Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ a5485ffc

History | View | Annotate | Download (475.4 kB)

1
#
2
#
3

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

    
21

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

    
24
# pylint: disable=W0201,C0302
25

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

    
29
# C0302: since we have waaaay 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
  REG_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: