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
11964

    
11965
      # Try to get all affected nodes' groups without having the group or node
11966
      # lock yet. Needs verification later in the code flow.
11967
      groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
11968

    
11969
      self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
11970

    
11971
  def CheckPrereq(self):
11972
    """Check prerequisites.
11973

11974
    """
11975
    assert self.needed_locks[locking.LEVEL_NODEGROUP]
11976
    assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
11977
            frozenset(self.op.nodes))
11978

    
11979
    expected_locks = (set([self.group_uuid]) |
11980
                      self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
11981
    actual_locks = self.owned_locks(locking.LEVEL_NODEGROUP)
11982
    if actual_locks != expected_locks:
11983
      raise errors.OpExecError("Nodes changed groups since locks were acquired,"
11984
                               " current groups are '%s', used to be '%s'" %
11985
                               (utils.CommaJoin(expected_locks),
11986
                                utils.CommaJoin(actual_locks)))
11987

    
11988
    self.node_data = self.cfg.GetAllNodesInfo()
11989
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
11990
    instance_data = self.cfg.GetAllInstancesInfo()
11991

    
11992
    if self.group is None:
11993
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
11994
                               (self.op.group_name, self.group_uuid))
11995

    
11996
    (new_splits, previous_splits) = \
11997
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
11998
                                             for node in self.op.nodes],
11999
                                            self.node_data, instance_data)
12000

    
12001
    if new_splits:
12002
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
12003

    
12004
      if not self.op.force:
12005
        raise errors.OpExecError("The following instances get split by this"
12006
                                 " change and --force was not given: %s" %
12007
                                 fmt_new_splits)
12008
      else:
12009
        self.LogWarning("This operation will split the following instances: %s",
12010
                        fmt_new_splits)
12011

    
12012
        if previous_splits:
12013
          self.LogWarning("In addition, these already-split instances continue"
12014
                          " to be split across groups: %s",
12015
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
12016

    
12017
  def Exec(self, feedback_fn):
12018
    """Assign nodes to a new group.
12019

12020
    """
12021
    mods = [(node_name, self.group_uuid) for node_name in self.op.nodes]
12022

    
12023
    self.cfg.AssignGroupNodes(mods)
12024

    
12025
  @staticmethod
12026
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
12027
    """Check for split instances after a node assignment.
12028

12029
    This method considers a series of node assignments as an atomic operation,
12030
    and returns information about split instances after applying the set of
12031
    changes.
12032

12033
    In particular, it returns information about newly split instances, and
12034
    instances that were already split, and remain so after the change.
12035

12036
    Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
12037
    considered.
12038

12039
    @type changes: list of (node_name, new_group_uuid) pairs.
12040
    @param changes: list of node assignments to consider.
12041
    @param node_data: a dict with data for all nodes
12042
    @param instance_data: a dict with all instances to consider
12043
    @rtype: a two-tuple
12044
    @return: a list of instances that were previously okay and result split as a
12045
      consequence of this change, and a list of instances that were previously
12046
      split and this change does not fix.
12047

12048
    """
12049
    changed_nodes = dict((node, group) for node, group in changes
12050
                         if node_data[node].group != group)
12051

    
12052
    all_split_instances = set()
12053
    previously_split_instances = set()
12054

    
12055
    def InstanceNodes(instance):
12056
      return [instance.primary_node] + list(instance.secondary_nodes)
12057

    
12058
    for inst in instance_data.values():
12059
      if inst.disk_template not in constants.DTS_INT_MIRROR:
12060
        continue
12061

    
12062
      instance_nodes = InstanceNodes(inst)
12063

    
12064
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
12065
        previously_split_instances.add(inst.name)
12066

    
12067
      if len(set(changed_nodes.get(node, node_data[node].group)
12068
                 for node in instance_nodes)) > 1:
12069
        all_split_instances.add(inst.name)
12070

    
12071
    return (list(all_split_instances - previously_split_instances),
12072
            list(previously_split_instances & all_split_instances))
12073

    
12074

    
12075
class _GroupQuery(_QueryBase):
12076
  FIELDS = query.GROUP_FIELDS
12077

    
12078
  def ExpandNames(self, lu):
12079
    lu.needed_locks = {}
12080

    
12081
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
12082
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
12083

    
12084
    if not self.names:
12085
      self.wanted = [name_to_uuid[name]
12086
                     for name in utils.NiceSort(name_to_uuid.keys())]
12087
    else:
12088
      # Accept names to be either names or UUIDs.
12089
      missing = []
12090
      self.wanted = []
12091
      all_uuid = frozenset(self._all_groups.keys())
12092

    
12093
      for name in self.names:
12094
        if name in all_uuid:
12095
          self.wanted.append(name)
12096
        elif name in name_to_uuid:
12097
          self.wanted.append(name_to_uuid[name])
12098
        else:
12099
          missing.append(name)
12100

    
12101
      if missing:
12102
        raise errors.OpPrereqError("Some groups do not exist: %s" %
12103
                                   utils.CommaJoin(missing),
12104
                                   errors.ECODE_NOENT)
12105

    
12106
  def DeclareLocks(self, lu, level):
12107
    pass
12108

    
12109
  def _GetQueryData(self, lu):
12110
    """Computes the list of node groups and their attributes.
12111

12112
    """
12113
    do_nodes = query.GQ_NODE in self.requested_data
12114
    do_instances = query.GQ_INST in self.requested_data
12115

    
12116
    group_to_nodes = None
12117
    group_to_instances = None
12118

    
12119
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
12120
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
12121
    # latter GetAllInstancesInfo() is not enough, for we have to go through
12122
    # instance->node. Hence, we will need to process nodes even if we only need
12123
    # instance information.
12124
    if do_nodes or do_instances:
12125
      all_nodes = lu.cfg.GetAllNodesInfo()
12126
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
12127
      node_to_group = {}
12128

    
12129
      for node in all_nodes.values():
12130
        if node.group in group_to_nodes:
12131
          group_to_nodes[node.group].append(node.name)
12132
          node_to_group[node.name] = node.group
12133

    
12134
      if do_instances:
12135
        all_instances = lu.cfg.GetAllInstancesInfo()
12136
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
12137

    
12138
        for instance in all_instances.values():
12139
          node = instance.primary_node
12140
          if node in node_to_group:
12141
            group_to_instances[node_to_group[node]].append(instance.name)
12142

    
12143
        if not do_nodes:
12144
          # Do not pass on node information if it was not requested.
12145
          group_to_nodes = None
12146

    
12147
    return query.GroupQueryData([self._all_groups[uuid]
12148
                                 for uuid in self.wanted],
12149
                                group_to_nodes, group_to_instances)
12150

    
12151

    
12152
class LUGroupQuery(NoHooksLU):
12153
  """Logical unit for querying node groups.
12154

12155
  """
12156
  REQ_BGL = False
12157

    
12158
  def CheckArguments(self):
12159
    self.gq = _GroupQuery(qlang.MakeSimpleFilter("name", self.op.names),
12160
                          self.op.output_fields, False)
12161

    
12162
  def ExpandNames(self):
12163
    self.gq.ExpandNames(self)
12164

    
12165
  def DeclareLocks(self, level):
12166
    self.gq.DeclareLocks(self, level)
12167

    
12168
  def Exec(self, feedback_fn):
12169
    return self.gq.OldStyleQuery(self)
12170

    
12171

    
12172
class LUGroupSetParams(LogicalUnit):
12173
  """Modifies the parameters of a node group.
12174

12175
  """
12176
  HPATH = "group-modify"
12177
  HTYPE = constants.HTYPE_GROUP
12178
  REQ_BGL = False
12179

    
12180
  def CheckArguments(self):
12181
    all_changes = [
12182
      self.op.ndparams,
12183
      self.op.alloc_policy,
12184
      ]
12185

    
12186
    if all_changes.count(None) == len(all_changes):
12187
      raise errors.OpPrereqError("Please pass at least one modification",
12188
                                 errors.ECODE_INVAL)
12189

    
12190
  def ExpandNames(self):
12191
    # This raises errors.OpPrereqError on its own:
12192
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12193

    
12194
    self.needed_locks = {
12195
      locking.LEVEL_NODEGROUP: [self.group_uuid],
12196
      }
12197

    
12198
  def CheckPrereq(self):
12199
    """Check prerequisites.
12200

12201
    """
12202
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
12203

    
12204
    if self.group is None:
12205
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
12206
                               (self.op.group_name, self.group_uuid))
12207

    
12208
    if self.op.ndparams:
12209
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
12210
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
12211
      self.new_ndparams = new_ndparams
12212

    
12213
  def BuildHooksEnv(self):
12214
    """Build hooks env.
12215

12216
    """
12217
    return {
12218
      "GROUP_NAME": self.op.group_name,
12219
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
12220
      }
12221

    
12222
  def BuildHooksNodes(self):
12223
    """Build hooks nodes.
12224

12225
    """
12226
    mn = self.cfg.GetMasterNode()
12227
    return ([mn], [mn])
12228

    
12229
  def Exec(self, feedback_fn):
12230
    """Modifies the node group.
12231

12232
    """
12233
    result = []
12234

    
12235
    if self.op.ndparams:
12236
      self.group.ndparams = self.new_ndparams
12237
      result.append(("ndparams", str(self.group.ndparams)))
12238

    
12239
    if self.op.alloc_policy:
12240
      self.group.alloc_policy = self.op.alloc_policy
12241

    
12242
    self.cfg.Update(self.group, feedback_fn)
12243
    return result
12244

    
12245

    
12246
class LUGroupRemove(LogicalUnit):
12247
  HPATH = "group-remove"
12248
  HTYPE = constants.HTYPE_GROUP
12249
  REQ_BGL = False
12250

    
12251
  def ExpandNames(self):
12252
    # This will raises errors.OpPrereqError on its own:
12253
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12254
    self.needed_locks = {
12255
      locking.LEVEL_NODEGROUP: [self.group_uuid],
12256
      }
12257

    
12258
  def CheckPrereq(self):
12259
    """Check prerequisites.
12260

12261
    This checks that the given group name exists as a node group, that is
12262
    empty (i.e., contains no nodes), and that is not the last group of the
12263
    cluster.
12264

12265
    """
12266
    # Verify that the group is empty.
12267
    group_nodes = [node.name
12268
                   for node in self.cfg.GetAllNodesInfo().values()
12269
                   if node.group == self.group_uuid]
12270

    
12271
    if group_nodes:
12272
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
12273
                                 " nodes: %s" %
12274
                                 (self.op.group_name,
12275
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
12276
                                 errors.ECODE_STATE)
12277

    
12278
    # Verify the cluster would not be left group-less.
12279
    if len(self.cfg.GetNodeGroupList()) == 1:
12280
      raise errors.OpPrereqError("Group '%s' is the only group,"
12281
                                 " cannot be removed" %
12282
                                 self.op.group_name,
12283
                                 errors.ECODE_STATE)
12284

    
12285
  def BuildHooksEnv(self):
12286
    """Build hooks env.
12287

12288
    """
12289
    return {
12290
      "GROUP_NAME": self.op.group_name,
12291
      }
12292

    
12293
  def BuildHooksNodes(self):
12294
    """Build hooks nodes.
12295

12296
    """
12297
    mn = self.cfg.GetMasterNode()
12298
    return ([mn], [mn])
12299

    
12300
  def Exec(self, feedback_fn):
12301
    """Remove the node group.
12302

12303
    """
12304
    try:
12305
      self.cfg.RemoveNodeGroup(self.group_uuid)
12306
    except errors.ConfigurationError:
12307
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
12308
                               (self.op.group_name, self.group_uuid))
12309

    
12310
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
12311

    
12312

    
12313
class LUGroupRename(LogicalUnit):
12314
  HPATH = "group-rename"
12315
  HTYPE = constants.HTYPE_GROUP
12316
  REQ_BGL = False
12317

    
12318
  def ExpandNames(self):
12319
    # This raises errors.OpPrereqError on its own:
12320
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12321

    
12322
    self.needed_locks = {
12323
      locking.LEVEL_NODEGROUP: [self.group_uuid],
12324
      }
12325

    
12326
  def CheckPrereq(self):
12327
    """Check prerequisites.
12328

12329
    Ensures requested new name is not yet used.
12330

12331
    """
12332
    try:
12333
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
12334
    except errors.OpPrereqError:
12335
      pass
12336
    else:
12337
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
12338
                                 " node group (UUID: %s)" %
12339
                                 (self.op.new_name, new_name_uuid),
12340
                                 errors.ECODE_EXISTS)
12341

    
12342
  def BuildHooksEnv(self):
12343
    """Build hooks env.
12344

12345
    """
12346
    return {
12347
      "OLD_NAME": self.op.group_name,
12348
      "NEW_NAME": self.op.new_name,
12349
      }
12350

    
12351
  def BuildHooksNodes(self):
12352
    """Build hooks nodes.
12353

12354
    """
12355
    mn = self.cfg.GetMasterNode()
12356

    
12357
    all_nodes = self.cfg.GetAllNodesInfo()
12358
    all_nodes.pop(mn, None)
12359

    
12360
    run_nodes = [mn]
12361
    run_nodes.extend(node.name for node in all_nodes.values()
12362
                     if node.group == self.group_uuid)
12363

    
12364
    return (run_nodes, run_nodes)
12365

    
12366
  def Exec(self, feedback_fn):
12367
    """Rename the node group.
12368

12369
    """
12370
    group = self.cfg.GetNodeGroup(self.group_uuid)
12371

    
12372
    if group is None:
12373
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
12374
                               (self.op.group_name, self.group_uuid))
12375

    
12376
    group.name = self.op.new_name
12377
    self.cfg.Update(group, feedback_fn)
12378

    
12379
    return self.op.new_name
12380

    
12381

    
12382
class LUGroupEvacuate(LogicalUnit):
12383
  HPATH = "group-evacuate"
12384
  HTYPE = constants.HTYPE_GROUP
12385
  REQ_BGL = False
12386

    
12387
  def ExpandNames(self):
12388
    # This raises errors.OpPrereqError on its own:
12389
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
12390

    
12391
    if self.op.target_groups:
12392
      self.req_target_uuids = map(self.cfg.LookupNodeGroup,
12393
                                  self.op.target_groups)
12394
    else:
12395
      self.req_target_uuids = []
12396

    
12397
    if self.group_uuid in self.req_target_uuids:
12398
      raise errors.OpPrereqError("Group to be evacuated (%s) can not be used"
12399
                                 " as a target group (targets are %s)" %
12400
                                 (self.group_uuid,
12401
                                  utils.CommaJoin(self.req_target_uuids)),
12402
                                 errors.ECODE_INVAL)
12403

    
12404
    self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
12405

    
12406
    self.share_locks = _ShareAll()
12407
    self.needed_locks = {
12408
      locking.LEVEL_INSTANCE: [],
12409
      locking.LEVEL_NODEGROUP: [],
12410
      locking.LEVEL_NODE: [],
12411
      }
12412

    
12413
  def DeclareLocks(self, level):
12414
    if level == locking.LEVEL_INSTANCE:
12415
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
12416

    
12417
      # Lock instances optimistically, needs verification once node and group
12418
      # locks have been acquired
12419
      self.needed_locks[locking.LEVEL_INSTANCE] = \
12420
        self.cfg.GetNodeGroupInstances(self.group_uuid)
12421

    
12422
    elif level == locking.LEVEL_NODEGROUP:
12423
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
12424

    
12425
      if self.req_target_uuids:
12426
        lock_groups = set([self.group_uuid] + self.req_target_uuids)
12427

    
12428
        # Lock all groups used by instances optimistically; this requires going
12429
        # via the node before it's locked, requiring verification later on
12430
        lock_groups.update(group_uuid
12431
                           for instance_name in
12432
                             self.owned_locks(locking.LEVEL_INSTANCE)
12433
                           for group_uuid in
12434
                             self.cfg.GetInstanceNodeGroups(instance_name))
12435
      else:
12436
        # No target groups, need to lock all of them
12437
        lock_groups = locking.ALL_SET
12438

    
12439
      self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
12440

    
12441
    elif level == locking.LEVEL_NODE:
12442
      # This will only lock the nodes in the group to be evacuated which
12443
      # contain actual instances
12444
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
12445
      self._LockInstancesNodes()
12446

    
12447
      # Lock all nodes in group to be evacuated and target groups
12448
      owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
12449
      assert self.group_uuid in owned_groups
12450
      member_nodes = [node_name
12451
                      for group in owned_groups
12452
                      for node_name in self.cfg.GetNodeGroup(group).members]
12453
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
12454

    
12455
  def CheckPrereq(self):
12456
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
12457
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
12458
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
12459

    
12460
    assert owned_groups.issuperset(self.req_target_uuids)
12461
    assert self.group_uuid in owned_groups
12462

    
12463
    # Check if locked instances are still correct
12464
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
12465

    
12466
    # Get instance information
12467
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
12468

    
12469
    # Check if node groups for locked instances are still correct
12470
    for instance_name in owned_instances:
12471
      inst = self.instances[instance_name]
12472
      assert owned_nodes.issuperset(inst.all_nodes), \
12473
        "Instance %s's nodes changed while we kept the lock" % instance_name
12474

    
12475
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
12476
                                             owned_groups)
12477

    
12478
      assert self.group_uuid in inst_groups, \
12479
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
12480

    
12481
    if self.req_target_uuids:
12482
      # User requested specific target groups
12483
      self.target_uuids = self.req_target_uuids
12484
    else:
12485
      # All groups except the one to be evacuated are potential targets
12486
      self.target_uuids = [group_uuid for group_uuid in owned_groups
12487
                           if group_uuid != self.group_uuid]
12488

    
12489
      if not self.target_uuids:
12490
        raise errors.OpPrereqError("There are no possible target groups",
12491
                                   errors.ECODE_INVAL)
12492

    
12493
  def BuildHooksEnv(self):
12494
    """Build hooks env.
12495

12496
    """
12497
    return {
12498
      "GROUP_NAME": self.op.group_name,
12499
      "TARGET_GROUPS": " ".join(self.target_uuids),
12500
      }
12501

    
12502
  def BuildHooksNodes(self):
12503
    """Build hooks nodes.
12504

12505
    """
12506
    mn = self.cfg.GetMasterNode()
12507

    
12508
    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
12509

    
12510
    run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members
12511

    
12512
    return (run_nodes, run_nodes)
12513

    
12514
  def Exec(self, feedback_fn):
12515
    instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
12516

    
12517
    assert self.group_uuid not in self.target_uuids
12518

    
12519
    ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
12520
                     instances=instances, target_groups=self.target_uuids)
12521

    
12522
    ial.Run(self.op.iallocator)
12523

    
12524
    if not ial.success:
12525
      raise errors.OpPrereqError("Can't compute group evacuation using"
12526
                                 " iallocator '%s': %s" %
12527
                                 (self.op.iallocator, ial.info),
12528
                                 errors.ECODE_NORES)
12529

    
12530
    jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
12531

    
12532
    self.LogInfo("Iallocator returned %s job(s) for evacuating node group %s",
12533
                 len(jobs), self.op.group_name)
12534

    
12535
    return ResultWithJobs(jobs)
12536

    
12537

    
12538
class TagsLU(NoHooksLU): # pylint: disable=W0223
12539
  """Generic tags LU.
12540

12541
  This is an abstract class which is the parent of all the other tags LUs.
12542

12543
  """
12544
  def ExpandNames(self):
12545
    self.group_uuid = None
12546
    self.needed_locks = {}
12547
    if self.op.kind == constants.TAG_NODE:
12548
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
12549
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
12550
    elif self.op.kind == constants.TAG_INSTANCE:
12551
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
12552
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
12553
    elif self.op.kind == constants.TAG_NODEGROUP:
12554
      self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)
12555

    
12556
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
12557
    # not possible to acquire the BGL based on opcode parameters)
12558

    
12559
  def CheckPrereq(self):
12560
    """Check prerequisites.
12561

12562
    """
12563
    if self.op.kind == constants.TAG_CLUSTER:
12564
      self.target = self.cfg.GetClusterInfo()
12565
    elif self.op.kind == constants.TAG_NODE:
12566
      self.target = self.cfg.GetNodeInfo(self.op.name)
12567
    elif self.op.kind == constants.TAG_INSTANCE:
12568
      self.target = self.cfg.GetInstanceInfo(self.op.name)
12569
    elif self.op.kind == constants.TAG_NODEGROUP:
12570
      self.target = self.cfg.GetNodeGroup(self.group_uuid)
12571
    else:
12572
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
12573
                                 str(self.op.kind), errors.ECODE_INVAL)
12574

    
12575

    
12576
class LUTagsGet(TagsLU):
12577
  """Returns the tags of a given object.
12578

12579
  """
12580
  REQ_BGL = False
12581

    
12582
  def ExpandNames(self):
12583
    TagsLU.ExpandNames(self)
12584

    
12585
    # Share locks as this is only a read operation
12586
    self.share_locks = _ShareAll()
12587

    
12588
  def Exec(self, feedback_fn):
12589
    """Returns the tag list.
12590

12591
    """
12592
    return list(self.target.GetTags())
12593

    
12594

    
12595
class LUTagsSearch(NoHooksLU):
12596
  """Searches the tags for a given pattern.
12597

12598
  """
12599
  REQ_BGL = False
12600

    
12601
  def ExpandNames(self):
12602
    self.needed_locks = {}
12603

    
12604
  def CheckPrereq(self):
12605
    """Check prerequisites.
12606

12607
    This checks the pattern passed for validity by compiling it.
12608

12609
    """
12610
    try:
12611
      self.re = re.compile(self.op.pattern)
12612
    except re.error, err:
12613
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
12614
                                 (self.op.pattern, err), errors.ECODE_INVAL)
12615

    
12616
  def Exec(self, feedback_fn):
12617
    """Returns the tag list.
12618

12619
    """
12620
    cfg = self.cfg
12621
    tgts = [("/cluster", cfg.GetClusterInfo())]
12622
    ilist = cfg.GetAllInstancesInfo().values()
12623
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
12624
    nlist = cfg.GetAllNodesInfo().values()
12625
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
12626
    tgts.extend(("/nodegroup/%s" % n.name, n)
12627
                for n in cfg.GetAllNodeGroupsInfo().values())
12628
    results = []
12629
    for path, target in tgts:
12630
      for tag in target.GetTags():
12631
        if self.re.search(tag):
12632
          results.append((path, tag))
12633
    return results
12634

    
12635

    
12636
class LUTagsSet(TagsLU):
12637
  """Sets a tag on a given object.
12638

12639
  """
12640
  REQ_BGL = False
12641

    
12642
  def CheckPrereq(self):
12643
    """Check prerequisites.
12644

12645
    This checks the type and length of the tag name and value.
12646

12647
    """
12648
    TagsLU.CheckPrereq(self)
12649
    for tag in self.op.tags:
12650
      objects.TaggableObject.ValidateTag(tag)
12651

    
12652
  def Exec(self, feedback_fn):
12653
    """Sets the tag.
12654

12655
    """
12656
    try:
12657
      for tag in self.op.tags:
12658
        self.target.AddTag(tag)
12659
    except errors.TagError, err:
12660
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
12661
    self.cfg.Update(self.target, feedback_fn)
12662

    
12663

    
12664
class LUTagsDel(TagsLU):
12665
  """Delete a list of tags from a given object.
12666

12667
  """
12668
  REQ_BGL = False
12669

    
12670
  def CheckPrereq(self):
12671
    """Check prerequisites.
12672

12673
    This checks that we have the given tag.
12674

12675
    """
12676
    TagsLU.CheckPrereq(self)
12677
    for tag in self.op.tags:
12678
      objects.TaggableObject.ValidateTag(tag)
12679
    del_tags = frozenset(self.op.tags)
12680
    cur_tags = self.target.GetTags()
12681

    
12682
    diff_tags = del_tags - cur_tags
12683
    if diff_tags:
12684
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
12685
      raise errors.OpPrereqError("Tag(s) %s not found" %
12686
                                 (utils.CommaJoin(diff_names), ),
12687
                                 errors.ECODE_NOENT)
12688

    
12689
  def Exec(self, feedback_fn):
12690
    """Remove the tag from the object.
12691

12692
    """
12693
    for tag in self.op.tags:
12694
      self.target.RemoveTag(tag)
12695
    self.cfg.Update(self.target, feedback_fn)
12696

    
12697

    
12698
class LUTestDelay(NoHooksLU):
12699
  """Sleep for a specified amount of time.
12700

12701
  This LU sleeps on the master and/or nodes for a specified amount of
12702
  time.
12703

12704
  """
12705
  REQ_BGL = False
12706

    
12707
  def ExpandNames(self):
12708
    """Expand names and set required locks.
12709

12710
    This expands the node list, if any.
12711

12712
    """
12713
    self.needed_locks = {}
12714
    if self.op.on_nodes:
12715
      # _GetWantedNodes can be used here, but is not always appropriate to use
12716
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
12717
      # more information.
12718
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
12719
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
12720

    
12721
  def _TestDelay(self):
12722
    """Do the actual sleep.
12723

12724
    """
12725
    if self.op.on_master:
12726
      if not utils.TestDelay(self.op.duration):
12727
        raise errors.OpExecError("Error during master delay test")
12728
    if self.op.on_nodes:
12729
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
12730
      for node, node_result in result.items():
12731
        node_result.Raise("Failure during rpc call to node %s" % node)
12732

    
12733
  def Exec(self, feedback_fn):
12734
    """Execute the test delay opcode, with the wanted repetitions.
12735

12736
    """
12737
    if self.op.repeat == 0:
12738
      self._TestDelay()
12739
    else:
12740
      top_value = self.op.repeat - 1
12741
      for i in range(self.op.repeat):
12742
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
12743
        self._TestDelay()
12744

    
12745

    
12746
class LUTestJqueue(NoHooksLU):
12747
  """Utility LU to test some aspects of the job queue.
12748

12749
  """
12750
  REQ_BGL = False
12751

    
12752
  # Must be lower than default timeout for WaitForJobChange to see whether it
12753
  # notices changed jobs
12754
  _CLIENT_CONNECT_TIMEOUT = 20.0
12755
  _CLIENT_CONFIRM_TIMEOUT = 60.0
12756

    
12757
  @classmethod
12758
  def _NotifyUsingSocket(cls, cb, errcls):
12759
    """Opens a Unix socket and waits for another program to connect.
12760

12761
    @type cb: callable
12762
    @param cb: Callback to send socket name to client
12763
    @type errcls: class
12764
    @param errcls: Exception class to use for errors
12765

12766
    """
12767
    # Using a temporary directory as there's no easy way to create temporary
12768
    # sockets without writing a custom loop around tempfile.mktemp and
12769
    # socket.bind
12770
    tmpdir = tempfile.mkdtemp()
12771
    try:
12772
      tmpsock = utils.PathJoin(tmpdir, "sock")
12773

    
12774
      logging.debug("Creating temporary socket at %s", tmpsock)
12775
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
12776
      try:
12777
        sock.bind(tmpsock)
12778
        sock.listen(1)
12779

    
12780
        # Send details to client
12781
        cb(tmpsock)
12782

    
12783
        # Wait for client to connect before continuing
12784
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
12785
        try:
12786
          (conn, _) = sock.accept()
12787
        except socket.error, err:
12788
          raise errcls("Client didn't connect in time (%s)" % err)
12789
      finally:
12790
        sock.close()
12791
    finally:
12792
      # Remove as soon as client is connected
12793
      shutil.rmtree(tmpdir)
12794

    
12795
    # Wait for client to close
12796
    try:
12797
      try:
12798
        # pylint: disable=E1101
12799
        # Instance of '_socketobject' has no ... member
12800
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
12801
        conn.recv(1)
12802
      except socket.error, err:
12803
        raise errcls("Client failed to confirm notification (%s)" % err)
12804
    finally:
12805
      conn.close()
12806

    
12807
  def _SendNotification(self, test, arg, sockname):
12808
    """Sends a notification to the client.
12809

12810
    @type test: string
12811
    @param test: Test name
12812
    @param arg: Test argument (depends on test)
12813
    @type sockname: string
12814
    @param sockname: Socket path
12815

12816
    """
12817
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
12818

    
12819
  def _Notify(self, prereq, test, arg):
12820
    """Notifies the client of a test.
12821

12822
    @type prereq: bool
12823
    @param prereq: Whether this is a prereq-phase test
12824
    @type test: string
12825
    @param test: Test name
12826
    @param arg: Test argument (depends on test)
12827

12828
    """
12829
    if prereq:
12830
      errcls = errors.OpPrereqError
12831
    else:
12832
      errcls = errors.OpExecError
12833

    
12834
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
12835
                                                  test, arg),
12836
                                   errcls)
12837

    
12838
  def CheckArguments(self):
12839
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
12840
    self.expandnames_calls = 0
12841

    
12842
  def ExpandNames(self):
12843
    checkargs_calls = getattr(self, "checkargs_calls", 0)
12844
    if checkargs_calls < 1:
12845
      raise errors.ProgrammerError("CheckArguments was not called")
12846

    
12847
    self.expandnames_calls += 1
12848

    
12849
    if self.op.notify_waitlock:
12850
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
12851

    
12852
    self.LogInfo("Expanding names")
12853

    
12854
    # Get lock on master node (just to get a lock, not for a particular reason)
12855
    self.needed_locks = {
12856
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
12857
      }
12858

    
12859
  def Exec(self, feedback_fn):
12860
    if self.expandnames_calls < 1:
12861
      raise errors.ProgrammerError("ExpandNames was not called")
12862

    
12863
    if self.op.notify_exec:
12864
      self._Notify(False, constants.JQT_EXEC, None)
12865

    
12866
    self.LogInfo("Executing")
12867

    
12868
    if self.op.log_messages:
12869
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
12870
      for idx, msg in enumerate(self.op.log_messages):
12871
        self.LogInfo("Sending log message %s", idx + 1)
12872
        feedback_fn(constants.JQT_MSGPREFIX + msg)
12873
        # Report how many test messages have been sent
12874
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
12875

    
12876
    if self.op.fail:
12877
      raise errors.OpExecError("Opcode failure was requested")
12878

    
12879
    return True
12880

    
12881

    
12882
class IAllocator(object):
12883
  """IAllocator framework.
12884

12885
  An IAllocator instance has three sets of attributes:
12886
    - cfg that is needed to query the cluster
12887
    - input data (all members of the _KEYS class attribute are required)
12888
    - four buffer attributes (in|out_data|text), that represent the
12889
      input (to the external script) in text and data structure format,
12890
      and the output from it, again in two formats
12891
    - the result variables from the script (success, info, nodes) for
12892
      easy usage
12893

12894
  """
12895
  # pylint: disable=R0902
12896
  # lots of instance attributes
12897

    
12898
  def __init__(self, cfg, rpc, mode, **kwargs):
12899
    self.cfg = cfg
12900
    self.rpc = rpc
12901
    # init buffer variables
12902
    self.in_text = self.out_text = self.in_data = self.out_data = None
12903
    # init all input fields so that pylint is happy
12904
    self.mode = mode
12905
    self.memory = self.disks = self.disk_template = None
12906
    self.os = self.tags = self.nics = self.vcpus = None
12907
    self.hypervisor = None
12908
    self.relocate_from = None
12909
    self.name = None
12910
    self.instances = None
12911
    self.evac_mode = None
12912
    self.target_groups = []
12913
    # computed fields
12914
    self.required_nodes = None
12915
    # init result fields
12916
    self.success = self.info = self.result = None
12917

    
12918
    try:
12919
      (fn, keydata, self._result_check) = self._MODE_DATA[self.mode]
12920
    except KeyError:
12921
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
12922
                                   " IAllocator" % self.mode)
12923

    
12924
    keyset = [n for (n, _) in keydata]
12925

    
12926
    for key in kwargs:
12927
      if key not in keyset:
12928
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
12929
                                     " IAllocator" % key)
12930
      setattr(self, key, kwargs[key])
12931

    
12932
    for key in keyset:
12933
      if key not in kwargs:
12934
        raise errors.ProgrammerError("Missing input parameter '%s' to"
12935
                                     " IAllocator" % key)
12936
    self._BuildInputData(compat.partial(fn, self), keydata)
12937

    
12938
  def _ComputeClusterData(self):
12939
    """Compute the generic allocator input data.
12940

12941
    This is the data that is independent of the actual operation.
12942

12943
    """
12944
    cfg = self.cfg
12945
    cluster_info = cfg.GetClusterInfo()
12946
    # cluster data
12947
    data = {
12948
      "version": constants.IALLOCATOR_VERSION,
12949
      "cluster_name": cfg.GetClusterName(),
12950
      "cluster_tags": list(cluster_info.GetTags()),
12951
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
12952
      # we don't have job IDs
12953
      }
12954
    ninfo = cfg.GetAllNodesInfo()
12955
    iinfo = cfg.GetAllInstancesInfo().values()
12956
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
12957

    
12958
    # node data
12959
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
12960

    
12961
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
12962
      hypervisor_name = self.hypervisor
12963
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
12964
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
12965
    else:
12966
      hypervisor_name = cluster_info.enabled_hypervisors[0]
12967

    
12968
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
12969
                                        hypervisor_name)
12970
    node_iinfo = \
12971
      self.rpc.call_all_instances_info(node_list,
12972
                                       cluster_info.enabled_hypervisors)
12973

    
12974
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
12975

    
12976
    config_ndata = self._ComputeBasicNodeData(ninfo)
12977
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
12978
                                                 i_list, config_ndata)
12979
    assert len(data["nodes"]) == len(ninfo), \
12980
        "Incomplete node data computed"
12981

    
12982
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
12983

    
12984
    self.in_data = data
12985

    
12986
  @staticmethod
12987
  def _ComputeNodeGroupData(cfg):
12988
    """Compute node groups data.
12989

12990
    """
12991
    ng = dict((guuid, {
12992
      "name": gdata.name,
12993
      "alloc_policy": gdata.alloc_policy,
12994
      })
12995
      for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
12996

    
12997
    return ng
12998

    
12999
  @staticmethod
13000
  def _ComputeBasicNodeData(node_cfg):
13001
    """Compute global node data.
13002

13003
    @rtype: dict
13004
    @returns: a dict of name: (node dict, node config)
13005

13006
    """
13007
    # fill in static (config-based) values
13008
    node_results = dict((ninfo.name, {
13009
      "tags": list(ninfo.GetTags()),
13010
      "primary_ip": ninfo.primary_ip,
13011
      "secondary_ip": ninfo.secondary_ip,
13012
      "offline": ninfo.offline,
13013
      "drained": ninfo.drained,
13014
      "master_candidate": ninfo.master_candidate,
13015
      "group": ninfo.group,
13016
      "master_capable": ninfo.master_capable,
13017
      "vm_capable": ninfo.vm_capable,
13018
      })
13019
      for ninfo in node_cfg.values())
13020

    
13021
    return node_results
13022

    
13023
  @staticmethod
13024
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
13025
                              node_results):
13026
    """Compute global node data.
13027

13028
    @param node_results: the basic node structures as filled from the config
13029

13030
    """
13031
    # make a copy of the current dict
13032
    node_results = dict(node_results)
13033
    for nname, nresult in node_data.items():
13034
      assert nname in node_results, "Missing basic data for node %s" % nname
13035
      ninfo = node_cfg[nname]
13036

    
13037
      if not (ninfo.offline or ninfo.drained):
13038
        nresult.Raise("Can't get data for node %s" % nname)
13039
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
13040
                                nname)
13041
        remote_info = nresult.payload
13042

    
13043
        for attr in ["memory_total", "memory_free", "memory_dom0",
13044
                     "vg_size", "vg_free", "cpu_total"]:
13045
          if attr not in remote_info:
13046
            raise errors.OpExecError("Node '%s' didn't return attribute"
13047
                                     " '%s'" % (nname, attr))
13048
          if not isinstance(remote_info[attr], int):
13049
            raise errors.OpExecError("Node '%s' returned invalid value"
13050
                                     " for '%s': %s" %
13051
                                     (nname, attr, remote_info[attr]))
13052
        # compute memory used by primary instances
13053
        i_p_mem = i_p_up_mem = 0
13054
        for iinfo, beinfo in i_list:
13055
          if iinfo.primary_node == nname:
13056
            i_p_mem += beinfo[constants.BE_MEMORY]
13057
            if iinfo.name not in node_iinfo[nname].payload:
13058
              i_used_mem = 0
13059
            else:
13060
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
13061
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
13062
            remote_info["memory_free"] -= max(0, i_mem_diff)
13063

    
13064
            if iinfo.admin_up:
13065
              i_p_up_mem += beinfo[constants.BE_MEMORY]
13066

    
13067
        # compute memory used by instances
13068
        pnr_dyn = {
13069
          "total_memory": remote_info["memory_total"],
13070
          "reserved_memory": remote_info["memory_dom0"],
13071
          "free_memory": remote_info["memory_free"],
13072
          "total_disk": remote_info["vg_size"],
13073
          "free_disk": remote_info["vg_free"],
13074
          "total_cpus": remote_info["cpu_total"],
13075
          "i_pri_memory": i_p_mem,
13076
          "i_pri_up_memory": i_p_up_mem,
13077
          }
13078
        pnr_dyn.update(node_results[nname])
13079
        node_results[nname] = pnr_dyn
13080

    
13081
    return node_results
13082

    
13083
  @staticmethod
13084
  def _ComputeInstanceData(cluster_info, i_list):
13085
    """Compute global instance data.
13086

13087
    """
13088
    instance_data = {}
13089
    for iinfo, beinfo in i_list:
13090
      nic_data = []
13091
      for nic in iinfo.nics:
13092
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
13093
        nic_dict = {
13094
          "mac": nic.mac,
13095
          "ip": nic.ip,
13096
          "mode": filled_params[constants.NIC_MODE],
13097
          "link": filled_params[constants.NIC_LINK],
13098
          }
13099
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
13100
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
13101
        nic_data.append(nic_dict)
13102
      pir = {
13103
        "tags": list(iinfo.GetTags()),
13104
        "admin_up": iinfo.admin_up,
13105
        "vcpus": beinfo[constants.BE_VCPUS],
13106
        "memory": beinfo[constants.BE_MEMORY],
13107
        "os": iinfo.os,
13108
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
13109
        "nics": nic_data,
13110
        "disks": [{constants.IDISK_SIZE: dsk.size,
13111
                   constants.IDISK_MODE: dsk.mode}
13112
                  for dsk in iinfo.disks],
13113
        "disk_template": iinfo.disk_template,
13114
        "hypervisor": iinfo.hypervisor,
13115
        }
13116
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
13117
                                                 pir["disks"])
13118
      instance_data[iinfo.name] = pir
13119

    
13120
    return instance_data
13121

    
13122
  def _AddNewInstance(self):
13123
    """Add new instance data to allocator structure.
13124

13125
    This in combination with _AllocatorGetClusterData will create the
13126
    correct structure needed as input for the allocator.
13127

13128
    The checks for the completeness of the opcode must have already been
13129
    done.
13130

13131
    """
13132
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
13133

    
13134
    if self.disk_template in constants.DTS_INT_MIRROR:
13135
      self.required_nodes = 2
13136
    else:
13137
      self.required_nodes = 1
13138

    
13139
    request = {
13140
      "name": self.name,
13141
      "disk_template": self.disk_template,
13142
      "tags": self.tags,
13143
      "os": self.os,
13144
      "vcpus": self.vcpus,
13145
      "memory": self.memory,
13146
      "disks": self.disks,
13147
      "disk_space_total": disk_space,
13148
      "nics": self.nics,
13149
      "required_nodes": self.required_nodes,
13150
      "hypervisor": self.hypervisor,
13151
      }
13152

    
13153
    return request
13154

    
13155
  def _AddRelocateInstance(self):
13156
    """Add relocate instance data to allocator structure.
13157

13158
    This in combination with _IAllocatorGetClusterData will create the
13159
    correct structure needed as input for the allocator.
13160

13161
    The checks for the completeness of the opcode must have already been
13162
    done.
13163

13164
    """
13165
    instance = self.cfg.GetInstanceInfo(self.name)
13166
    if instance is None:
13167
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
13168
                                   " IAllocator" % self.name)
13169

    
13170
    if instance.disk_template not in constants.DTS_MIRRORED:
13171
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
13172
                                 errors.ECODE_INVAL)
13173

    
13174
    if instance.disk_template in constants.DTS_INT_MIRROR and \
13175
        len(instance.secondary_nodes) != 1:
13176
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
13177
                                 errors.ECODE_STATE)
13178

    
13179
    self.required_nodes = 1
13180
    disk_sizes = [{constants.IDISK_SIZE: disk.size} for disk in instance.disks]
13181
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
13182

    
13183
    request = {
13184
      "name": self.name,
13185
      "disk_space_total": disk_space,
13186
      "required_nodes": self.required_nodes,
13187
      "relocate_from": self.relocate_from,
13188
      }
13189
    return request
13190

    
13191
  def _AddNodeEvacuate(self):
13192
    """Get data for node-evacuate requests.
13193

13194
    """
13195
    return {
13196
      "instances": self.instances,
13197
      "evac_mode": self.evac_mode,
13198
      }
13199

    
13200
  def _AddChangeGroup(self):
13201
    """Get data for node-evacuate requests.
13202

13203
    """
13204
    return {
13205
      "instances": self.instances,
13206
      "target_groups": self.target_groups,
13207
      }
13208

    
13209
  def _BuildInputData(self, fn, keydata):
13210
    """Build input data structures.
13211

13212
    """
13213
    self._ComputeClusterData()
13214

    
13215
    request = fn()
13216
    request["type"] = self.mode
13217
    for keyname, keytype in keydata:
13218
      if keyname not in request:
13219
        raise errors.ProgrammerError("Request parameter %s is missing" %
13220
                                     keyname)
13221
      val = request[keyname]
13222
      if not keytype(val):
13223
        raise errors.ProgrammerError("Request parameter %s doesn't pass"
13224
                                     " validation, value %s, expected"
13225
                                     " type %s" % (keyname, val, keytype))
13226
    self.in_data["request"] = request
13227

    
13228
    self.in_text = serializer.Dump(self.in_data)
13229

    
13230
  _STRING_LIST = ht.TListOf(ht.TString)
13231
  _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
13232
     # pylint: disable=E1101
13233
     # Class '...' has no 'OP_ID' member
13234
     "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
13235
                          opcodes.OpInstanceMigrate.OP_ID,
13236
                          opcodes.OpInstanceReplaceDisks.OP_ID])
13237
     })))
13238

    
13239
  _NEVAC_MOVED = \
13240
    ht.TListOf(ht.TAnd(ht.TIsLength(3),
13241
                       ht.TItems([ht.TNonEmptyString,
13242
                                  ht.TNonEmptyString,
13243
                                  ht.TListOf(ht.TNonEmptyString),
13244
                                 ])))
13245
  _NEVAC_FAILED = \
13246
    ht.TListOf(ht.TAnd(ht.TIsLength(2),
13247
                       ht.TItems([ht.TNonEmptyString,
13248
                                  ht.TMaybeString,
13249
                                 ])))
13250
  _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
13251
                          ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
13252

    
13253
  _MODE_DATA = {
13254
    constants.IALLOCATOR_MODE_ALLOC:
13255
      (_AddNewInstance,
13256
       [
13257
        ("name", ht.TString),
13258
        ("memory", ht.TInt),
13259
        ("disks", ht.TListOf(ht.TDict)),
13260
        ("disk_template", ht.TString),
13261
        ("os", ht.TString),
13262
        ("tags", _STRING_LIST),
13263
        ("nics", ht.TListOf(ht.TDict)),
13264
        ("vcpus", ht.TInt),
13265
        ("hypervisor", ht.TString),
13266
        ], ht.TList),
13267
    constants.IALLOCATOR_MODE_RELOC:
13268
      (_AddRelocateInstance,
13269
       [("name", ht.TString), ("relocate_from", _STRING_LIST)],
13270
       ht.TList),
13271
     constants.IALLOCATOR_MODE_NODE_EVAC:
13272
      (_AddNodeEvacuate, [
13273
        ("instances", _STRING_LIST),
13274
        ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
13275
        ], _NEVAC_RESULT),
13276
     constants.IALLOCATOR_MODE_CHG_GROUP:
13277
      (_AddChangeGroup, [
13278
        ("instances", _STRING_LIST),
13279
        ("target_groups", _STRING_LIST),
13280
        ], _NEVAC_RESULT),
13281
    }
13282

    
13283
  def Run(self, name, validate=True, call_fn=None):
13284
    """Run an instance allocator and return the results.
13285

13286
    """
13287
    if call_fn is None:
13288
      call_fn = self.rpc.call_iallocator_runner
13289

    
13290
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
13291
    result.Raise("Failure while running the iallocator script")
13292

    
13293
    self.out_text = result.payload
13294
    if validate:
13295
      self._ValidateResult()
13296

    
13297
  def _ValidateResult(self):
13298
    """Process the allocator results.
13299

13300
    This will process and if successful save the result in
13301
    self.out_data and the other parameters.
13302

13303
    """
13304
    try:
13305
      rdict = serializer.Load(self.out_text)
13306
    except Exception, err:
13307
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
13308

    
13309
    if not isinstance(rdict, dict):
13310
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
13311

    
13312
    # TODO: remove backwards compatiblity in later versions
13313
    if "nodes" in rdict and "result" not in rdict:
13314
      rdict["result"] = rdict["nodes"]
13315
      del rdict["nodes"]
13316

    
13317
    for key in "success", "info", "result":
13318
      if key not in rdict:
13319
        raise errors.OpExecError("Can't parse iallocator results:"
13320
                                 " missing key '%s'" % key)
13321
      setattr(self, key, rdict[key])
13322

    
13323
    if not self._result_check(self.result):
13324
      raise errors.OpExecError("Iallocator returned invalid result,"
13325
                               " expected %s, got %s" %
13326
                               (self._result_check, self.result),
13327
                               errors.ECODE_INVAL)
13328

    
13329
    if self.mode == constants.IALLOCATOR_MODE_RELOC:
13330
      assert self.relocate_from is not None
13331
      assert self.required_nodes == 1
13332

    
13333
      node2group = dict((name, ndata["group"])
13334
                        for (name, ndata) in self.in_data["nodes"].items())
13335

    
13336
      fn = compat.partial(self._NodesToGroups, node2group,
13337
                          self.in_data["nodegroups"])
13338

    
13339
      instance = self.cfg.GetInstanceInfo(self.name)
13340
      request_groups = fn(self.relocate_from + [instance.primary_node])
13341
      result_groups = fn(rdict["result"] + [instance.primary_node])
13342

    
13343
      if self.success and not set(result_groups).issubset(request_groups):
13344
        raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
13345
                                 " differ from original groups (%s)" %
13346
                                 (utils.CommaJoin(result_groups),
13347
                                  utils.CommaJoin(request_groups)))
13348

    
13349
    elif self.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
13350
      assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
13351

    
13352
    self.out_data = rdict
13353

    
13354
  @staticmethod
13355
  def _NodesToGroups(node2group, groups, nodes):
13356
    """Returns a list of unique group names for a list of nodes.
13357

13358
    @type node2group: dict
13359
    @param node2group: Map from node name to group UUID
13360
    @type groups: dict
13361
    @param groups: Group information
13362
    @type nodes: list
13363
    @param nodes: Node names
13364

13365
    """
13366
    result = set()
13367

    
13368
    for node in nodes:
13369
      try:
13370
        group_uuid = node2group[node]
13371
      except KeyError:
13372
        # Ignore unknown node
13373
        pass
13374
      else:
13375
        try:
13376
          group = groups[group_uuid]
13377
        except KeyError:
13378
          # Can't find group, let's use UUID
13379
          group_name = group_uuid
13380
        else:
13381
          group_name = group["name"]
13382

    
13383
        result.add(group_name)
13384

    
13385
    return sorted(result)
13386

    
13387

    
13388
class LUTestAllocator(NoHooksLU):
13389
  """Run allocator tests.
13390

13391
  This LU runs the allocator tests
13392

13393
  """
13394
  def CheckPrereq(self):
13395
    """Check prerequisites.
13396

13397
    This checks the opcode parameters depending on the director and mode test.
13398

13399
    """
13400
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
13401
      for attr in ["memory", "disks", "disk_template",
13402
                   "os", "tags", "nics", "vcpus"]:
13403
        if not hasattr(self.op, attr):
13404
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
13405
                                     attr, errors.ECODE_INVAL)
13406
      iname = self.cfg.ExpandInstanceName(self.op.name)
13407
      if iname is not None:
13408
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
13409
                                   iname, errors.ECODE_EXISTS)
13410
      if not isinstance(self.op.nics, list):
13411
        raise errors.OpPrereqError("Invalid parameter 'nics'",
13412
                                   errors.ECODE_INVAL)
13413
      if not isinstance(self.op.disks, list):
13414
        raise errors.OpPrereqError("Invalid parameter 'disks'",
13415
                                   errors.ECODE_INVAL)
13416
      for row in self.op.disks:
13417
        if (not isinstance(row, dict) or
13418
            constants.IDISK_SIZE not in row or
13419
            not isinstance(row[constants.IDISK_SIZE], int) or
13420
            constants.IDISK_MODE not in row or
13421
            row[constants.IDISK_MODE] not in constants.DISK_ACCESS_SET):
13422
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
13423
                                     " parameter", errors.ECODE_INVAL)
13424
      if self.op.hypervisor is None:
13425
        self.op.hypervisor = self.cfg.GetHypervisorType()
13426
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
13427
      fname = _ExpandInstanceName(self.cfg, self.op.name)
13428
      self.op.name = fname
13429
      self.relocate_from = \
13430
          list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
13431
    elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
13432
                          constants.IALLOCATOR_MODE_NODE_EVAC):
13433
      if not self.op.instances:
13434
        raise errors.OpPrereqError("Missing instances", errors.ECODE_INVAL)
13435
      self.op.instances = _GetWantedInstances(self, self.op.instances)
13436
    else:
13437
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
13438
                                 self.op.mode, errors.ECODE_INVAL)
13439

    
13440
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
13441
      if self.op.allocator is None:
13442
        raise errors.OpPrereqError("Missing allocator name",
13443
                                   errors.ECODE_INVAL)
13444
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
13445
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
13446
                                 self.op.direction, errors.ECODE_INVAL)
13447

    
13448
  def Exec(self, feedback_fn):
13449
    """Run the allocator test.
13450

13451
    """
13452
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
13453
      ial = IAllocator(self.cfg, self.rpc,
13454
                       mode=self.op.mode,
13455
                       name=self.op.name,
13456
                       memory=self.op.memory,
13457
                       disks=self.op.disks,
13458
                       disk_template=self.op.disk_template,
13459
                       os=self.op.os,
13460
                       tags=self.op.tags,
13461
                       nics=self.op.nics,
13462
                       vcpus=self.op.vcpus,
13463
                       hypervisor=self.op.hypervisor,
13464
                       )
13465
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
13466
      ial = IAllocator(self.cfg, self.rpc,
13467
                       mode=self.op.mode,
13468
                       name=self.op.name,
13469
                       relocate_from=list(self.relocate_from),
13470
                       )
13471
    elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
13472
      ial = IAllocator(self.cfg, self.rpc,
13473
                       mode=self.op.mode,
13474
                       instances=self.op.instances,
13475
                       target_groups=self.op.target_groups)
13476
    elif self.op.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
13477
      ial = IAllocator(self.cfg, self.rpc,
13478
                       mode=self.op.mode,
13479
                       instances=self.op.instances,
13480
                       evac_mode=self.op.evac_mode)
13481
    else:
13482
      raise errors.ProgrammerError("Uncatched mode %s in"
13483
                                   " LUTestAllocator.Exec", self.op.mode)
13484

    
13485
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
13486
      result = ial.in_text
13487
    else:
13488
      ial.Run(self.op.allocator, validate=False)
13489
      result = ial.out_text
13490
    return result
13491

    
13492

    
13493
#: Query type implementations
13494
_QUERY_IMPL = {
13495
  constants.QR_INSTANCE: _InstanceQuery,
13496
  constants.QR_NODE: _NodeQuery,
13497
  constants.QR_GROUP: _GroupQuery,
13498
  constants.QR_OS: _OsQuery,
13499
  }
13500

    
13501
assert set(_QUERY_IMPL.keys()) == constants.QR_VIA_OP
13502

    
13503

    
13504
def _GetQueryImplementation(name):
13505
  """Returns the implemtnation for a query type.
13506

13507
  @param name: Query type, must be one of L{constants.QR_VIA_OP}
13508

13509
  """
13510
  try:
13511
    return _QUERY_IMPL[name]
13512
  except KeyError:
13513
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
13514
                               errors.ECODE_INVAL)