Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 666e013f

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 = True
1559

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

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

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

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

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

    
1590
    feedback_fn("* Verifying cluster config")
1591

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

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

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

    
1601
    feedback_fn("* Verifying hypervisor parameters")
1602

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

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

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

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

    
1615
    dangling_instances = {}
1616
    no_node_instances = []
1617

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

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

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

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

    
1639
    return not self.bad
1640

    
1641

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

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

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

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

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

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

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

    
1706
    # Get instances in node group; this is unsafe and needs verification later
1707
    inst_names = \
1708
      self.cfg.GetNodeGroupInstances(self.group_uuid, primary_only=True)
1709

    
1710
    self.needed_locks = {
1711
      locking.LEVEL_INSTANCE: inst_names,
1712
      locking.LEVEL_NODEGROUP: [self.group_uuid],
1713
      locking.LEVEL_NODE: [],
1714
      }
1715

    
1716
    self.share_locks = _ShareAll()
1717

    
1718
  def DeclareLocks(self, level):
1719
    if level == locking.LEVEL_NODE:
1720
      # Get members of node group; this is unsafe and needs verification later
1721
      nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
1722

    
1723
      all_inst_info = self.cfg.GetAllInstancesInfo()
1724

    
1725
      # In Exec(), we warn about mirrored instances that have primary and
1726
      # secondary living in separate node groups. To fully verify that
1727
      # volumes for these instances are healthy, we will need to do an
1728
      # extra call to their secondaries. We ensure here those nodes will
1729
      # be locked.
1730
      for inst in self.owned_locks(locking.LEVEL_INSTANCE):
1731
        # Important: access only the instances whose lock is owned
1732
        if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
1733
          nodes.update(all_inst_info[inst].secondary_nodes)
1734

    
1735
      self.needed_locks[locking.LEVEL_NODE] = nodes
1736

    
1737
  def CheckPrereq(self):
1738
    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
1739
    self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
1740

    
1741
    group_nodes = set(self.group_info.members)
1742
    group_instances = \
1743
      self.cfg.GetNodeGroupInstances(self.group_uuid, primary_only=True)
1744

    
1745
    unlocked_nodes = \
1746
        group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1747

    
1748
    unlocked_instances = \
1749
        group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
1750

    
1751
    if unlocked_nodes:
1752
      raise errors.OpPrereqError("Missing lock for nodes: %s" %
1753
                                 utils.CommaJoin(unlocked_nodes),
1754
                                 errors.ECODE_STATE)
1755

    
1756
    if unlocked_instances:
1757
      raise errors.OpPrereqError("Missing lock for instances: %s" %
1758
                                 utils.CommaJoin(unlocked_instances),
1759
                                 errors.ECODE_STATE)
1760

    
1761
    self.all_node_info = self.cfg.GetAllNodesInfo()
1762
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1763

    
1764
    self.my_node_names = utils.NiceSort(group_nodes)
1765
    self.my_inst_names = utils.NiceSort(group_instances)
1766

    
1767
    self.my_node_info = dict((name, self.all_node_info[name])
1768
                             for name in self.my_node_names)
1769

    
1770
    self.my_inst_info = dict((name, self.all_inst_info[name])
1771
                             for name in self.my_inst_names)
1772

    
1773
    # We detect here the nodes that will need the extra RPC calls for verifying
1774
    # split LV volumes; they should be locked.
1775
    extra_lv_nodes = set()
1776

    
1777
    for inst in self.my_inst_info.values():
1778
      if inst.disk_template in constants.DTS_INT_MIRROR:
1779
        for nname in inst.all_nodes:
1780
          if self.all_node_info[nname].group != self.group_uuid:
1781
            extra_lv_nodes.add(nname)
1782

    
1783
    unlocked_lv_nodes = \
1784
        extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1785

    
1786
    if unlocked_lv_nodes:
1787
      raise errors.OpPrereqError("Missing node locks for LV check: %s" %
1788
                                 utils.CommaJoin(unlocked_lv_nodes),
1789
                                 errors.ECODE_STATE)
1790
    self.extra_lv_nodes = list(extra_lv_nodes)
1791

    
1792
  def _VerifyNode(self, ninfo, nresult):
1793
    """Perform some basic validation on data returned from a node.
1794

1795
      - check the result data structure is well formed and has all the
1796
        mandatory fields
1797
      - check ganeti version
1798

1799
    @type ninfo: L{objects.Node}
1800
    @param ninfo: the node to check
1801
    @param nresult: the results from the node
1802
    @rtype: boolean
1803
    @return: whether overall this call was successful (and we can expect
1804
         reasonable values in the respose)
1805

1806
    """
1807
    node = ninfo.name
1808
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1809

    
1810
    # main result, nresult should be a non-empty dict
1811
    test = not nresult or not isinstance(nresult, dict)
1812
    _ErrorIf(test, self.ENODERPC, node,
1813
                  "unable to verify node: no data returned")
1814
    if test:
1815
      return False
1816

    
1817
    # compares ganeti version
1818
    local_version = constants.PROTOCOL_VERSION
1819
    remote_version = nresult.get("version", None)
1820
    test = not (remote_version and
1821
                isinstance(remote_version, (list, tuple)) and
1822
                len(remote_version) == 2)
1823
    _ErrorIf(test, self.ENODERPC, node,
1824
             "connection to node returned invalid data")
1825
    if test:
1826
      return False
1827

    
1828
    test = local_version != remote_version[0]
1829
    _ErrorIf(test, self.ENODEVERSION, node,
1830
             "incompatible protocol versions: master %s,"
1831
             " node %s", local_version, remote_version[0])
1832
    if test:
1833
      return False
1834

    
1835
    # node seems compatible, we can actually try to look into its results
1836

    
1837
    # full package version
1838
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1839
                  self.ENODEVERSION, node,
1840
                  "software version mismatch: master %s, node %s",
1841
                  constants.RELEASE_VERSION, remote_version[1],
1842
                  code=self.ETYPE_WARNING)
1843

    
1844
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1845
    if ninfo.vm_capable and isinstance(hyp_result, dict):
1846
      for hv_name, hv_result in hyp_result.iteritems():
1847
        test = hv_result is not None
1848
        _ErrorIf(test, self.ENODEHV, node,
1849
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1850

    
1851
    hvp_result = nresult.get(constants.NV_HVPARAMS, None)
1852
    if ninfo.vm_capable and isinstance(hvp_result, list):
1853
      for item, hv_name, hv_result in hvp_result:
1854
        _ErrorIf(True, self.ENODEHV, node,
1855
                 "hypervisor %s parameter verify failure (source %s): %s",
1856
                 hv_name, item, hv_result)
1857

    
1858
    test = nresult.get(constants.NV_NODESETUP,
1859
                       ["Missing NODESETUP results"])
1860
    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
1861
             "; ".join(test))
1862

    
1863
    return True
1864

    
1865
  def _VerifyNodeTime(self, ninfo, nresult,
1866
                      nvinfo_starttime, nvinfo_endtime):
1867
    """Check the node time.
1868

1869
    @type ninfo: L{objects.Node}
1870
    @param ninfo: the node to check
1871
    @param nresult: the remote results for the node
1872
    @param nvinfo_starttime: the start time of the RPC call
1873
    @param nvinfo_endtime: the end time of the RPC call
1874

1875
    """
1876
    node = ninfo.name
1877
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1878

    
1879
    ntime = nresult.get(constants.NV_TIME, None)
1880
    try:
1881
      ntime_merged = utils.MergeTime(ntime)
1882
    except (ValueError, TypeError):
1883
      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
1884
      return
1885

    
1886
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1887
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1888
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1889
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1890
    else:
1891
      ntime_diff = None
1892

    
1893
    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
1894
             "Node time diverges by at least %s from master node time",
1895
             ntime_diff)
1896

    
1897
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1898
    """Check the node LVM results.
1899

1900
    @type ninfo: L{objects.Node}
1901
    @param ninfo: the node to check
1902
    @param nresult: the remote results for the node
1903
    @param vg_name: the configured VG name
1904

1905
    """
1906
    if vg_name is None:
1907
      return
1908

    
1909
    node = ninfo.name
1910
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1911

    
1912
    # checks vg existence and size > 20G
1913
    vglist = nresult.get(constants.NV_VGLIST, None)
1914
    test = not vglist
1915
    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
1916
    if not test:
1917
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
1918
                                            constants.MIN_VG_SIZE)
1919
      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
1920

    
1921
    # check pv names
1922
    pvlist = nresult.get(constants.NV_PVLIST, None)
1923
    test = pvlist is None
1924
    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
1925
    if not test:
1926
      # check that ':' is not present in PV names, since it's a
1927
      # special character for lvcreate (denotes the range of PEs to
1928
      # use on the PV)
1929
      for _, pvname, owner_vg in pvlist:
1930
        test = ":" in pvname
1931
        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
1932
                 " '%s' of VG '%s'", pvname, owner_vg)
1933

    
1934
  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
1935
    """Check the node bridges.
1936

1937
    @type ninfo: L{objects.Node}
1938
    @param ninfo: the node to check
1939
    @param nresult: the remote results for the node
1940
    @param bridges: the expected list of bridges
1941

1942
    """
1943
    if not bridges:
1944
      return
1945

    
1946
    node = ninfo.name
1947
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1948

    
1949
    missing = nresult.get(constants.NV_BRIDGES, None)
1950
    test = not isinstance(missing, list)
1951
    _ErrorIf(test, self.ENODENET, node,
1952
             "did not return valid bridge information")
1953
    if not test:
1954
      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
1955
               utils.CommaJoin(sorted(missing)))
1956

    
1957
  def _VerifyNodeNetwork(self, ninfo, nresult):
1958
    """Check the node network connectivity results.
1959

1960
    @type ninfo: L{objects.Node}
1961
    @param ninfo: the node to check
1962
    @param nresult: the remote results for the node
1963

1964
    """
1965
    node = ninfo.name
1966
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
1967

    
1968
    test = constants.NV_NODELIST not in nresult
1969
    _ErrorIf(test, self.ENODESSH, node,
1970
             "node hasn't returned node ssh connectivity data")
1971
    if not test:
1972
      if nresult[constants.NV_NODELIST]:
1973
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1974
          _ErrorIf(True, self.ENODESSH, node,
1975
                   "ssh communication with node '%s': %s", a_node, a_msg)
1976

    
1977
    test = constants.NV_NODENETTEST not in nresult
1978
    _ErrorIf(test, self.ENODENET, node,
1979
             "node hasn't returned node tcp connectivity data")
1980
    if not test:
1981
      if nresult[constants.NV_NODENETTEST]:
1982
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1983
        for anode in nlist:
1984
          _ErrorIf(True, self.ENODENET, node,
1985
                   "tcp communication with node '%s': %s",
1986
                   anode, nresult[constants.NV_NODENETTEST][anode])
1987

    
1988
    test = constants.NV_MASTERIP not in nresult
1989
    _ErrorIf(test, self.ENODENET, node,
1990
             "node hasn't returned node master IP reachability data")
1991
    if not test:
1992
      if not nresult[constants.NV_MASTERIP]:
1993
        if node == self.master_node:
1994
          msg = "the master node cannot reach the master IP (not configured?)"
1995
        else:
1996
          msg = "cannot reach the master IP"
1997
        _ErrorIf(True, self.ENODENET, node, msg)
1998

    
1999
  def _VerifyInstance(self, instance, instanceconfig, node_image,
2000
                      diskstatus):
2001
    """Verify an instance.
2002

2003
    This function checks to see if the required block devices are
2004
    available on the instance's node.
2005

2006
    """
2007
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2008
    node_current = instanceconfig.primary_node
2009

    
2010
    node_vol_should = {}
2011
    instanceconfig.MapLVsByNode(node_vol_should)
2012

    
2013
    for node in node_vol_should:
2014
      n_img = node_image[node]
2015
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2016
        # ignore missing volumes on offline or broken nodes
2017
        continue
2018
      for volume in node_vol_should[node]:
2019
        test = volume not in n_img.volumes
2020
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
2021
                 "volume %s missing on node %s", volume, node)
2022

    
2023
    if instanceconfig.admin_up:
2024
      pri_img = node_image[node_current]
2025
      test = instance not in pri_img.instances and not pri_img.offline
2026
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
2027
               "instance not running on its primary node %s",
2028
               node_current)
2029

    
2030
    diskdata = [(nname, success, status, idx)
2031
                for (nname, disks) in diskstatus.items()
2032
                for idx, (success, status) in enumerate(disks)]
2033

    
2034
    for nname, success, bdev_status, idx in diskdata:
2035
      # the 'ghost node' construction in Exec() ensures that we have a
2036
      # node here
2037
      snode = node_image[nname]
2038
      bad_snode = snode.ghost or snode.offline
2039
      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
2040
               self.EINSTANCEFAULTYDISK, instance,
2041
               "couldn't retrieve status for disk/%s on %s: %s",
2042
               idx, nname, bdev_status)
2043
      _ErrorIf((instanceconfig.admin_up and success and
2044
                bdev_status.ldisk_status == constants.LDS_FAULTY),
2045
               self.EINSTANCEFAULTYDISK, instance,
2046
               "disk/%s on %s is faulty", idx, nname)
2047

    
2048
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
2049
    """Verify if there are any unknown volumes in the cluster.
2050

2051
    The .os, .swap and backup volumes are ignored. All other volumes are
2052
    reported as unknown.
2053

2054
    @type reserved: L{ganeti.utils.FieldSet}
2055
    @param reserved: a FieldSet of reserved volume names
2056

2057
    """
2058
    for node, n_img in node_image.items():
2059
      if (n_img.offline or n_img.rpc_fail or n_img.lvm_fail or
2060
          self.all_node_info[node].group != self.group_uuid):
2061
        # skip non-healthy nodes
2062
        continue
2063
      for volume in n_img.volumes:
2064
        test = ((node not in node_vol_should or
2065
                volume not in node_vol_should[node]) and
2066
                not reserved.Matches(volume))
2067
        self._ErrorIf(test, self.ENODEORPHANLV, node,
2068
                      "volume %s is unknown", volume)
2069

    
2070
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
2071
    """Verify N+1 Memory Resilience.
2072

2073
    Check that if one single node dies we can still start all the
2074
    instances it was primary for.
2075

2076
    """
2077
    cluster_info = self.cfg.GetClusterInfo()
2078
    for node, n_img in node_image.items():
2079
      # This code checks that every node which is now listed as
2080
      # secondary has enough memory to host all instances it is
2081
      # supposed to should a single other node in the cluster fail.
2082
      # FIXME: not ready for failover to an arbitrary node
2083
      # FIXME: does not support file-backed instances
2084
      # WARNING: we currently take into account down instances as well
2085
      # as up ones, considering that even if they're down someone
2086
      # might want to start them even in the event of a node failure.
2087
      if n_img.offline or self.all_node_info[node].group != self.group_uuid:
2088
        # we're skipping nodes marked offline and nodes in other groups from
2089
        # the N+1 warning, since most likely we don't have good memory
2090
        # infromation from them; we already list instances living on such
2091
        # nodes, and that's enough warning
2092
        continue
2093
      for prinode, instances in n_img.sbp.items():
2094
        needed_mem = 0
2095
        for instance in instances:
2096
          bep = cluster_info.FillBE(instance_cfg[instance])
2097
          if bep[constants.BE_AUTO_BALANCE]:
2098
            needed_mem += bep[constants.BE_MEMORY]
2099
        test = n_img.mfree < needed_mem
2100
        self._ErrorIf(test, self.ENODEN1, node,
2101
                      "not enough memory to accomodate instance failovers"
2102
                      " should node %s fail (%dMiB needed, %dMiB available)",
2103
                      prinode, needed_mem, n_img.mfree)
2104

    
2105
  @classmethod
2106
  def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
2107
                   (files_all, files_all_opt, files_mc, files_vm)):
2108
    """Verifies file checksums collected from all nodes.
2109

2110
    @param errorif: Callback for reporting errors
2111
    @param nodeinfo: List of L{objects.Node} objects
2112
    @param master_node: Name of master node
2113
    @param all_nvinfo: RPC results
2114

2115
    """
2116
    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
2117
            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
2118
           "Found file listed in more than one file list"
2119

    
2120
    # Define functions determining which nodes to consider for a file
2121
    files2nodefn = [
2122
      (files_all, None),
2123
      (files_all_opt, None),
2124
      (files_mc, lambda node: (node.master_candidate or
2125
                               node.name == master_node)),
2126
      (files_vm, lambda node: node.vm_capable),
2127
      ]
2128

    
2129
    # Build mapping from filename to list of nodes which should have the file
2130
    nodefiles = {}
2131
    for (files, fn) in files2nodefn:
2132
      if fn is None:
2133
        filenodes = nodeinfo
2134
      else:
2135
        filenodes = filter(fn, nodeinfo)
2136
      nodefiles.update((filename,
2137
                        frozenset(map(operator.attrgetter("name"), filenodes)))
2138
                       for filename in files)
2139

    
2140
    assert set(nodefiles) == (files_all | files_all_opt | files_mc | files_vm)
2141

    
2142
    fileinfo = dict((filename, {}) for filename in nodefiles)
2143
    ignore_nodes = set()
2144

    
2145
    for node in nodeinfo:
2146
      if node.offline:
2147
        ignore_nodes.add(node.name)
2148
        continue
2149

    
2150
      nresult = all_nvinfo[node.name]
2151

    
2152
      if nresult.fail_msg or not nresult.payload:
2153
        node_files = None
2154
      else:
2155
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
2156

    
2157
      test = not (node_files and isinstance(node_files, dict))
2158
      errorif(test, cls.ENODEFILECHECK, node.name,
2159
              "Node did not return file checksum data")
2160
      if test:
2161
        ignore_nodes.add(node.name)
2162
        continue
2163

    
2164
      # Build per-checksum mapping from filename to nodes having it
2165
      for (filename, checksum) in node_files.items():
2166
        assert filename in nodefiles
2167
        fileinfo[filename].setdefault(checksum, set()).add(node.name)
2168

    
2169
    for (filename, checksums) in fileinfo.items():
2170
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2171

    
2172
      # Nodes having the file
2173
      with_file = frozenset(node_name
2174
                            for nodes in fileinfo[filename].values()
2175
                            for node_name in nodes) - ignore_nodes
2176

    
2177
      expected_nodes = nodefiles[filename] - ignore_nodes
2178

    
2179
      # Nodes missing file
2180
      missing_file = expected_nodes - with_file
2181

    
2182
      if filename in files_all_opt:
2183
        # All or no nodes
2184
        errorif(missing_file and missing_file != expected_nodes,
2185
                cls.ECLUSTERFILECHECK, None,
2186
                "File %s is optional, but it must exist on all or no"
2187
                " nodes (not found on %s)",
2188
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2189
      else:
2190
        # Non-optional files
2191
        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
2192
                "File %s is missing from node(s) %s", filename,
2193
                utils.CommaJoin(utils.NiceSort(missing_file)))
2194

    
2195
        # Warn if a node has a file it shouldn't
2196
        unexpected = with_file - expected_nodes
2197
        errorif(unexpected,
2198
                cls.ECLUSTERFILECHECK, None,
2199
                "File %s should not exist on node(s) %s",
2200
                filename, utils.CommaJoin(utils.NiceSort(unexpected)))
2201

    
2202
      # See if there are multiple versions of the file
2203
      test = len(checksums) > 1
2204
      if test:
2205
        variants = ["variant %s on %s" %
2206
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2207
                    for (idx, (checksum, nodes)) in
2208
                      enumerate(sorted(checksums.items()))]
2209
      else:
2210
        variants = []
2211

    
2212
      errorif(test, cls.ECLUSTERFILECHECK, None,
2213
              "File %s found with %s different checksums (%s)",
2214
              filename, len(checksums), "; ".join(variants))
2215

    
2216
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2217
                      drbd_map):
2218
    """Verifies and the node DRBD status.
2219

2220
    @type ninfo: L{objects.Node}
2221
    @param ninfo: the node to check
2222
    @param nresult: the remote results for the node
2223
    @param instanceinfo: the dict of instances
2224
    @param drbd_helper: the configured DRBD usermode helper
2225
    @param drbd_map: the DRBD map as returned by
2226
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2227

2228
    """
2229
    node = ninfo.name
2230
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2231

    
2232
    if drbd_helper:
2233
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2234
      test = (helper_result == None)
2235
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
2236
               "no drbd usermode helper returned")
2237
      if helper_result:
2238
        status, payload = helper_result
2239
        test = not status
2240
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2241
                 "drbd usermode helper check unsuccessful: %s", payload)
2242
        test = status and (payload != drbd_helper)
2243
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2244
                 "wrong drbd usermode helper: %s", payload)
2245

    
2246
    # compute the DRBD minors
2247
    node_drbd = {}
2248
    for minor, instance in drbd_map[node].items():
2249
      test = instance not in instanceinfo
2250
      _ErrorIf(test, self.ECLUSTERCFG, None,
2251
               "ghost instance '%s' in temporary DRBD map", instance)
2252
        # ghost instance should not be running, but otherwise we
2253
        # don't give double warnings (both ghost instance and
2254
        # unallocated minor in use)
2255
      if test:
2256
        node_drbd[minor] = (instance, False)
2257
      else:
2258
        instance = instanceinfo[instance]
2259
        node_drbd[minor] = (instance.name, instance.admin_up)
2260

    
2261
    # and now check them
2262
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2263
    test = not isinstance(used_minors, (tuple, list))
2264
    _ErrorIf(test, self.ENODEDRBD, node,
2265
             "cannot parse drbd status file: %s", str(used_minors))
2266
    if test:
2267
      # we cannot check drbd status
2268
      return
2269

    
2270
    for minor, (iname, must_exist) in node_drbd.items():
2271
      test = minor not in used_minors and must_exist
2272
      _ErrorIf(test, self.ENODEDRBD, node,
2273
               "drbd minor %d of instance %s is not active", minor, iname)
2274
    for minor in used_minors:
2275
      test = minor not in node_drbd
2276
      _ErrorIf(test, self.ENODEDRBD, node,
2277
               "unallocated drbd minor %d is in use", minor)
2278

    
2279
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2280
    """Builds the node OS structures.
2281

2282
    @type ninfo: L{objects.Node}
2283
    @param ninfo: the node to check
2284
    @param nresult: the remote results for the node
2285
    @param nimg: the node image object
2286

2287
    """
2288
    node = ninfo.name
2289
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2290

    
2291
    remote_os = nresult.get(constants.NV_OSLIST, None)
2292
    test = (not isinstance(remote_os, list) or
2293
            not compat.all(isinstance(v, list) and len(v) == 7
2294
                           for v in remote_os))
2295

    
2296
    _ErrorIf(test, self.ENODEOS, node,
2297
             "node hasn't returned valid OS data")
2298

    
2299
    nimg.os_fail = test
2300

    
2301
    if test:
2302
      return
2303

    
2304
    os_dict = {}
2305

    
2306
    for (name, os_path, status, diagnose,
2307
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2308

    
2309
      if name not in os_dict:
2310
        os_dict[name] = []
2311

    
2312
      # parameters is a list of lists instead of list of tuples due to
2313
      # JSON lacking a real tuple type, fix it:
2314
      parameters = [tuple(v) for v in parameters]
2315
      os_dict[name].append((os_path, status, diagnose,
2316
                            set(variants), set(parameters), set(api_ver)))
2317

    
2318
    nimg.oslist = os_dict
2319

    
2320
  def _VerifyNodeOS(self, ninfo, nimg, base):
2321
    """Verifies the node OS list.
2322

2323
    @type ninfo: L{objects.Node}
2324
    @param ninfo: the node to check
2325
    @param nimg: the node image object
2326
    @param base: the 'template' node we match against (e.g. from the master)
2327

2328
    """
2329
    node = ninfo.name
2330
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2331

    
2332
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2333

    
2334
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2335
    for os_name, os_data in nimg.oslist.items():
2336
      assert os_data, "Empty OS status for OS %s?!" % os_name
2337
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2338
      _ErrorIf(not f_status, self.ENODEOS, node,
2339
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2340
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
2341
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2342
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2343
      # comparisons with the 'base' image
2344
      test = os_name not in base.oslist
2345
      _ErrorIf(test, self.ENODEOS, node,
2346
               "Extra OS %s not present on reference node (%s)",
2347
               os_name, base.name)
2348
      if test:
2349
        continue
2350
      assert base.oslist[os_name], "Base node has empty OS status?"
2351
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2352
      if not b_status:
2353
        # base OS is invalid, skipping
2354
        continue
2355
      for kind, a, b in [("API version", f_api, b_api),
2356
                         ("variants list", f_var, b_var),
2357
                         ("parameters", beautify_params(f_param),
2358
                          beautify_params(b_param))]:
2359
        _ErrorIf(a != b, self.ENODEOS, node,
2360
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2361
                 kind, os_name, base.name,
2362
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2363

    
2364
    # check any missing OSes
2365
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2366
    _ErrorIf(missing, self.ENODEOS, node,
2367
             "OSes present on reference node %s but missing on this node: %s",
2368
             base.name, utils.CommaJoin(missing))
2369

    
2370
  def _VerifyOob(self, ninfo, nresult):
2371
    """Verifies out of band functionality of a node.
2372

2373
    @type ninfo: L{objects.Node}
2374
    @param ninfo: the node to check
2375
    @param nresult: the remote results for the node
2376

2377
    """
2378
    node = ninfo.name
2379
    # We just have to verify the paths on master and/or master candidates
2380
    # as the oob helper is invoked on the master
2381
    if ((ninfo.master_candidate or ninfo.master_capable) and
2382
        constants.NV_OOB_PATHS in nresult):
2383
      for path_result in nresult[constants.NV_OOB_PATHS]:
2384
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
2385

    
2386
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2387
    """Verifies and updates the node volume data.
2388

2389
    This function will update a L{NodeImage}'s internal structures
2390
    with data from the remote call.
2391

2392
    @type ninfo: L{objects.Node}
2393
    @param ninfo: the node to check
2394
    @param nresult: the remote results for the node
2395
    @param nimg: the node image object
2396
    @param vg_name: the configured VG name
2397

2398
    """
2399
    node = ninfo.name
2400
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2401

    
2402
    nimg.lvm_fail = True
2403
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2404
    if vg_name is None:
2405
      pass
2406
    elif isinstance(lvdata, basestring):
2407
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
2408
               utils.SafeEncode(lvdata))
2409
    elif not isinstance(lvdata, dict):
2410
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
2411
    else:
2412
      nimg.volumes = lvdata
2413
      nimg.lvm_fail = False
2414

    
2415
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2416
    """Verifies and updates the node instance list.
2417

2418
    If the listing was successful, then updates this node's instance
2419
    list. Otherwise, it marks the RPC call as failed for the instance
2420
    list key.
2421

2422
    @type ninfo: L{objects.Node}
2423
    @param ninfo: the node to check
2424
    @param nresult: the remote results for the node
2425
    @param nimg: the node image object
2426

2427
    """
2428
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2429
    test = not isinstance(idata, list)
2430
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
2431
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
2432
    if test:
2433
      nimg.hyp_fail = True
2434
    else:
2435
      nimg.instances = idata
2436

    
2437
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2438
    """Verifies and computes a node information map
2439

2440
    @type ninfo: L{objects.Node}
2441
    @param ninfo: the node to check
2442
    @param nresult: the remote results for the node
2443
    @param nimg: the node image object
2444
    @param vg_name: the configured VG name
2445

2446
    """
2447
    node = ninfo.name
2448
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2449

    
2450
    # try to read free memory (from the hypervisor)
2451
    hv_info = nresult.get(constants.NV_HVINFO, None)
2452
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2453
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
2454
    if not test:
2455
      try:
2456
        nimg.mfree = int(hv_info["memory_free"])
2457
      except (ValueError, TypeError):
2458
        _ErrorIf(True, self.ENODERPC, node,
2459
                 "node returned invalid nodeinfo, check hypervisor")
2460

    
2461
    # FIXME: devise a free space model for file based instances as well
2462
    if vg_name is not None:
2463
      test = (constants.NV_VGLIST not in nresult or
2464
              vg_name not in nresult[constants.NV_VGLIST])
2465
      _ErrorIf(test, self.ENODELVM, node,
2466
               "node didn't return data for the volume group '%s'"
2467
               " - it is either missing or broken", vg_name)
2468
      if not test:
2469
        try:
2470
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2471
        except (ValueError, TypeError):
2472
          _ErrorIf(True, self.ENODERPC, node,
2473
                   "node returned invalid LVM info, check LVM status")
2474

    
2475
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2476
    """Gets per-disk status information for all instances.
2477

2478
    @type nodelist: list of strings
2479
    @param nodelist: Node names
2480
    @type node_image: dict of (name, L{objects.Node})
2481
    @param node_image: Node objects
2482
    @type instanceinfo: dict of (name, L{objects.Instance})
2483
    @param instanceinfo: Instance objects
2484
    @rtype: {instance: {node: [(succes, payload)]}}
2485
    @return: a dictionary of per-instance dictionaries with nodes as
2486
        keys and disk information as values; the disk information is a
2487
        list of tuples (success, payload)
2488

2489
    """
2490
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2491

    
2492
    node_disks = {}
2493
    node_disks_devonly = {}
2494
    diskless_instances = set()
2495
    diskless = constants.DT_DISKLESS
2496

    
2497
    for nname in nodelist:
2498
      node_instances = list(itertools.chain(node_image[nname].pinst,
2499
                                            node_image[nname].sinst))
2500
      diskless_instances.update(inst for inst in node_instances
2501
                                if instanceinfo[inst].disk_template == diskless)
2502
      disks = [(inst, disk)
2503
               for inst in node_instances
2504
               for disk in instanceinfo[inst].disks]
2505

    
2506
      if not disks:
2507
        # No need to collect data
2508
        continue
2509

    
2510
      node_disks[nname] = disks
2511

    
2512
      # Creating copies as SetDiskID below will modify the objects and that can
2513
      # lead to incorrect data returned from nodes
2514
      devonly = [dev.Copy() for (_, dev) in disks]
2515

    
2516
      for dev in devonly:
2517
        self.cfg.SetDiskID(dev, nname)
2518

    
2519
      node_disks_devonly[nname] = devonly
2520

    
2521
    assert len(node_disks) == len(node_disks_devonly)
2522

    
2523
    # Collect data from all nodes with disks
2524
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2525
                                                          node_disks_devonly)
2526

    
2527
    assert len(result) == len(node_disks)
2528

    
2529
    instdisk = {}
2530

    
2531
    for (nname, nres) in result.items():
2532
      disks = node_disks[nname]
2533

    
2534
      if nres.offline:
2535
        # No data from this node
2536
        data = len(disks) * [(False, "node offline")]
2537
      else:
2538
        msg = nres.fail_msg
2539
        _ErrorIf(msg, self.ENODERPC, nname,
2540
                 "while getting disk information: %s", msg)
2541
        if msg:
2542
          # No data from this node
2543
          data = len(disks) * [(False, msg)]
2544
        else:
2545
          data = []
2546
          for idx, i in enumerate(nres.payload):
2547
            if isinstance(i, (tuple, list)) and len(i) == 2:
2548
              data.append(i)
2549
            else:
2550
              logging.warning("Invalid result from node %s, entry %d: %s",
2551
                              nname, idx, i)
2552
              data.append((False, "Invalid result from the remote node"))
2553

    
2554
      for ((inst, _), status) in zip(disks, data):
2555
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2556

    
2557
    # Add empty entries for diskless instances.
2558
    for inst in diskless_instances:
2559
      assert inst not in instdisk
2560
      instdisk[inst] = {}
2561

    
2562
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2563
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2564
                      compat.all(isinstance(s, (tuple, list)) and
2565
                                 len(s) == 2 for s in statuses)
2566
                      for inst, nnames in instdisk.items()
2567
                      for nname, statuses in nnames.items())
2568
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2569

    
2570
    return instdisk
2571

    
2572
  @staticmethod
2573
  def _SshNodeSelector(group_uuid, all_nodes):
2574
    """Create endless iterators for all potential SSH check hosts.
2575

2576
    """
2577
    nodes = [node for node in all_nodes
2578
             if (node.group != group_uuid and
2579
                 not node.offline)]
2580
    keyfunc = operator.attrgetter("group")
2581

    
2582
    return map(itertools.cycle,
2583
               [sorted(map(operator.attrgetter("name"), names))
2584
                for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
2585
                                                  keyfunc)])
2586

    
2587
  @classmethod
2588
  def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
2589
    """Choose which nodes should talk to which other nodes.
2590

2591
    We will make nodes contact all nodes in their group, and one node from
2592
    every other group.
2593

2594
    @warning: This algorithm has a known issue if one node group is much
2595
      smaller than others (e.g. just one node). In such a case all other
2596
      nodes will talk to the single node.
2597

2598
    """
2599
    online_nodes = sorted(node.name for node in group_nodes if not node.offline)
2600
    sel = cls._SshNodeSelector(group_uuid, all_nodes)
2601

    
2602
    return (online_nodes,
2603
            dict((name, sorted([i.next() for i in sel]))
2604
                 for name in online_nodes))
2605

    
2606
  def BuildHooksEnv(self):
2607
    """Build hooks env.
2608

2609
    Cluster-Verify hooks just ran in the post phase and their failure makes
2610
    the output be logged in the verify output and the verification to fail.
2611

2612
    """
2613
    env = {
2614
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2615
      }
2616

    
2617
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2618
               for node in self.my_node_info.values())
2619

    
2620
    return env
2621

    
2622
  def BuildHooksNodes(self):
2623
    """Build hooks nodes.
2624

2625
    """
2626
    return ([], self.my_node_names)
2627

    
2628
  def Exec(self, feedback_fn):
2629
    """Verify integrity of the node group, performing various test on nodes.
2630

2631
    """
2632
    # This method has too many local variables. pylint: disable=R0914
2633
    feedback_fn("* Verifying group '%s'" % self.group_info.name)
2634

    
2635
    if not self.my_node_names:
2636
      # empty node group
2637
      feedback_fn("* Empty node group, skipping verification")
2638
      return True
2639

    
2640
    self.bad = False
2641
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2642
    verbose = self.op.verbose
2643
    self._feedback_fn = feedback_fn
2644

    
2645
    vg_name = self.cfg.GetVGName()
2646
    drbd_helper = self.cfg.GetDRBDHelper()
2647
    cluster = self.cfg.GetClusterInfo()
2648
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2649
    hypervisors = cluster.enabled_hypervisors
2650
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2651

    
2652
    i_non_redundant = [] # Non redundant instances
2653
    i_non_a_balanced = [] # Non auto-balanced instances
2654
    n_offline = 0 # Count of offline nodes
2655
    n_drained = 0 # Count of nodes being drained
2656
    node_vol_should = {}
2657

    
2658
    # FIXME: verify OS list
2659

    
2660
    # File verification
2661
    filemap = _ComputeAncillaryFiles(cluster, False)
2662

    
2663
    # do local checksums
2664
    master_node = self.master_node = self.cfg.GetMasterNode()
2665
    master_ip = self.cfg.GetMasterIP()
2666

    
2667
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2668

    
2669
    node_verify_param = {
2670
      constants.NV_FILELIST:
2671
        utils.UniqueSequence(filename
2672
                             for files in filemap
2673
                             for filename in files),
2674
      constants.NV_NODELIST:
2675
        self._SelectSshCheckNodes(node_data_list, self.group_uuid,
2676
                                  self.all_node_info.values()),
2677
      constants.NV_HYPERVISOR: hypervisors,
2678
      constants.NV_HVPARAMS:
2679
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2680
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2681
                                 for node in node_data_list
2682
                                 if not node.offline],
2683
      constants.NV_INSTANCELIST: hypervisors,
2684
      constants.NV_VERSION: None,
2685
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2686
      constants.NV_NODESETUP: None,
2687
      constants.NV_TIME: None,
2688
      constants.NV_MASTERIP: (master_node, master_ip),
2689
      constants.NV_OSLIST: None,
2690
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2691
      }
2692

    
2693
    if vg_name is not None:
2694
      node_verify_param[constants.NV_VGLIST] = None
2695
      node_verify_param[constants.NV_LVLIST] = vg_name
2696
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2697
      node_verify_param[constants.NV_DRBDLIST] = None
2698

    
2699
    if drbd_helper:
2700
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2701

    
2702
    # bridge checks
2703
    # FIXME: this needs to be changed per node-group, not cluster-wide
2704
    bridges = set()
2705
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2706
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2707
      bridges.add(default_nicpp[constants.NIC_LINK])
2708
    for instance in self.my_inst_info.values():
2709
      for nic in instance.nics:
2710
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2711
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2712
          bridges.add(full_nic[constants.NIC_LINK])
2713

    
2714
    if bridges:
2715
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2716

    
2717
    # Build our expected cluster state
2718
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2719
                                                 name=node.name,
2720
                                                 vm_capable=node.vm_capable))
2721
                      for node in node_data_list)
2722

    
2723
    # Gather OOB paths
2724
    oob_paths = []
2725
    for node in self.all_node_info.values():
2726
      path = _SupportsOob(self.cfg, node)
2727
      if path and path not in oob_paths:
2728
        oob_paths.append(path)
2729

    
2730
    if oob_paths:
2731
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2732

    
2733
    for instance in self.my_inst_names:
2734
      inst_config = self.my_inst_info[instance]
2735

    
2736
      for nname in inst_config.all_nodes:
2737
        if nname not in node_image:
2738
          gnode = self.NodeImage(name=nname)
2739
          gnode.ghost = (nname not in self.all_node_info)
2740
          node_image[nname] = gnode
2741

    
2742
      inst_config.MapLVsByNode(node_vol_should)
2743

    
2744
      pnode = inst_config.primary_node
2745
      node_image[pnode].pinst.append(instance)
2746

    
2747
      for snode in inst_config.secondary_nodes:
2748
        nimg = node_image[snode]
2749
        nimg.sinst.append(instance)
2750
        if pnode not in nimg.sbp:
2751
          nimg.sbp[pnode] = []
2752
        nimg.sbp[pnode].append(instance)
2753

    
2754
    # At this point, we have the in-memory data structures complete,
2755
    # except for the runtime information, which we'll gather next
2756

    
2757
    # Due to the way our RPC system works, exact response times cannot be
2758
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2759
    # time before and after executing the request, we can at least have a time
2760
    # window.
2761
    nvinfo_starttime = time.time()
2762
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2763
                                           node_verify_param,
2764
                                           self.cfg.GetClusterName())
2765
    nvinfo_endtime = time.time()
2766

    
2767
    if self.extra_lv_nodes and vg_name is not None:
2768
      extra_lv_nvinfo = \
2769
          self.rpc.call_node_verify(self.extra_lv_nodes,
2770
                                    {constants.NV_LVLIST: vg_name},
2771
                                    self.cfg.GetClusterName())
2772
    else:
2773
      extra_lv_nvinfo = {}
2774

    
2775
    all_drbd_map = self.cfg.ComputeDRBDMap()
2776

    
2777
    feedback_fn("* Gathering disk information (%s nodes)" %
2778
                len(self.my_node_names))
2779
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2780
                                     self.my_inst_info)
2781

    
2782
    feedback_fn("* Verifying configuration file consistency")
2783

    
2784
    # If not all nodes are being checked, we need to make sure the master node
2785
    # and a non-checked vm_capable node are in the list.
2786
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2787
    if absent_nodes:
2788
      vf_nvinfo = all_nvinfo.copy()
2789
      vf_node_info = list(self.my_node_info.values())
2790
      additional_nodes = []
2791
      if master_node not in self.my_node_info:
2792
        additional_nodes.append(master_node)
2793
        vf_node_info.append(self.all_node_info[master_node])
2794
      # Add the first vm_capable node we find which is not included
2795
      for node in absent_nodes:
2796
        nodeinfo = self.all_node_info[node]
2797
        if nodeinfo.vm_capable and not nodeinfo.offline:
2798
          additional_nodes.append(node)
2799
          vf_node_info.append(self.all_node_info[node])
2800
          break
2801
      key = constants.NV_FILELIST
2802
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2803
                                                 {key: node_verify_param[key]},
2804
                                                 self.cfg.GetClusterName()))
2805
    else:
2806
      vf_nvinfo = all_nvinfo
2807
      vf_node_info = self.my_node_info.values()
2808

    
2809
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2810

    
2811
    feedback_fn("* Verifying node status")
2812

    
2813
    refos_img = None
2814

    
2815
    for node_i in node_data_list:
2816
      node = node_i.name
2817
      nimg = node_image[node]
2818

    
2819
      if node_i.offline:
2820
        if verbose:
2821
          feedback_fn("* Skipping offline node %s" % (node,))
2822
        n_offline += 1
2823
        continue
2824

    
2825
      if node == master_node:
2826
        ntype = "master"
2827
      elif node_i.master_candidate:
2828
        ntype = "master candidate"
2829
      elif node_i.drained:
2830
        ntype = "drained"
2831
        n_drained += 1
2832
      else:
2833
        ntype = "regular"
2834
      if verbose:
2835
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2836

    
2837
      msg = all_nvinfo[node].fail_msg
2838
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2839
      if msg:
2840
        nimg.rpc_fail = True
2841
        continue
2842

    
2843
      nresult = all_nvinfo[node].payload
2844

    
2845
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2846
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2847
      self._VerifyNodeNetwork(node_i, nresult)
2848
      self._VerifyOob(node_i, nresult)
2849

    
2850
      if nimg.vm_capable:
2851
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2852
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2853
                             all_drbd_map)
2854

    
2855
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2856
        self._UpdateNodeInstances(node_i, nresult, nimg)
2857
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2858
        self._UpdateNodeOS(node_i, nresult, nimg)
2859

    
2860
        if not nimg.os_fail:
2861
          if refos_img is None:
2862
            refos_img = nimg
2863
          self._VerifyNodeOS(node_i, nimg, refos_img)
2864
        self._VerifyNodeBridges(node_i, nresult, bridges)
2865

    
2866
        # Check whether all running instancies are primary for the node. (This
2867
        # can no longer be done from _VerifyInstance below, since some of the
2868
        # wrong instances could be from other node groups.)
2869
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2870

    
2871
        for inst in non_primary_inst:
2872
          test = inst in self.all_inst_info
2873
          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
2874
                   "instance should not run on node %s", node_i.name)
2875
          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
2876
                   "node is running unknown instance %s", inst)
2877

    
2878
    for node, result in extra_lv_nvinfo.items():
2879
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
2880
                              node_image[node], vg_name)
2881

    
2882
    feedback_fn("* Verifying instance status")
2883
    for instance in self.my_inst_names:
2884
      if verbose:
2885
        feedback_fn("* Verifying instance %s" % instance)
2886
      inst_config = self.my_inst_info[instance]
2887
      self._VerifyInstance(instance, inst_config, node_image,
2888
                           instdisk[instance])
2889
      inst_nodes_offline = []
2890

    
2891
      pnode = inst_config.primary_node
2892
      pnode_img = node_image[pnode]
2893
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2894
               self.ENODERPC, pnode, "instance %s, connection to"
2895
               " primary node failed", instance)
2896

    
2897
      _ErrorIf(inst_config.admin_up and pnode_img.offline,
2898
               self.EINSTANCEBADNODE, instance,
2899
               "instance is marked as running and lives on offline node %s",
2900
               inst_config.primary_node)
2901

    
2902
      # If the instance is non-redundant we cannot survive losing its primary
2903
      # node, so we are not N+1 compliant. On the other hand we have no disk
2904
      # templates with more than one secondary so that situation is not well
2905
      # supported either.
2906
      # FIXME: does not support file-backed instances
2907
      if not inst_config.secondary_nodes:
2908
        i_non_redundant.append(instance)
2909

    
2910
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2911
               instance, "instance has multiple secondary nodes: %s",
2912
               utils.CommaJoin(inst_config.secondary_nodes),
2913
               code=self.ETYPE_WARNING)
2914

    
2915
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2916
        pnode = inst_config.primary_node
2917
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2918
        instance_groups = {}
2919

    
2920
        for node in instance_nodes:
2921
          instance_groups.setdefault(self.all_node_info[node].group,
2922
                                     []).append(node)
2923

    
2924
        pretty_list = [
2925
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2926
          # Sort so that we always list the primary node first.
2927
          for group, nodes in sorted(instance_groups.items(),
2928
                                     key=lambda (_, nodes): pnode in nodes,
2929
                                     reverse=True)]
2930

    
2931
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2932
                      instance, "instance has primary and secondary nodes in"
2933
                      " different groups: %s", utils.CommaJoin(pretty_list),
2934
                      code=self.ETYPE_WARNING)
2935

    
2936
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2937
        i_non_a_balanced.append(instance)
2938

    
2939
      for snode in inst_config.secondary_nodes:
2940
        s_img = node_image[snode]
2941
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2942
                 "instance %s, connection to secondary node failed", instance)
2943

    
2944
        if s_img.offline:
2945
          inst_nodes_offline.append(snode)
2946

    
2947
      # warn that the instance lives on offline nodes
2948
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2949
               "instance has offline secondary node(s) %s",
2950
               utils.CommaJoin(inst_nodes_offline))
2951
      # ... or ghost/non-vm_capable nodes
2952
      for node in inst_config.all_nodes:
2953
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2954
                 "instance lives on ghost node %s", node)
2955
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2956
                 instance, "instance lives on non-vm_capable node %s", node)
2957

    
2958
    feedback_fn("* Verifying orphan volumes")
2959
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2960

    
2961
    # We will get spurious "unknown volume" warnings if any node of this group
2962
    # is secondary for an instance whose primary is in another group. To avoid
2963
    # them, we find these instances and add their volumes to node_vol_should.
2964
    for inst in self.all_inst_info.values():
2965
      for secondary in inst.secondary_nodes:
2966
        if (secondary in self.my_node_info
2967
            and inst.name not in self.my_inst_info):
2968
          inst.MapLVsByNode(node_vol_should)
2969
          break
2970

    
2971
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2972

    
2973
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2974
      feedback_fn("* Verifying N+1 Memory redundancy")
2975
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
2976

    
2977
    feedback_fn("* Other Notes")
2978
    if i_non_redundant:
2979
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2980
                  % len(i_non_redundant))
2981

    
2982
    if i_non_a_balanced:
2983
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2984
                  % len(i_non_a_balanced))
2985

    
2986
    if n_offline:
2987
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2988

    
2989
    if n_drained:
2990
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2991

    
2992
    return not self.bad
2993

    
2994
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2995
    """Analyze the post-hooks' result
2996

2997
    This method analyses the hook result, handles it, and sends some
2998
    nicely-formatted feedback back to the user.
2999

3000
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
3001
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
3002
    @param hooks_results: the results of the multi-node hooks rpc call
3003
    @param feedback_fn: function used send feedback back to the caller
3004
    @param lu_result: previous Exec result
3005
    @return: the new Exec result, based on the previous result
3006
        and hook results
3007

3008
    """
3009
    # We only really run POST phase hooks, only for non-empty groups,
3010
    # and are only interested in their results
3011
    if not self.my_node_names:
3012
      # empty node group
3013
      pass
3014
    elif phase == constants.HOOKS_PHASE_POST:
3015
      # Used to change hooks' output to proper indentation
3016
      feedback_fn("* Hooks Results")
3017
      assert hooks_results, "invalid result from hooks"
3018

    
3019
      for node_name in hooks_results:
3020
        res = hooks_results[node_name]
3021
        msg = res.fail_msg
3022
        test = msg and not res.offline
3023
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
3024
                      "Communication failure in hooks execution: %s", msg)
3025
        if res.offline or msg:
3026
          # No need to investigate payload if node is offline or gave
3027
          # an error.
3028
          continue
3029
        for script, hkr, output in res.payload:
3030
          test = hkr == constants.HKR_FAIL
3031
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
3032
                        "Script %s failed, output:", script)
3033
          if test:
3034
            output = self._HOOKS_INDENT_RE.sub("      ", output)
3035
            feedback_fn("%s" % output)
3036
            lu_result = False
3037

    
3038
    return lu_result
3039

    
3040

    
3041
class LUClusterVerifyDisks(NoHooksLU):
3042
  """Verifies the cluster disks status.
3043

3044
  """
3045
  REQ_BGL = False
3046

    
3047
  def ExpandNames(self):
3048
    self.share_locks = _ShareAll()
3049
    self.needed_locks = {
3050
      locking.LEVEL_NODEGROUP: locking.ALL_SET,
3051
      }
3052

    
3053
  def Exec(self, feedback_fn):
3054
    group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3055

    
3056
    # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3057
    return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3058
                           for group in group_names])
3059

    
3060

    
3061
class LUGroupVerifyDisks(NoHooksLU):
3062
  """Verifies the status of all disks in a node group.
3063

3064
  """
3065
  REQ_BGL = False
3066

    
3067
  def ExpandNames(self):
3068
    # Raises errors.OpPrereqError on its own if group can't be found
3069
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3070

    
3071
    self.share_locks = _ShareAll()
3072
    self.needed_locks = {
3073
      locking.LEVEL_INSTANCE: [],
3074
      locking.LEVEL_NODEGROUP: [],
3075
      locking.LEVEL_NODE: [],
3076
      }
3077

    
3078
  def DeclareLocks(self, level):
3079
    if level == locking.LEVEL_INSTANCE:
3080
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
3081

    
3082
      # Lock instances optimistically, needs verification once node and group
3083
      # locks have been acquired
3084
      self.needed_locks[locking.LEVEL_INSTANCE] = \
3085
        self.cfg.GetNodeGroupInstances(self.group_uuid)
3086

    
3087
    elif level == locking.LEVEL_NODEGROUP:
3088
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3089

    
3090
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
3091
        set([self.group_uuid] +
3092
            # Lock all groups used by instances optimistically; this requires
3093
            # going via the node before it's locked, requiring verification
3094
            # later on
3095
            [group_uuid
3096
             for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3097
             for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3098

    
3099
    elif level == locking.LEVEL_NODE:
3100
      # This will only lock the nodes in the group to be verified which contain
3101
      # actual instances
3102
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3103
      self._LockInstancesNodes()
3104

    
3105
      # Lock all nodes in group to be verified
3106
      assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3107
      member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3108
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3109

    
3110
  def CheckPrereq(self):
3111
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3112
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3113
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3114

    
3115
    assert self.group_uuid in owned_groups
3116

    
3117
    # Check if locked instances are still correct
3118
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3119

    
3120
    # Get instance information
3121
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3122

    
3123
    # Check if node groups for locked instances are still correct
3124
    for (instance_name, inst) in self.instances.items():
3125
      assert owned_nodes.issuperset(inst.all_nodes), \
3126
        "Instance %s's nodes changed while we kept the lock" % instance_name
3127

    
3128
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
3129
                                             owned_groups)
3130

    
3131
      assert self.group_uuid in inst_groups, \
3132
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
3133

    
3134
  def Exec(self, feedback_fn):
3135
    """Verify integrity of cluster disks.
3136

3137
    @rtype: tuple of three items
3138
    @return: a tuple of (dict of node-to-node_error, list of instances
3139
        which need activate-disks, dict of instance: (node, volume) for
3140
        missing volumes
3141

3142
    """
3143
    res_nodes = {}
3144
    res_instances = set()
3145
    res_missing = {}
3146

    
3147
    nv_dict = _MapInstanceDisksToNodes([inst
3148
                                        for inst in self.instances.values()
3149
                                        if inst.admin_up])
3150

    
3151
    if nv_dict:
3152
      nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3153
                             set(self.cfg.GetVmCapableNodeList()))
3154

    
3155
      node_lvs = self.rpc.call_lv_list(nodes, [])
3156

    
3157
      for (node, node_res) in node_lvs.items():
3158
        if node_res.offline:
3159
          continue
3160

    
3161
        msg = node_res.fail_msg
3162
        if msg:
3163
          logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3164
          res_nodes[node] = msg
3165
          continue
3166

    
3167
        for lv_name, (_, _, lv_online) in node_res.payload.items():
3168
          inst = nv_dict.pop((node, lv_name), None)
3169
          if not (lv_online or inst is None):
3170
            res_instances.add(inst)
3171

    
3172
      # any leftover items in nv_dict are missing LVs, let's arrange the data
3173
      # better
3174
      for key, inst in nv_dict.iteritems():
3175
        res_missing.setdefault(inst, []).append(list(key))
3176

    
3177
    return (res_nodes, list(res_instances), res_missing)
3178

    
3179

    
3180
class LUClusterRepairDiskSizes(NoHooksLU):
3181
  """Verifies the cluster disks sizes.
3182

3183
  """
3184
  REQ_BGL = False
3185

    
3186
  def ExpandNames(self):
3187
    if self.op.instances:
3188
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
3189
      self.needed_locks = {
3190
        locking.LEVEL_NODE: [],
3191
        locking.LEVEL_INSTANCE: self.wanted_names,
3192
        }
3193
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3194
    else:
3195
      self.wanted_names = None
3196
      self.needed_locks = {
3197
        locking.LEVEL_NODE: locking.ALL_SET,
3198
        locking.LEVEL_INSTANCE: locking.ALL_SET,
3199
        }
3200
    self.share_locks = {
3201
      locking.LEVEL_NODE: 1,
3202
      locking.LEVEL_INSTANCE: 0,
3203
      }
3204

    
3205
  def DeclareLocks(self, level):
3206
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
3207
      self._LockInstancesNodes(primary_only=True)
3208

    
3209
  def CheckPrereq(self):
3210
    """Check prerequisites.
3211

3212
    This only checks the optional instance list against the existing names.
3213

3214
    """
3215
    if self.wanted_names is None:
3216
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3217

    
3218
    self.wanted_instances = \
3219
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3220

    
3221
  def _EnsureChildSizes(self, disk):
3222
    """Ensure children of the disk have the needed disk size.
3223

3224
    This is valid mainly for DRBD8 and fixes an issue where the
3225
    children have smaller disk size.
3226

3227
    @param disk: an L{ganeti.objects.Disk} object
3228

3229
    """
3230
    if disk.dev_type == constants.LD_DRBD8:
3231
      assert disk.children, "Empty children for DRBD8?"
3232
      fchild = disk.children[0]
3233
      mismatch = fchild.size < disk.size
3234
      if mismatch:
3235
        self.LogInfo("Child disk has size %d, parent %d, fixing",
3236
                     fchild.size, disk.size)
3237
        fchild.size = disk.size
3238

    
3239
      # and we recurse on this child only, not on the metadev
3240
      return self._EnsureChildSizes(fchild) or mismatch
3241
    else:
3242
      return False
3243

    
3244
  def Exec(self, feedback_fn):
3245
    """Verify the size of cluster disks.
3246

3247
    """
3248
    # TODO: check child disks too
3249
    # TODO: check differences in size between primary/secondary nodes
3250
    per_node_disks = {}
3251
    for instance in self.wanted_instances:
3252
      pnode = instance.primary_node
3253
      if pnode not in per_node_disks:
3254
        per_node_disks[pnode] = []
3255
      for idx, disk in enumerate(instance.disks):
3256
        per_node_disks[pnode].append((instance, idx, disk))
3257

    
3258
    changed = []
3259
    for node, dskl in per_node_disks.items():
3260
      newl = [v[2].Copy() for v in dskl]
3261
      for dsk in newl:
3262
        self.cfg.SetDiskID(dsk, node)
3263
      result = self.rpc.call_blockdev_getsize(node, newl)
3264
      if result.fail_msg:
3265
        self.LogWarning("Failure in blockdev_getsize call to node"
3266
                        " %s, ignoring", node)
3267
        continue
3268
      if len(result.payload) != len(dskl):
3269
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
3270
                        " result.payload=%s", node, len(dskl), result.payload)
3271
        self.LogWarning("Invalid result from node %s, ignoring node results",
3272
                        node)
3273
        continue
3274
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
3275
        if size is None:
3276
          self.LogWarning("Disk %d of instance %s did not return size"
3277
                          " information, ignoring", idx, instance.name)
3278
          continue
3279
        if not isinstance(size, (int, long)):
3280
          self.LogWarning("Disk %d of instance %s did not return valid"
3281
                          " size information, ignoring", idx, instance.name)
3282
          continue
3283
        size = size >> 20
3284
        if size != disk.size:
3285
          self.LogInfo("Disk %d of instance %s has mismatched size,"
3286
                       " correcting: recorded %d, actual %d", idx,
3287
                       instance.name, disk.size, size)
3288
          disk.size = size
3289
          self.cfg.Update(instance, feedback_fn)
3290
          changed.append((instance.name, idx, size))
3291
        if self._EnsureChildSizes(disk):
3292
          self.cfg.Update(instance, feedback_fn)
3293
          changed.append((instance.name, idx, disk.size))
3294
    return changed
3295

    
3296

    
3297
class LUClusterRename(LogicalUnit):
3298
  """Rename the cluster.
3299

3300
  """
3301
  HPATH = "cluster-rename"
3302
  HTYPE = constants.HTYPE_CLUSTER
3303

    
3304
  def BuildHooksEnv(self):
3305
    """Build hooks env.
3306

3307
    """
3308
    return {
3309
      "OP_TARGET": self.cfg.GetClusterName(),
3310
      "NEW_NAME": self.op.name,
3311
      }
3312

    
3313
  def BuildHooksNodes(self):
3314
    """Build hooks nodes.
3315

3316
    """
3317
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3318

    
3319
  def CheckPrereq(self):
3320
    """Verify that the passed name is a valid one.
3321

3322
    """
3323
    hostname = netutils.GetHostname(name=self.op.name,
3324
                                    family=self.cfg.GetPrimaryIPFamily())
3325

    
3326
    new_name = hostname.name
3327
    self.ip = new_ip = hostname.ip
3328
    old_name = self.cfg.GetClusterName()
3329
    old_ip = self.cfg.GetMasterIP()
3330
    if new_name == old_name and new_ip == old_ip:
3331
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3332
                                 " cluster has changed",
3333
                                 errors.ECODE_INVAL)
3334
    if new_ip != old_ip:
3335
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3336
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3337
                                   " reachable on the network" %
3338
                                   new_ip, errors.ECODE_NOTUNIQUE)
3339

    
3340
    self.op.name = new_name
3341

    
3342
  def Exec(self, feedback_fn):
3343
    """Rename the cluster.
3344

3345
    """
3346
    clustername = self.op.name
3347
    ip = self.ip
3348

    
3349
    # shutdown the master IP
3350
    master = self.cfg.GetMasterNode()
3351
    result = self.rpc.call_node_stop_master(master, False)
3352
    result.Raise("Could not disable the master role")
3353

    
3354
    try:
3355
      cluster = self.cfg.GetClusterInfo()
3356
      cluster.cluster_name = clustername
3357
      cluster.master_ip = ip
3358
      self.cfg.Update(cluster, feedback_fn)
3359

    
3360
      # update the known hosts file
3361
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3362
      node_list = self.cfg.GetOnlineNodeList()
3363
      try:
3364
        node_list.remove(master)
3365
      except ValueError:
3366
        pass
3367
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3368
    finally:
3369
      result = self.rpc.call_node_start_master(master, False, False)
3370
      msg = result.fail_msg
3371
      if msg:
3372
        self.LogWarning("Could not re-enable the master role on"
3373
                        " the master, please restart manually: %s", msg)
3374

    
3375
    return clustername
3376

    
3377

    
3378
class LUClusterSetParams(LogicalUnit):
3379
  """Change the parameters of the cluster.
3380

3381
  """
3382
  HPATH = "cluster-modify"
3383
  HTYPE = constants.HTYPE_CLUSTER
3384
  REQ_BGL = False
3385

    
3386
  def CheckArguments(self):
3387
    """Check parameters
3388

3389
    """
3390
    if self.op.uid_pool:
3391
      uidpool.CheckUidPool(self.op.uid_pool)
3392

    
3393
    if self.op.add_uids:
3394
      uidpool.CheckUidPool(self.op.add_uids)
3395

    
3396
    if self.op.remove_uids:
3397
      uidpool.CheckUidPool(self.op.remove_uids)
3398

    
3399
  def ExpandNames(self):
3400
    # FIXME: in the future maybe other cluster params won't require checking on
3401
    # all nodes to be modified.
3402
    self.needed_locks = {
3403
      locking.LEVEL_NODE: locking.ALL_SET,
3404
    }
3405
    self.share_locks[locking.LEVEL_NODE] = 1
3406

    
3407
  def BuildHooksEnv(self):
3408
    """Build hooks env.
3409

3410
    """
3411
    return {
3412
      "OP_TARGET": self.cfg.GetClusterName(),
3413
      "NEW_VG_NAME": self.op.vg_name,
3414
      }
3415

    
3416
  def BuildHooksNodes(self):
3417
    """Build hooks nodes.
3418

3419
    """
3420
    mn = self.cfg.GetMasterNode()
3421
    return ([mn], [mn])
3422

    
3423
  def CheckPrereq(self):
3424
    """Check prerequisites.
3425

3426
    This checks whether the given params don't conflict and
3427
    if the given volume group is valid.
3428

3429
    """
3430
    if self.op.vg_name is not None and not self.op.vg_name:
3431
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3432
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3433
                                   " instances exist", errors.ECODE_INVAL)
3434

    
3435
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3436
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3437
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3438
                                   " drbd-based instances exist",
3439
                                   errors.ECODE_INVAL)
3440

    
3441
    node_list = self.owned_locks(locking.LEVEL_NODE)
3442

    
3443
    # if vg_name not None, checks given volume group on all nodes
3444
    if self.op.vg_name:
3445
      vglist = self.rpc.call_vg_list(node_list)
3446
      for node in node_list:
3447
        msg = vglist[node].fail_msg
3448
        if msg:
3449
          # ignoring down node
3450
          self.LogWarning("Error while gathering data on node %s"
3451
                          " (ignoring node): %s", node, msg)
3452
          continue
3453
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3454
                                              self.op.vg_name,
3455
                                              constants.MIN_VG_SIZE)
3456
        if vgstatus:
3457
          raise errors.OpPrereqError("Error on node '%s': %s" %
3458
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3459

    
3460
    if self.op.drbd_helper:
3461
      # checks given drbd helper on all nodes
3462
      helpers = self.rpc.call_drbd_helper(node_list)
3463
      for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3464
        if ninfo.offline:
3465
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3466
          continue
3467
        msg = helpers[node].fail_msg
3468
        if msg:
3469
          raise errors.OpPrereqError("Error checking drbd helper on node"
3470
                                     " '%s': %s" % (node, msg),
3471
                                     errors.ECODE_ENVIRON)
3472
        node_helper = helpers[node].payload
3473
        if node_helper != self.op.drbd_helper:
3474
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3475
                                     (node, node_helper), errors.ECODE_ENVIRON)
3476

    
3477
    self.cluster = cluster = self.cfg.GetClusterInfo()
3478
    # validate params changes
3479
    if self.op.beparams:
3480
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3481
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3482

    
3483
    if self.op.ndparams:
3484
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3485
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3486

    
3487
      # TODO: we need a more general way to handle resetting
3488
      # cluster-level parameters to default values
3489
      if self.new_ndparams["oob_program"] == "":
3490
        self.new_ndparams["oob_program"] = \
3491
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3492

    
3493
    if self.op.nicparams:
3494
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3495
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3496
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3497
      nic_errors = []
3498

    
3499
      # check all instances for consistency
3500
      for instance in self.cfg.GetAllInstancesInfo().values():
3501
        for nic_idx, nic in enumerate(instance.nics):
3502
          params_copy = copy.deepcopy(nic.nicparams)
3503
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3504

    
3505
          # check parameter syntax
3506
          try:
3507
            objects.NIC.CheckParameterSyntax(params_filled)
3508
          except errors.ConfigurationError, err:
3509
            nic_errors.append("Instance %s, nic/%d: %s" %
3510
                              (instance.name, nic_idx, err))
3511

    
3512
          # if we're moving instances to routed, check that they have an ip
3513
          target_mode = params_filled[constants.NIC_MODE]
3514
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3515
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3516
                              " address" % (instance.name, nic_idx))
3517
      if nic_errors:
3518
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3519
                                   "\n".join(nic_errors))
3520

    
3521
    # hypervisor list/parameters
3522
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3523
    if self.op.hvparams:
3524
      for hv_name, hv_dict in self.op.hvparams.items():
3525
        if hv_name not in self.new_hvparams:
3526
          self.new_hvparams[hv_name] = hv_dict
3527
        else:
3528
          self.new_hvparams[hv_name].update(hv_dict)
3529

    
3530
    # os hypervisor parameters
3531
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3532
    if self.op.os_hvp:
3533
      for os_name, hvs in self.op.os_hvp.items():
3534
        if os_name not in self.new_os_hvp:
3535
          self.new_os_hvp[os_name] = hvs
3536
        else:
3537
          for hv_name, hv_dict in hvs.items():
3538
            if hv_name not in self.new_os_hvp[os_name]:
3539
              self.new_os_hvp[os_name][hv_name] = hv_dict
3540
            else:
3541
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3542

    
3543
    # os parameters
3544
    self.new_osp = objects.FillDict(cluster.osparams, {})
3545
    if self.op.osparams:
3546
      for os_name, osp in self.op.osparams.items():
3547
        if os_name not in self.new_osp:
3548
          self.new_osp[os_name] = {}
3549

    
3550
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3551
                                                  use_none=True)
3552

    
3553
        if not self.new_osp[os_name]:
3554
          # we removed all parameters
3555
          del self.new_osp[os_name]
3556
        else:
3557
          # check the parameter validity (remote check)
3558
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3559
                         os_name, self.new_osp[os_name])
3560

    
3561
    # changes to the hypervisor list
3562
    if self.op.enabled_hypervisors is not None:
3563
      self.hv_list = self.op.enabled_hypervisors
3564
      for hv in self.hv_list:
3565
        # if the hypervisor doesn't already exist in the cluster
3566
        # hvparams, we initialize it to empty, and then (in both
3567
        # cases) we make sure to fill the defaults, as we might not
3568
        # have a complete defaults list if the hypervisor wasn't
3569
        # enabled before
3570
        if hv not in new_hvp:
3571
          new_hvp[hv] = {}
3572
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3573
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3574
    else:
3575
      self.hv_list = cluster.enabled_hypervisors
3576

    
3577
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3578
      # either the enabled list has changed, or the parameters have, validate
3579
      for hv_name, hv_params in self.new_hvparams.items():
3580
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3581
            (self.op.enabled_hypervisors and
3582
             hv_name in self.op.enabled_hypervisors)):
3583
          # either this is a new hypervisor, or its parameters have changed
3584
          hv_class = hypervisor.GetHypervisor(hv_name)
3585
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3586
          hv_class.CheckParameterSyntax(hv_params)
3587
          _CheckHVParams(self, node_list, hv_name, hv_params)
3588

    
3589
    if self.op.os_hvp:
3590
      # no need to check any newly-enabled hypervisors, since the
3591
      # defaults have already been checked in the above code-block
3592
      for os_name, os_hvp in self.new_os_hvp.items():
3593
        for hv_name, hv_params in os_hvp.items():
3594
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3595
          # we need to fill in the new os_hvp on top of the actual hv_p
3596
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3597
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3598
          hv_class = hypervisor.GetHypervisor(hv_name)
3599
          hv_class.CheckParameterSyntax(new_osp)
3600
          _CheckHVParams(self, node_list, hv_name, new_osp)
3601

    
3602
    if self.op.default_iallocator:
3603
      alloc_script = utils.FindFile(self.op.default_iallocator,
3604
                                    constants.IALLOCATOR_SEARCH_PATH,
3605
                                    os.path.isfile)
3606
      if alloc_script is None:
3607
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3608
                                   " specified" % self.op.default_iallocator,
3609
                                   errors.ECODE_INVAL)
3610

    
3611
  def Exec(self, feedback_fn):
3612
    """Change the parameters of the cluster.
3613

3614
    """
3615
    if self.op.vg_name is not None:
3616
      new_volume = self.op.vg_name
3617
      if not new_volume:
3618
        new_volume = None
3619
      if new_volume != self.cfg.GetVGName():
3620
        self.cfg.SetVGName(new_volume)
3621
      else:
3622
        feedback_fn("Cluster LVM configuration already in desired"
3623
                    " state, not changing")
3624
    if self.op.drbd_helper is not None:
3625
      new_helper = self.op.drbd_helper
3626
      if not new_helper:
3627
        new_helper = None
3628
      if new_helper != self.cfg.GetDRBDHelper():
3629
        self.cfg.SetDRBDHelper(new_helper)
3630
      else:
3631
        feedback_fn("Cluster DRBD helper already in desired state,"
3632
                    " not changing")
3633
    if self.op.hvparams:
3634
      self.cluster.hvparams = self.new_hvparams
3635
    if self.op.os_hvp:
3636
      self.cluster.os_hvp = self.new_os_hvp
3637
    if self.op.enabled_hypervisors is not None:
3638
      self.cluster.hvparams = self.new_hvparams
3639
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3640
    if self.op.beparams:
3641
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3642
    if self.op.nicparams:
3643
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3644
    if self.op.osparams:
3645
      self.cluster.osparams = self.new_osp
3646
    if self.op.ndparams:
3647
      self.cluster.ndparams = self.new_ndparams
3648

    
3649
    if self.op.candidate_pool_size is not None:
3650
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3651
      # we need to update the pool size here, otherwise the save will fail
3652
      _AdjustCandidatePool(self, [])
3653

    
3654
    if self.op.maintain_node_health is not None:
3655
      self.cluster.maintain_node_health = self.op.maintain_node_health
3656

    
3657
    if self.op.prealloc_wipe_disks is not None:
3658
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3659

    
3660
    if self.op.add_uids is not None:
3661
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3662

    
3663
    if self.op.remove_uids is not None:
3664
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3665

    
3666
    if self.op.uid_pool is not None:
3667
      self.cluster.uid_pool = self.op.uid_pool
3668

    
3669
    if self.op.default_iallocator is not None:
3670
      self.cluster.default_iallocator = self.op.default_iallocator
3671

    
3672
    if self.op.reserved_lvs is not None:
3673
      self.cluster.reserved_lvs = self.op.reserved_lvs
3674

    
3675
    def helper_os(aname, mods, desc):
3676
      desc += " OS list"
3677
      lst = getattr(self.cluster, aname)
3678
      for key, val in mods:
3679
        if key == constants.DDM_ADD:
3680
          if val in lst:
3681
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3682
          else:
3683
            lst.append(val)
3684
        elif key == constants.DDM_REMOVE:
3685
          if val in lst:
3686
            lst.remove(val)
3687
          else:
3688
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3689
        else:
3690
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3691

    
3692
    if self.op.hidden_os:
3693
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3694

    
3695
    if self.op.blacklisted_os:
3696
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3697

    
3698
    if self.op.master_netdev:
3699
      master = self.cfg.GetMasterNode()
3700
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3701
                  self.cluster.master_netdev)
3702
      result = self.rpc.call_node_stop_master(master, False)
3703
      result.Raise("Could not disable the master ip")
3704
      feedback_fn("Changing master_netdev from %s to %s" %
3705
                  (self.cluster.master_netdev, self.op.master_netdev))
3706
      self.cluster.master_netdev = self.op.master_netdev
3707

    
3708
    self.cfg.Update(self.cluster, feedback_fn)
3709

    
3710
    if self.op.master_netdev:
3711
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3712
                  self.op.master_netdev)
3713
      result = self.rpc.call_node_start_master(master, False, False)
3714
      if result.fail_msg:
3715
        self.LogWarning("Could not re-enable the master ip on"
3716
                        " the master, please restart manually: %s",
3717
                        result.fail_msg)
3718

    
3719

    
3720
def _UploadHelper(lu, nodes, fname):
3721
  """Helper for uploading a file and showing warnings.
3722

3723
  """
3724
  if os.path.exists(fname):
3725
    result = lu.rpc.call_upload_file(nodes, fname)
3726
    for to_node, to_result in result.items():
3727
      msg = to_result.fail_msg
3728
      if msg:
3729
        msg = ("Copy of file %s to node %s failed: %s" %
3730
               (fname, to_node, msg))
3731
        lu.proc.LogWarning(msg)
3732

    
3733

    
3734
def _ComputeAncillaryFiles(cluster, redist):
3735
  """Compute files external to Ganeti which need to be consistent.
3736

3737
  @type redist: boolean
3738
  @param redist: Whether to include files which need to be redistributed
3739

3740
  """
3741
  # Compute files for all nodes
3742
  files_all = set([
3743
    constants.SSH_KNOWN_HOSTS_FILE,
3744
    constants.CONFD_HMAC_KEY,
3745
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3746
    ])
3747

    
3748
  if not redist:
3749
    files_all.update(constants.ALL_CERT_FILES)
3750
    files_all.update(ssconf.SimpleStore().GetFileList())
3751
  else:
3752
    # we need to ship at least the RAPI certificate
3753
    files_all.add(constants.RAPI_CERT_FILE)
3754

    
3755
  if cluster.modify_etc_hosts:
3756
    files_all.add(constants.ETC_HOSTS)
3757

    
3758
  # Files which must either exist on all nodes or on none
3759
  files_all_opt = set([
3760
    constants.RAPI_USERS_FILE,
3761
    ])
3762

    
3763
  # Files which should only be on master candidates
3764
  files_mc = set()
3765
  if not redist:
3766
    files_mc.add(constants.CLUSTER_CONF_FILE)
3767

    
3768
  # Files which should only be on VM-capable nodes
3769
  files_vm = set(filename
3770
    for hv_name in cluster.enabled_hypervisors
3771
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
3772

    
3773
  # Filenames must be unique
3774
  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
3775
          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
3776
         "Found file listed in more than one file list"
3777

    
3778
  return (files_all, files_all_opt, files_mc, files_vm)
3779

    
3780

    
3781
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3782
  """Distribute additional files which are part of the cluster configuration.
3783

3784
  ConfigWriter takes care of distributing the config and ssconf files, but
3785
  there are more files which should be distributed to all nodes. This function
3786
  makes sure those are copied.
3787

3788
  @param lu: calling logical unit
3789
  @param additional_nodes: list of nodes not in the config to distribute to
3790
  @type additional_vm: boolean
3791
  @param additional_vm: whether the additional nodes are vm-capable or not
3792

3793
  """
3794
  # Gather target nodes
3795
  cluster = lu.cfg.GetClusterInfo()
3796
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3797

    
3798
  online_nodes = lu.cfg.GetOnlineNodeList()
3799
  vm_nodes = lu.cfg.GetVmCapableNodeList()
3800

    
3801
  if additional_nodes is not None:
3802
    online_nodes.extend(additional_nodes)
3803
    if additional_vm:
3804
      vm_nodes.extend(additional_nodes)
3805

    
3806
  # Never distribute to master node
3807
  for nodelist in [online_nodes, vm_nodes]:
3808
    if master_info.name in nodelist:
3809
      nodelist.remove(master_info.name)
3810

    
3811
  # Gather file lists
3812
  (files_all, files_all_opt, files_mc, files_vm) = \
3813
    _ComputeAncillaryFiles(cluster, True)
3814

    
3815
  # Never re-distribute configuration file from here
3816
  assert not (constants.CLUSTER_CONF_FILE in files_all or
3817
              constants.CLUSTER_CONF_FILE in files_vm)
3818
  assert not files_mc, "Master candidates not handled in this function"
3819

    
3820
  filemap = [
3821
    (online_nodes, files_all),
3822
    (online_nodes, files_all_opt),
3823
    (vm_nodes, files_vm),
3824
    ]
3825

    
3826
  # Upload the files
3827
  for (node_list, files) in filemap:
3828
    for fname in files:
3829
      _UploadHelper(lu, node_list, fname)
3830

    
3831

    
3832
class LUClusterRedistConf(NoHooksLU):
3833
  """Force the redistribution of cluster configuration.
3834

3835
  This is a very simple LU.
3836

3837
  """
3838
  REQ_BGL = False
3839

    
3840
  def ExpandNames(self):
3841
    self.needed_locks = {
3842
      locking.LEVEL_NODE: locking.ALL_SET,
3843
    }
3844
    self.share_locks[locking.LEVEL_NODE] = 1
3845

    
3846
  def Exec(self, feedback_fn):
3847
    """Redistribute the configuration.
3848

3849
    """
3850
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3851
    _RedistributeAncillaryFiles(self)
3852

    
3853

    
3854
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3855
  """Sleep and poll for an instance's disk to sync.
3856

3857
  """
3858
  if not instance.disks or disks is not None and not disks:
3859
    return True
3860

    
3861
  disks = _ExpandCheckDisks(instance, disks)
3862

    
3863
  if not oneshot:
3864
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3865

    
3866
  node = instance.primary_node
3867

    
3868
  for dev in disks:
3869
    lu.cfg.SetDiskID(dev, node)
3870

    
3871
  # TODO: Convert to utils.Retry
3872

    
3873
  retries = 0
3874
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3875
  while True:
3876
    max_time = 0
3877
    done = True
3878
    cumul_degraded = False
3879
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3880
    msg = rstats.fail_msg
3881
    if msg:
3882
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3883
      retries += 1
3884
      if retries >= 10:
3885
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3886
                                 " aborting." % node)
3887
      time.sleep(6)
3888
      continue
3889
    rstats = rstats.payload
3890
    retries = 0
3891
    for i, mstat in enumerate(rstats):
3892
      if mstat is None:
3893
        lu.LogWarning("Can't compute data for node %s/%s",
3894
                           node, disks[i].iv_name)
3895
        continue
3896

    
3897
      cumul_degraded = (cumul_degraded or
3898
                        (mstat.is_degraded and mstat.sync_percent is None))
3899
      if mstat.sync_percent is not None:
3900
        done = False
3901
        if mstat.estimated_time is not None:
3902
          rem_time = ("%s remaining (estimated)" %
3903
                      utils.FormatSeconds(mstat.estimated_time))
3904
          max_time = mstat.estimated_time
3905
        else:
3906
          rem_time = "no time estimate"
3907
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3908
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3909

    
3910
    # if we're done but degraded, let's do a few small retries, to
3911
    # make sure we see a stable and not transient situation; therefore
3912
    # we force restart of the loop
3913
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3914
      logging.info("Degraded disks found, %d retries left", degr_retries)
3915
      degr_retries -= 1
3916
      time.sleep(1)
3917
      continue
3918

    
3919
    if done or oneshot:
3920
      break
3921

    
3922
    time.sleep(min(60, max_time))
3923

    
3924
  if done:
3925
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3926
  return not cumul_degraded
3927

    
3928

    
3929
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3930
  """Check that mirrors are not degraded.
3931

3932
  The ldisk parameter, if True, will change the test from the
3933
  is_degraded attribute (which represents overall non-ok status for
3934
  the device(s)) to the ldisk (representing the local storage status).
3935

3936
  """
3937
  lu.cfg.SetDiskID(dev, node)
3938

    
3939
  result = True
3940

    
3941
  if on_primary or dev.AssembleOnSecondary():
3942
    rstats = lu.rpc.call_blockdev_find(node, dev)
3943
    msg = rstats.fail_msg
3944
    if msg:
3945
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3946
      result = False
3947
    elif not rstats.payload:
3948
      lu.LogWarning("Can't find disk on node %s", node)
3949
      result = False
3950
    else:
3951
      if ldisk:
3952
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3953
      else:
3954
        result = result and not rstats.payload.is_degraded
3955

    
3956
  if dev.children:
3957
    for child in dev.children:
3958
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3959

    
3960
  return result
3961

    
3962

    
3963
class LUOobCommand(NoHooksLU):
3964
  """Logical unit for OOB handling.
3965

3966
  """
3967
  REG_BGL = False
3968
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
3969

    
3970
  def ExpandNames(self):
3971
    """Gather locks we need.
3972

3973
    """
3974
    if self.op.node_names:
3975
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
3976
      lock_names = self.op.node_names
3977
    else:
3978
      lock_names = locking.ALL_SET
3979

    
3980
    self.needed_locks = {
3981
      locking.LEVEL_NODE: lock_names,
3982
      }
3983

    
3984
  def CheckPrereq(self):
3985
    """Check prerequisites.
3986

3987
    This checks:
3988
     - the node exists in the configuration
3989
     - OOB is supported
3990

3991
    Any errors are signaled by raising errors.OpPrereqError.
3992

3993
    """
3994
    self.nodes = []
3995
    self.master_node = self.cfg.GetMasterNode()
3996

    
3997
    assert self.op.power_delay >= 0.0
3998

    
3999
    if self.op.node_names:
4000
      if (self.op.command in self._SKIP_MASTER and
4001
          self.master_node in self.op.node_names):
4002
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
4003
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
4004

    
4005
        if master_oob_handler:
4006
          additional_text = ("run '%s %s %s' if you want to operate on the"
4007
                             " master regardless") % (master_oob_handler,
4008
                                                      self.op.command,
4009
                                                      self.master_node)
4010
        else:
4011
          additional_text = "it does not support out-of-band operations"
4012

    
4013
        raise errors.OpPrereqError(("Operating on the master node %s is not"
4014
                                    " allowed for %s; %s") %
4015
                                   (self.master_node, self.op.command,
4016
                                    additional_text), errors.ECODE_INVAL)
4017
    else:
4018
      self.op.node_names = self.cfg.GetNodeList()
4019
      if self.op.command in self._SKIP_MASTER:
4020
        self.op.node_names.remove(self.master_node)
4021

    
4022
    if self.op.command in self._SKIP_MASTER:
4023
      assert self.master_node not in self.op.node_names
4024

    
4025
    for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
4026
      if node is None:
4027
        raise errors.OpPrereqError("Node %s not found" % node_name,
4028
                                   errors.ECODE_NOENT)
4029
      else:
4030
        self.nodes.append(node)
4031

    
4032
      if (not self.op.ignore_status and
4033
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
4034
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
4035
                                    " not marked offline") % node_name,
4036
                                   errors.ECODE_STATE)
4037

    
4038
  def Exec(self, feedback_fn):
4039
    """Execute OOB and return result if we expect any.
4040

4041
    """
4042
    master_node = self.master_node
4043
    ret = []
4044

    
4045
    for idx, node in enumerate(utils.NiceSort(self.nodes,
4046
                                              key=lambda node: node.name)):
4047
      node_entry = [(constants.RS_NORMAL, node.name)]
4048
      ret.append(node_entry)
4049

    
4050
      oob_program = _SupportsOob(self.cfg, node)
4051

    
4052
      if not oob_program:
4053
        node_entry.append((constants.RS_UNAVAIL, None))
4054
        continue
4055

    
4056
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
4057
                   self.op.command, oob_program, node.name)
4058
      result = self.rpc.call_run_oob(master_node, oob_program,
4059
                                     self.op.command, node.name,
4060
                                     self.op.timeout)
4061

    
4062
      if result.fail_msg:
4063
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4064
                        node.name, result.fail_msg)
4065
        node_entry.append((constants.RS_NODATA, None))
4066
      else:
4067
        try:
4068
          self._CheckPayload(result)
4069
        except errors.OpExecError, err:
4070
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
4071
                          node.name, err)