Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 6a1434d7

History | View | Annotate | Download (476.6 kB)

1
#
2
#
3

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

    
21

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

    
24
# pylint: disable=W0201,C0302
25

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

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

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

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

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

    
65

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

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

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

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

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

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

    
87

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

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

101
  Note that all commands require root permissions.
102

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

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

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

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

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

    
145
    # Tasklets
146
    self.tasklets = None
147

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

    
151
    self.CheckArguments()
152

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

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

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

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

168
    """
169
    pass
170

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

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

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

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

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

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

196
    Examples::
197

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

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

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

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

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

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

235
    """
236

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

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

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

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

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

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

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

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

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

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

287
    """
288
    raise NotImplementedError
289

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

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

301
    """
302
    raise NotImplementedError
303

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
390
    del self.recalculate_locks[locking.LEVEL_NODE]
391

    
392

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

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

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

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

406
    This just raises an error.
407

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

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

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

    
417

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

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

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

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

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

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

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

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

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

450
    """
451
    pass
452

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

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

460
    """
461
    raise NotImplementedError
462

    
463

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

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

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

474
    """
475
    self.use_locking = use_locking
476

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

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

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

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

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

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

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

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

    
511
    # Return expanded names
512
    return self.wanted
513

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

517
    See L{LogicalUnit.ExpandNames}.
518

519
    """
520
    raise NotImplementedError()
521

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

525
    See L{LogicalUnit.DeclareLocks}.
526

527
    """
528
    raise NotImplementedError()
529

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

533
    @return: Query data object
534

535
    """
536
    raise NotImplementedError()
537

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

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

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

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

    
552

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

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

    
559

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

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

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

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

    
583
  return inst_groups
584

    
585

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

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

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

    
607
  return wanted_instances
608

    
609

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

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

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

    
622

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

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

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

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

    
640

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

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

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

    
660

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

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

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

    
693

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

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

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

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

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

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

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

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

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

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

    
738

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

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

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

    
751

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

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

    
763

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

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

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

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

    
782

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

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

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

    
797

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

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

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

    
812

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

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

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

    
825

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

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

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

    
838

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

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

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

    
856

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

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

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

    
883

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

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

    
891

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

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

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

    
907

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

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

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

    
924

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

    
929

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

    
934

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

940
  This builds the hook environment from individual variables.
941

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

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

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

    
1006
  env["INSTANCE_NIC_COUNT"] = nic_count
1007

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

    
1016
  env["INSTANCE_DISK_COUNT"] = disk_count
1017

    
1018
  if not tags:
1019
    tags = []
1020

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

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

    
1027
  return env
1028

    
1029

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

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

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

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

    
1053

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

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

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

    
1092

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

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

    
1108

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

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

    
1119

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

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

    
1133

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

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

    
1142

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

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

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

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

    
1166

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

    
1170

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

1174
  """
1175

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

    
1178

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

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

    
1186

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

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

    
1194

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

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

    
1204
  return []
1205

    
1206

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

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

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

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

    
1221
  return faulty
1222

    
1223

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

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

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

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

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

    
1255

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

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

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

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

    
1276
  return iallocator
1277

    
1278

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

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

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

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

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

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

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

1303
    """
1304
    return True
1305

    
1306

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

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

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

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

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

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

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

1331
    This checks whether the cluster is empty.
1332

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

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

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

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

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

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

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

    
1361
    return master
1362

    
1363

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

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

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

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

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

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

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

    
1396

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

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

1408
  """
1409
  hvp_data = []
1410

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

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

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

    
1426
  return hvp_data
1427

    
1428

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

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

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

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

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

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

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

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

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

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

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

    
1512

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

1516
  """
1517
  REQ_BGL = False
1518

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

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

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

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

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

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

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

    
1551
    return ResultWithJobs(jobs)
1552

    
1553

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

1557
  """
1558
  REQ_BGL = True
1559

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

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

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

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

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

    
1590
    feedback_fn("* Verifying cluster config")
1591

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

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

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

    
1601
    feedback_fn("* Verifying hypervisor parameters")
1602

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

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

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

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

    
1615
    dangling_instances = {}
1616
    no_node_instances = []
1617

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

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

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

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

    
1639
    return not self.bad
1640

    
1641

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

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

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

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

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

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

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

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

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

    
1715
    self.share_locks = _ShareAll()
1716

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

    
1722
      all_inst_info = self.cfg.GetAllInstancesInfo()
1723

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
1859
    return True
1860

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2133
      nresult = all_nvinfo[node.name]
2134

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2272
    nimg.os_fail = test
2273

    
2274
    if test:
2275
      return
2276

    
2277
    os_dict = {}
2278

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

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

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

    
2291
    nimg.oslist = os_dict
2292

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2483
      node_disks[nname] = disks
2484

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

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

    
2492
      node_disks_devonly[nname] = devonly
2493

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

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

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

    
2502
    instdisk = {}
2503

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

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

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

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

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

    
2543
    return instdisk
2544

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

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

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

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

    
2559
    return env
2560

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

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

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

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

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

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

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

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

    
2597
    # FIXME: verify OS list
2598

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2692
      inst_config.MapLVsByNode(node_vol_should)
2693

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

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

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

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

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

    
2725
    all_drbd_map = self.cfg.ComputeDRBDMap()
2726

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

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

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

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

    
2761
    feedback_fn("* Verifying node status")
2762

    
2763
    refos_img = None
2764

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

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

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

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

    
2793
      nresult = all_nvinfo[node].payload
2794

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2942
    return not self.bad
2943

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

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

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

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

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

    
2990
    return lu_result
2991

    
2992

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

2996
  """
2997
  REQ_BGL = False
2998

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

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

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

    
3012

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

3016
  """
3017
  REQ_BGL = False
3018

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

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

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

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

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

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

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

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

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

    
3067
    assert self.group_uuid in owned_groups
3068

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
3131

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

3135
  """
3136
  REQ_BGL = False
3137

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
3245

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

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

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

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

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

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

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

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

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

    
3289
    self.op.name = new_name
3290

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

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

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

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

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

    
3324
    return clustername
3325

    
3326

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
3668

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

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

    
3682

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

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

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

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

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

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

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

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

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

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

    
3726

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

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

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

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

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

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

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

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

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

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

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

    
3777

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

3781
  This is a very simple LU.
3782

3783
  """
3784
  REQ_BGL = False
3785

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

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

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

    
3799

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

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

    
3807
  disks = _ExpandCheckDisks(instance, disks)
3808

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

    
3812
  node = instance.primary_node
3813

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

    
3817
  # TODO: Convert to utils.Retry
3818

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

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

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

    
3865
    if done or oneshot:
3866
      break
3867

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

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

    
3874

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

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

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

    
3885
  result = True
3886

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

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

    
3906
  return result
3907

    
3908

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

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

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

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

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

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

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

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

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

    
3943
    assert self.op.power_delay >= 0.0
3944

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
4050
    return ret
4051

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

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

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

    
4070