Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 69ab2e12

History | View | Annotate | Download (474.4 kB)

1
#
2
#
3

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

    
21

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

    
24
# pylint: disable=W0201,C0302
25

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

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

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

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

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

    
65

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

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

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

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

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

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

    
87

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

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

101
  Note that all commands require root permissions.
102

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

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

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

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

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

    
145
    # Tasklets
146
    self.tasklets = None
147

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

    
151
    self.CheckArguments()
152

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

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

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

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

168
    """
169
    pass
170

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

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

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

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

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

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

196
    Examples::
197

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

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

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

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

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

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

235
    """
236

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

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

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

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

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

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

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

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

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

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

287
    """
288
    raise NotImplementedError
289

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

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

301
    """
302
    raise NotImplementedError
303

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
390
    del self.recalculate_locks[locking.LEVEL_NODE]
391

    
392

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

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

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

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

406
    This just raises an error.
407

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

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

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

    
417

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

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

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

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

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

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

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

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

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

450
    """
451
    pass
452

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

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

460
    """
461
    raise NotImplementedError
462

    
463

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

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

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

474
    """
475
    self.use_locking = use_locking
476

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

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

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

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

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

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

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

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

    
511
    # Return expanded names
512
    return self.wanted
513

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

517
    See L{LogicalUnit.ExpandNames}.
518

519
    """
520
    raise NotImplementedError()
521

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

525
    See L{LogicalUnit.DeclareLocks}.
526

527
    """
528
    raise NotImplementedError()
529

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

533
    @return: Query data object
534

535
    """
536
    raise NotImplementedError()
537

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

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

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

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

    
552

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

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

    
559

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

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

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

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

    
583
  return inst_groups
584

    
585

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

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

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

    
607
  return wanted_instances
608

    
609

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

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

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

    
622

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

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

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

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

    
640

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

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

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

    
660

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

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

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

    
693

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

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

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

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

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

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

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

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

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

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

    
738

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

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

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

    
751

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

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

    
763

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

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

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

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

    
782

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

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

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

    
797

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

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

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

    
812

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

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

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

    
825

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

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

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

    
838

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

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

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

    
856

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

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

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

    
883

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

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

    
891

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

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

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

    
907

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

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

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

    
924

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

    
929

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

    
934

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

940
  This builds the hook environment from individual variables.
941

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

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

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

    
1006
  env["INSTANCE_NIC_COUNT"] = nic_count
1007

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

    
1016
  env["INSTANCE_DISK_COUNT"] = disk_count
1017

    
1018
  if not tags:
1019
    tags = []
1020

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

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

    
1027
  return env
1028

    
1029

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

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

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

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

    
1053

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

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

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

    
1092

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

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

    
1108

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

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

    
1119

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

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

    
1133

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

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

    
1142

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

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

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

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

    
1166

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

    
1170

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

1174
  """
1175

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

    
1178

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

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

    
1186

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

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

    
1194

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

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

    
1204
  return []
1205

    
1206

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

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

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

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

    
1221
  return faulty
1222

    
1223

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

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

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

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

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

    
1255

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

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

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

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

    
1276
  return iallocator
1277

    
1278

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

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

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

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

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

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

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

1303
    """
1304
    return True
1305

    
1306

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

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

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

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

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

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

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

1331
    This checks whether the cluster is empty.
1332

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

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

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

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

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

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

    
1358
    result = self.rpc.call_node_deactivate_master_ip(master)
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
    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
2112
            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
2113
           "Found file listed in more than one file list"
2114

    
2115
    # Define functions determining which nodes to consider for a file
2116
    files2nodefn = [
2117
      (files_all, None),
2118
      (files_all_opt, None),
2119
      (files_mc, lambda node: (node.master_candidate or
2120
                               node.name == master_node)),
2121
      (files_vm, lambda node: node.vm_capable),
2122
      ]
2123

    
2124
    # Build mapping from filename to list of nodes which should have the file
2125
    nodefiles = {}
2126
    for (files, fn) in files2nodefn:
2127
      if fn is None:
2128
        filenodes = nodeinfo
2129
      else:
2130
        filenodes = filter(fn, nodeinfo)
2131
      nodefiles.update((filename,
2132
                        frozenset(map(operator.attrgetter("name"), filenodes)))
2133
                       for filename in files)
2134

    
2135
    assert set(nodefiles) == (files_all | files_all_opt | files_mc | files_vm)
2136

    
2137
    fileinfo = dict((filename, {}) for filename in nodefiles)
2138
    ignore_nodes = set()
2139

    
2140
    for node in nodeinfo:
2141
      if node.offline:
2142
        ignore_nodes.add(node.name)
2143
        continue
2144

    
2145
      nresult = all_nvinfo[node.name]
2146

    
2147
      if nresult.fail_msg or not nresult.payload:
2148
        node_files = None
2149
      else:
2150
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
2151

    
2152
      test = not (node_files and isinstance(node_files, dict))
2153
      errorif(test, cls.ENODEFILECHECK, node.name,
2154
              "Node did not return file checksum data")
2155
      if test:
2156
        ignore_nodes.add(node.name)
2157
        continue
2158

    
2159
      # Build per-checksum mapping from filename to nodes having it
2160
      for (filename, checksum) in node_files.items():
2161
        assert filename in nodefiles
2162
        fileinfo[filename].setdefault(checksum, set()).add(node.name)
2163

    
2164
    for (filename, checksums) in fileinfo.items():
2165
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2166

    
2167
      # Nodes having the file
2168
      with_file = frozenset(node_name
2169
                            for nodes in fileinfo[filename].values()
2170
                            for node_name in nodes) - ignore_nodes
2171

    
2172
      expected_nodes = nodefiles[filename] - ignore_nodes
2173

    
2174
      # Nodes missing file
2175
      missing_file = expected_nodes - with_file
2176

    
2177
      if filename in files_all_opt:
2178
        # All or no nodes
2179
        errorif(missing_file and missing_file != expected_nodes,
2180
                cls.ECLUSTERFILECHECK, None,
2181
                "File %s is optional, but it must exist on all or no"
2182
                " nodes (not found on %s)",
2183
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2184
      else:
2185
        # Non-optional files
2186
        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
2187
                "File %s is missing from node(s) %s", filename,
2188
                utils.CommaJoin(utils.NiceSort(missing_file)))
2189

    
2190
        # Warn if a node has a file it shouldn't
2191
        unexpected = with_file - expected_nodes
2192
        errorif(unexpected,
2193
                cls.ECLUSTERFILECHECK, None,
2194
                "File %s should not exist on node(s) %s",
2195
                filename, utils.CommaJoin(utils.NiceSort(unexpected)))
2196

    
2197
      # See if there are multiple versions of the file
2198
      test = len(checksums) > 1
2199
      if test:
2200
        variants = ["variant %s on %s" %
2201
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2202
                    for (idx, (checksum, nodes)) in
2203
                      enumerate(sorted(checksums.items()))]
2204
      else:
2205
        variants = []
2206

    
2207
      errorif(test, cls.ECLUSTERFILECHECK, None,
2208
              "File %s found with %s different checksums (%s)",
2209
              filename, len(checksums), "; ".join(variants))
2210

    
2211
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2212
                      drbd_map):
2213
    """Verifies and the node DRBD status.
2214

2215
    @type ninfo: L{objects.Node}
2216
    @param ninfo: the node to check
2217
    @param nresult: the remote results for the node
2218
    @param instanceinfo: the dict of instances
2219
    @param drbd_helper: the configured DRBD usermode helper
2220
    @param drbd_map: the DRBD map as returned by
2221
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2222

2223
    """
2224
    node = ninfo.name
2225
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2226

    
2227
    if drbd_helper:
2228
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2229
      test = (helper_result == None)
2230
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
2231
               "no drbd usermode helper returned")
2232
      if helper_result:
2233
        status, payload = helper_result
2234
        test = not status
2235
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2236
                 "drbd usermode helper check unsuccessful: %s", payload)
2237
        test = status and (payload != drbd_helper)
2238
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2239
                 "wrong drbd usermode helper: %s", payload)
2240

    
2241
    # compute the DRBD minors
2242
    node_drbd = {}
2243
    for minor, instance in drbd_map[node].items():
2244
      test = instance not in instanceinfo
2245
      _ErrorIf(test, self.ECLUSTERCFG, None,
2246
               "ghost instance '%s' in temporary DRBD map", instance)
2247
        # ghost instance should not be running, but otherwise we
2248
        # don't give double warnings (both ghost instance and
2249
        # unallocated minor in use)
2250
      if test:
2251
        node_drbd[minor] = (instance, False)
2252
      else:
2253
        instance = instanceinfo[instance]
2254
        node_drbd[minor] = (instance.name, instance.admin_up)
2255

    
2256
    # and now check them
2257
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2258
    test = not isinstance(used_minors, (tuple, list))
2259
    _ErrorIf(test, self.ENODEDRBD, node,
2260
             "cannot parse drbd status file: %s", str(used_minors))
2261
    if test:
2262
      # we cannot check drbd status
2263
      return
2264

    
2265
    for minor, (iname, must_exist) in node_drbd.items():
2266
      test = minor not in used_minors and must_exist
2267
      _ErrorIf(test, self.ENODEDRBD, node,
2268
               "drbd minor %d of instance %s is not active", minor, iname)
2269
    for minor in used_minors:
2270
      test = minor not in node_drbd
2271
      _ErrorIf(test, self.ENODEDRBD, node,
2272
               "unallocated drbd minor %d is in use", minor)
2273

    
2274
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2275
    """Builds the node OS structures.
2276

2277
    @type ninfo: L{objects.Node}
2278
    @param ninfo: the node to check
2279
    @param nresult: the remote results for the node
2280
    @param nimg: the node image object
2281

2282
    """
2283
    node = ninfo.name
2284
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2285

    
2286
    remote_os = nresult.get(constants.NV_OSLIST, None)
2287
    test = (not isinstance(remote_os, list) or
2288
            not compat.all(isinstance(v, list) and len(v) == 7
2289
                           for v in remote_os))
2290

    
2291
    _ErrorIf(test, self.ENODEOS, node,
2292
             "node hasn't returned valid OS data")
2293

    
2294
    nimg.os_fail = test
2295

    
2296
    if test:
2297
      return
2298

    
2299
    os_dict = {}
2300

    
2301
    for (name, os_path, status, diagnose,
2302
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2303

    
2304
      if name not in os_dict:
2305
        os_dict[name] = []
2306

    
2307
      # parameters is a list of lists instead of list of tuples due to
2308
      # JSON lacking a real tuple type, fix it:
2309
      parameters = [tuple(v) for v in parameters]
2310
      os_dict[name].append((os_path, status, diagnose,
2311
                            set(variants), set(parameters), set(api_ver)))
2312

    
2313
    nimg.oslist = os_dict
2314

    
2315
  def _VerifyNodeOS(self, ninfo, nimg, base):
2316
    """Verifies the node OS list.
2317

2318
    @type ninfo: L{objects.Node}
2319
    @param ninfo: the node to check
2320
    @param nimg: the node image object
2321
    @param base: the 'template' node we match against (e.g. from the master)
2322

2323
    """
2324
    node = ninfo.name
2325
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2326

    
2327
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2328

    
2329
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2330
    for os_name, os_data in nimg.oslist.items():
2331
      assert os_data, "Empty OS status for OS %s?!" % os_name
2332
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2333
      _ErrorIf(not f_status, self.ENODEOS, node,
2334
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2335
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
2336
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2337
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2338
      # comparisons with the 'base' image
2339
      test = os_name not in base.oslist
2340
      _ErrorIf(test, self.ENODEOS, node,
2341
               "Extra OS %s not present on reference node (%s)",
2342
               os_name, base.name)
2343
      if test:
2344
        continue
2345
      assert base.oslist[os_name], "Base node has empty OS status?"
2346
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2347
      if not b_status:
2348
        # base OS is invalid, skipping
2349
        continue
2350
      for kind, a, b in [("API version", f_api, b_api),
2351
                         ("variants list", f_var, b_var),
2352
                         ("parameters", beautify_params(f_param),
2353
                          beautify_params(b_param))]:
2354
        _ErrorIf(a != b, self.ENODEOS, node,
2355
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2356
                 kind, os_name, base.name,
2357
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2358

    
2359
    # check any missing OSes
2360
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2361
    _ErrorIf(missing, self.ENODEOS, node,
2362
             "OSes present on reference node %s but missing on this node: %s",
2363
             base.name, utils.CommaJoin(missing))
2364

    
2365
  def _VerifyOob(self, ninfo, nresult):
2366
    """Verifies out of band functionality of a node.
2367

2368
    @type ninfo: L{objects.Node}
2369
    @param ninfo: the node to check
2370
    @param nresult: the remote results for the node
2371

2372
    """
2373
    node = ninfo.name
2374
    # We just have to verify the paths on master and/or master candidates
2375
    # as the oob helper is invoked on the master
2376
    if ((ninfo.master_candidate or ninfo.master_capable) and
2377
        constants.NV_OOB_PATHS in nresult):
2378
      for path_result in nresult[constants.NV_OOB_PATHS]:
2379
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
2380

    
2381
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2382
    """Verifies and updates the node volume data.
2383

2384
    This function will update a L{NodeImage}'s internal structures
2385
    with data from the remote call.
2386

2387
    @type ninfo: L{objects.Node}
2388
    @param ninfo: the node to check
2389
    @param nresult: the remote results for the node
2390
    @param nimg: the node image object
2391
    @param vg_name: the configured VG name
2392

2393
    """
2394
    node = ninfo.name
2395
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2396

    
2397
    nimg.lvm_fail = True
2398
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2399
    if vg_name is None:
2400
      pass
2401
    elif isinstance(lvdata, basestring):
2402
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
2403
               utils.SafeEncode(lvdata))
2404
    elif not isinstance(lvdata, dict):
2405
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
2406
    else:
2407
      nimg.volumes = lvdata
2408
      nimg.lvm_fail = False
2409

    
2410
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2411
    """Verifies and updates the node instance list.
2412

2413
    If the listing was successful, then updates this node's instance
2414
    list. Otherwise, it marks the RPC call as failed for the instance
2415
    list key.
2416

2417
    @type ninfo: L{objects.Node}
2418
    @param ninfo: the node to check
2419
    @param nresult: the remote results for the node
2420
    @param nimg: the node image object
2421

2422
    """
2423
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2424
    test = not isinstance(idata, list)
2425
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
2426
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
2427
    if test:
2428
      nimg.hyp_fail = True
2429
    else:
2430
      nimg.instances = idata
2431

    
2432
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2433
    """Verifies and computes a node information map
2434

2435
    @type ninfo: L{objects.Node}
2436
    @param ninfo: the node to check
2437
    @param nresult: the remote results for the node
2438
    @param nimg: the node image object
2439
    @param vg_name: the configured VG name
2440

2441
    """
2442
    node = ninfo.name
2443
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2444

    
2445
    # try to read free memory (from the hypervisor)
2446
    hv_info = nresult.get(constants.NV_HVINFO, None)
2447
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2448
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
2449
    if not test:
2450
      try:
2451
        nimg.mfree = int(hv_info["memory_free"])
2452
      except (ValueError, TypeError):
2453
        _ErrorIf(True, self.ENODERPC, node,
2454
                 "node returned invalid nodeinfo, check hypervisor")
2455

    
2456
    # FIXME: devise a free space model for file based instances as well
2457
    if vg_name is not None:
2458
      test = (constants.NV_VGLIST not in nresult or
2459
              vg_name not in nresult[constants.NV_VGLIST])
2460
      _ErrorIf(test, self.ENODELVM, node,
2461
               "node didn't return data for the volume group '%s'"
2462
               " - it is either missing or broken", vg_name)
2463
      if not test:
2464
        try:
2465
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2466
        except (ValueError, TypeError):
2467
          _ErrorIf(True, self.ENODERPC, node,
2468
                   "node returned invalid LVM info, check LVM status")
2469

    
2470
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2471
    """Gets per-disk status information for all instances.
2472

2473
    @type nodelist: list of strings
2474
    @param nodelist: Node names
2475
    @type node_image: dict of (name, L{objects.Node})
2476
    @param node_image: Node objects
2477
    @type instanceinfo: dict of (name, L{objects.Instance})
2478
    @param instanceinfo: Instance objects
2479
    @rtype: {instance: {node: [(succes, payload)]}}
2480
    @return: a dictionary of per-instance dictionaries with nodes as
2481
        keys and disk information as values; the disk information is a
2482
        list of tuples (success, payload)
2483

2484
    """
2485
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2486

    
2487
    node_disks = {}
2488
    node_disks_devonly = {}
2489
    diskless_instances = set()
2490
    diskless = constants.DT_DISKLESS
2491

    
2492
    for nname in nodelist:
2493
      node_instances = list(itertools.chain(node_image[nname].pinst,
2494
                                            node_image[nname].sinst))
2495
      diskless_instances.update(inst for inst in node_instances
2496
                                if instanceinfo[inst].disk_template == diskless)
2497
      disks = [(inst, disk)
2498
               for inst in node_instances
2499
               for disk in instanceinfo[inst].disks]
2500

    
2501
      if not disks:
2502
        # No need to collect data
2503
        continue
2504

    
2505
      node_disks[nname] = disks
2506

    
2507
      # Creating copies as SetDiskID below will modify the objects and that can
2508
      # lead to incorrect data returned from nodes
2509
      devonly = [dev.Copy() for (_, dev) in disks]
2510

    
2511
      for dev in devonly:
2512
        self.cfg.SetDiskID(dev, nname)
2513

    
2514
      node_disks_devonly[nname] = devonly
2515

    
2516
    assert len(node_disks) == len(node_disks_devonly)
2517

    
2518
    # Collect data from all nodes with disks
2519
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2520
                                                          node_disks_devonly)
2521

    
2522
    assert len(result) == len(node_disks)
2523

    
2524
    instdisk = {}
2525

    
2526
    for (nname, nres) in result.items():
2527
      disks = node_disks[nname]
2528

    
2529
      if nres.offline:
2530
        # No data from this node
2531
        data = len(disks) * [(False, "node offline")]
2532
      else:
2533
        msg = nres.fail_msg
2534
        _ErrorIf(msg, self.ENODERPC, nname,
2535
                 "while getting disk information: %s", msg)
2536
        if msg:
2537
          # No data from this node
2538
          data = len(disks) * [(False, msg)]
2539
        else:
2540
          data = []
2541
          for idx, i in enumerate(nres.payload):
2542
            if isinstance(i, (tuple, list)) and len(i) == 2:
2543
              data.append(i)
2544
            else:
2545
              logging.warning("Invalid result from node %s, entry %d: %s",
2546
                              nname, idx, i)
2547
              data.append((False, "Invalid result from the remote node"))
2548

    
2549
      for ((inst, _), status) in zip(disks, data):
2550
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2551

    
2552
    # Add empty entries for diskless instances.
2553
    for inst in diskless_instances:
2554
      assert inst not in instdisk
2555
      instdisk[inst] = {}
2556

    
2557
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2558
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2559
                      compat.all(isinstance(s, (tuple, list)) and
2560
                                 len(s) == 2 for s in statuses)
2561
                      for inst, nnames in instdisk.items()
2562
                      for nname, statuses in nnames.items())
2563
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2564

    
2565
    return instdisk
2566

    
2567
  @staticmethod
2568
  def _SshNodeSelector(group_uuid, all_nodes):
2569
    """Create endless iterators for all potential SSH check hosts.
2570

2571
    """
2572
    nodes = [node for node in all_nodes
2573
             if (node.group != group_uuid and
2574
                 not node.offline)]
2575
    keyfunc = operator.attrgetter("group")
2576

    
2577
    return map(itertools.cycle,
2578
               [sorted(map(operator.attrgetter("name"), names))
2579
                for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
2580
                                                  keyfunc)])
2581

    
2582
  @classmethod
2583
  def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
2584
    """Choose which nodes should talk to which other nodes.
2585

2586
    We will make nodes contact all nodes in their group, and one node from
2587
    every other group.
2588

2589
    @warning: This algorithm has a known issue if one node group is much
2590
      smaller than others (e.g. just one node). In such a case all other
2591
      nodes will talk to the single node.
2592

2593
    """
2594
    online_nodes = sorted(node.name for node in group_nodes if not node.offline)
2595
    sel = cls._SshNodeSelector(group_uuid, all_nodes)
2596

    
2597
    return (online_nodes,
2598
            dict((name, sorted([i.next() for i in sel]))
2599
                 for name in online_nodes))
2600

    
2601
  def BuildHooksEnv(self):
2602
    """Build hooks env.
2603

2604
    Cluster-Verify hooks just ran in the post phase and their failure makes
2605
    the output be logged in the verify output and the verification to fail.
2606

2607
    """
2608
    env = {
2609
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2610
      }
2611

    
2612
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2613
               for node in self.my_node_info.values())
2614

    
2615
    return env
2616

    
2617
  def BuildHooksNodes(self):
2618
    """Build hooks nodes.
2619

2620
    """
2621
    return ([], self.my_node_names)
2622

    
2623
  def Exec(self, feedback_fn):
2624
    """Verify integrity of the node group, performing various test on nodes.
2625

2626
    """
2627
    # This method has too many local variables. pylint: disable=R0914
2628
    feedback_fn("* Verifying group '%s'" % self.group_info.name)
2629

    
2630
    if not self.my_node_names:
2631
      # empty node group
2632
      feedback_fn("* Empty node group, skipping verification")
2633
      return True
2634

    
2635
    self.bad = False
2636
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2637
    verbose = self.op.verbose
2638
    self._feedback_fn = feedback_fn
2639

    
2640
    vg_name = self.cfg.GetVGName()
2641
    drbd_helper = self.cfg.GetDRBDHelper()
2642
    cluster = self.cfg.GetClusterInfo()
2643
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2644
    hypervisors = cluster.enabled_hypervisors
2645
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2646

    
2647
    i_non_redundant = [] # Non redundant instances
2648
    i_non_a_balanced = [] # Non auto-balanced instances
2649
    n_offline = 0 # Count of offline nodes
2650
    n_drained = 0 # Count of nodes being drained
2651
    node_vol_should = {}
2652

    
2653
    # FIXME: verify OS list
2654

    
2655
    # File verification
2656
    filemap = _ComputeAncillaryFiles(cluster, False)
2657

    
2658
    # do local checksums
2659
    master_node = self.master_node = self.cfg.GetMasterNode()
2660
    master_ip = self.cfg.GetMasterIP()
2661

    
2662
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2663

    
2664
    node_verify_param = {
2665
      constants.NV_FILELIST:
2666
        utils.UniqueSequence(filename
2667
                             for files in filemap
2668
                             for filename in files),
2669
      constants.NV_NODELIST:
2670
        self._SelectSshCheckNodes(node_data_list, self.group_uuid,
2671
                                  self.all_node_info.values()),
2672
      constants.NV_HYPERVISOR: hypervisors,
2673
      constants.NV_HVPARAMS:
2674
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2675
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2676
                                 for node in node_data_list
2677
                                 if not node.offline],
2678
      constants.NV_INSTANCELIST: hypervisors,
2679
      constants.NV_VERSION: None,
2680
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2681
      constants.NV_NODESETUP: None,
2682
      constants.NV_TIME: None,
2683
      constants.NV_MASTERIP: (master_node, master_ip),
2684
      constants.NV_OSLIST: None,
2685
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2686
      }
2687

    
2688
    if vg_name is not None:
2689
      node_verify_param[constants.NV_VGLIST] = None
2690
      node_verify_param[constants.NV_LVLIST] = vg_name
2691
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2692
      node_verify_param[constants.NV_DRBDLIST] = None
2693

    
2694
    if drbd_helper:
2695
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2696

    
2697
    # bridge checks
2698
    # FIXME: this needs to be changed per node-group, not cluster-wide
2699
    bridges = set()
2700
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2701
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2702
      bridges.add(default_nicpp[constants.NIC_LINK])
2703
    for instance in self.my_inst_info.values():
2704
      for nic in instance.nics:
2705
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2706
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2707
          bridges.add(full_nic[constants.NIC_LINK])
2708

    
2709
    if bridges:
2710
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2711

    
2712
    # Build our expected cluster state
2713
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2714
                                                 name=node.name,
2715
                                                 vm_capable=node.vm_capable))
2716
                      for node in node_data_list)
2717

    
2718
    # Gather OOB paths
2719
    oob_paths = []
2720
    for node in self.all_node_info.values():
2721
      path = _SupportsOob(self.cfg, node)
2722
      if path and path not in oob_paths:
2723
        oob_paths.append(path)
2724

    
2725
    if oob_paths:
2726
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2727

    
2728
    for instance in self.my_inst_names:
2729
      inst_config = self.my_inst_info[instance]
2730

    
2731
      for nname in inst_config.all_nodes:
2732
        if nname not in node_image:
2733
          gnode = self.NodeImage(name=nname)
2734
          gnode.ghost = (nname not in self.all_node_info)
2735
          node_image[nname] = gnode
2736

    
2737
      inst_config.MapLVsByNode(node_vol_should)
2738

    
2739
      pnode = inst_config.primary_node
2740
      node_image[pnode].pinst.append(instance)
2741

    
2742
      for snode in inst_config.secondary_nodes:
2743
        nimg = node_image[snode]
2744
        nimg.sinst.append(instance)
2745
        if pnode not in nimg.sbp:
2746
          nimg.sbp[pnode] = []
2747
        nimg.sbp[pnode].append(instance)
2748

    
2749
    # At this point, we have the in-memory data structures complete,
2750
    # except for the runtime information, which we'll gather next
2751

    
2752
    # Due to the way our RPC system works, exact response times cannot be
2753
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2754
    # time before and after executing the request, we can at least have a time
2755
    # window.
2756
    nvinfo_starttime = time.time()
2757
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2758
                                           node_verify_param,
2759
                                           self.cfg.GetClusterName())
2760
    nvinfo_endtime = time.time()
2761

    
2762
    if self.extra_lv_nodes and vg_name is not None:
2763
      extra_lv_nvinfo = \
2764
          self.rpc.call_node_verify(self.extra_lv_nodes,
2765
                                    {constants.NV_LVLIST: vg_name},
2766
                                    self.cfg.GetClusterName())
2767
    else:
2768
      extra_lv_nvinfo = {}
2769

    
2770
    all_drbd_map = self.cfg.ComputeDRBDMap()
2771

    
2772
    feedback_fn("* Gathering disk information (%s nodes)" %
2773
                len(self.my_node_names))
2774
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2775
                                     self.my_inst_info)
2776

    
2777
    feedback_fn("* Verifying configuration file consistency")
2778

    
2779
    # If not all nodes are being checked, we need to make sure the master node
2780
    # and a non-checked vm_capable node are in the list.
2781
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2782
    if absent_nodes:
2783
      vf_nvinfo = all_nvinfo.copy()
2784
      vf_node_info = list(self.my_node_info.values())
2785
      additional_nodes = []
2786
      if master_node not in self.my_node_info:
2787
        additional_nodes.append(master_node)
2788
        vf_node_info.append(self.all_node_info[master_node])
2789
      # Add the first vm_capable node we find which is not included
2790
      for node in absent_nodes:
2791
        nodeinfo = self.all_node_info[node]
2792
        if nodeinfo.vm_capable and not nodeinfo.offline:
2793
          additional_nodes.append(node)
2794
          vf_node_info.append(self.all_node_info[node])
2795
          break
2796
      key = constants.NV_FILELIST
2797
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2798
                                                 {key: node_verify_param[key]},
2799
                                                 self.cfg.GetClusterName()))
2800
    else:
2801
      vf_nvinfo = all_nvinfo
2802
      vf_node_info = self.my_node_info.values()
2803

    
2804
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2805

    
2806
    feedback_fn("* Verifying node status")
2807

    
2808
    refos_img = None
2809

    
2810
    for node_i in node_data_list:
2811
      node = node_i.name
2812
      nimg = node_image[node]
2813

    
2814
      if node_i.offline:
2815
        if verbose:
2816
          feedback_fn("* Skipping offline node %s" % (node,))
2817
        n_offline += 1
2818
        continue
2819

    
2820
      if node == master_node:
2821
        ntype = "master"
2822
      elif node_i.master_candidate:
2823
        ntype = "master candidate"
2824
      elif node_i.drained:
2825
        ntype = "drained"
2826
        n_drained += 1
2827
      else:
2828
        ntype = "regular"
2829
      if verbose:
2830
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2831

    
2832
      msg = all_nvinfo[node].fail_msg
2833
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2834
      if msg:
2835
        nimg.rpc_fail = True
2836
        continue
2837

    
2838
      nresult = all_nvinfo[node].payload
2839

    
2840
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2841
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2842
      self._VerifyNodeNetwork(node_i, nresult)
2843
      self._VerifyOob(node_i, nresult)
2844

    
2845
      if nimg.vm_capable:
2846
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2847
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2848
                             all_drbd_map)
2849

    
2850
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2851
        self._UpdateNodeInstances(node_i, nresult, nimg)
2852
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2853
        self._UpdateNodeOS(node_i, nresult, nimg)
2854

    
2855
        if not nimg.os_fail:
2856
          if refos_img is None:
2857
            refos_img = nimg
2858
          self._VerifyNodeOS(node_i, nimg, refos_img)
2859
        self._VerifyNodeBridges(node_i, nresult, bridges)
2860

    
2861
        # Check whether all running instancies are primary for the node. (This
2862
        # can no longer be done from _VerifyInstance below, since some of the
2863
        # wrong instances could be from other node groups.)
2864
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2865

    
2866
        for inst in non_primary_inst:
2867
          test = inst in self.all_inst_info
2868
          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
2869
                   "instance should not run on node %s", node_i.name)
2870
          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
2871
                   "node is running unknown instance %s", inst)
2872

    
2873
    for node, result in extra_lv_nvinfo.items():
2874
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
2875
                              node_image[node], vg_name)
2876

    
2877
    feedback_fn("* Verifying instance status")
2878
    for instance in self.my_inst_names:
2879
      if verbose:
2880
        feedback_fn("* Verifying instance %s" % instance)
2881
      inst_config = self.my_inst_info[instance]
2882
      self._VerifyInstance(instance, inst_config, node_image,
2883
                           instdisk[instance])
2884
      inst_nodes_offline = []
2885

    
2886
      pnode = inst_config.primary_node
2887
      pnode_img = node_image[pnode]
2888
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2889
               self.ENODERPC, pnode, "instance %s, connection to"
2890
               " primary node failed", instance)
2891

    
2892
      _ErrorIf(inst_config.admin_up and pnode_img.offline,
2893
               self.EINSTANCEBADNODE, instance,
2894
               "instance is marked as running and lives on offline node %s",
2895
               inst_config.primary_node)
2896

    
2897
      # If the instance is non-redundant we cannot survive losing its primary
2898
      # node, so we are not N+1 compliant. On the other hand we have no disk
2899
      # templates with more than one secondary so that situation is not well
2900
      # supported either.
2901
      # FIXME: does not support file-backed instances
2902
      if not inst_config.secondary_nodes:
2903
        i_non_redundant.append(instance)
2904

    
2905
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2906
               instance, "instance has multiple secondary nodes: %s",
2907
               utils.CommaJoin(inst_config.secondary_nodes),
2908
               code=self.ETYPE_WARNING)
2909

    
2910
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2911
        pnode = inst_config.primary_node
2912
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2913
        instance_groups = {}
2914

    
2915
        for node in instance_nodes:
2916
          instance_groups.setdefault(self.all_node_info[node].group,
2917
                                     []).append(node)
2918

    
2919
        pretty_list = [
2920
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2921
          # Sort so that we always list the primary node first.
2922
          for group, nodes in sorted(instance_groups.items(),
2923
                                     key=lambda (_, nodes): pnode in nodes,
2924
                                     reverse=True)]
2925

    
2926
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2927
                      instance, "instance has primary and secondary nodes in"
2928
                      " different groups: %s", utils.CommaJoin(pretty_list),
2929
                      code=self.ETYPE_WARNING)
2930

    
2931
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2932
        i_non_a_balanced.append(instance)
2933

    
2934
      for snode in inst_config.secondary_nodes:
2935
        s_img = node_image[snode]
2936
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2937
                 "instance %s, connection to secondary node failed", instance)
2938

    
2939
        if s_img.offline:
2940
          inst_nodes_offline.append(snode)
2941

    
2942
      # warn that the instance lives on offline nodes
2943
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2944
               "instance has offline secondary node(s) %s",
2945
               utils.CommaJoin(inst_nodes_offline))
2946
      # ... or ghost/non-vm_capable nodes
2947
      for node in inst_config.all_nodes:
2948
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2949
                 "instance lives on ghost node %s", node)
2950
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2951
                 instance, "instance lives on non-vm_capable node %s", node)
2952

    
2953
    feedback_fn("* Verifying orphan volumes")
2954
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2955

    
2956
    # We will get spurious "unknown volume" warnings if any node of this group
2957
    # is secondary for an instance whose primary is in another group. To avoid
2958
    # them, we find these instances and add their volumes to node_vol_should.
2959
    for inst in self.all_inst_info.values():
2960
      for secondary in inst.secondary_nodes:
2961
        if (secondary in self.my_node_info
2962
            and inst.name not in self.my_inst_info):
2963
          inst.MapLVsByNode(node_vol_should)
2964
          break
2965

    
2966
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2967

    
2968
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2969
      feedback_fn("* Verifying N+1 Memory redundancy")
2970
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
2971

    
2972
    feedback_fn("* Other Notes")
2973
    if i_non_redundant:
2974
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2975
                  % len(i_non_redundant))
2976

    
2977
    if i_non_a_balanced:
2978
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2979
                  % len(i_non_a_balanced))
2980

    
2981
    if n_offline:
2982
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2983

    
2984
    if n_drained:
2985
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2986

    
2987
    return not self.bad
2988

    
2989
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2990
    """Analyze the post-hooks' result
2991

2992
    This method analyses the hook result, handles it, and sends some
2993
    nicely-formatted feedback back to the user.
2994

2995
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
2996
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
2997
    @param hooks_results: the results of the multi-node hooks rpc call
2998
    @param feedback_fn: function used send feedback back to the caller
2999
    @param lu_result: previous Exec result
3000
    @return: the new Exec result, based on the previous result
3001
        and hook results
3002

3003
    """
3004
    # We only really run POST phase hooks, only for non-empty groups,
3005
    # and are only interested in their results
3006
    if not self.my_node_names:
3007
      # empty node group
3008
      pass
3009
    elif phase == constants.HOOKS_PHASE_POST:
3010
      # Used to change hooks' output to proper indentation
3011
      feedback_fn("* Hooks Results")
3012
      assert hooks_results, "invalid result from hooks"
3013

    
3014
      for node_name in hooks_results:
3015
        res = hooks_results[node_name]
3016
        msg = res.fail_msg
3017
        test = msg and not res.offline
3018
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
3019
                      "Communication failure in hooks execution: %s", msg)
3020
        if res.offline or msg:
3021
          # No need to investigate payload if node is offline or gave
3022
          # an error.
3023
          continue
3024
        for script, hkr, output in res.payload:
3025
          test = hkr == constants.HKR_FAIL
3026
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
3027
                        "Script %s failed, output:", script)
3028
          if test:
3029
            output = self._HOOKS_INDENT_RE.sub("      ", output)
3030
            feedback_fn("%s" % output)
3031
            lu_result = False
3032

    
3033
    return lu_result
3034

    
3035

    
3036
class LUClusterVerifyDisks(NoHooksLU):
3037
  """Verifies the cluster disks status.
3038

3039
  """
3040
  REQ_BGL = False
3041

    
3042
  def ExpandNames(self):
3043
    self.share_locks = _ShareAll()
3044
    self.needed_locks = {
3045
      locking.LEVEL_NODEGROUP: locking.ALL_SET,
3046
      }
3047

    
3048
  def Exec(self, feedback_fn):
3049
    group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3050

    
3051
    # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3052
    return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3053
                           for group in group_names])
3054

    
3055

    
3056
class LUGroupVerifyDisks(NoHooksLU):
3057
  """Verifies the status of all disks in a node group.
3058

3059
  """
3060
  REQ_BGL = False
3061

    
3062
  def ExpandNames(self):
3063
    # Raises errors.OpPrereqError on its own if group can't be found
3064
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3065

    
3066
    self.share_locks = _ShareAll()
3067
    self.needed_locks = {
3068
      locking.LEVEL_INSTANCE: [],
3069
      locking.LEVEL_NODEGROUP: [],
3070
      locking.LEVEL_NODE: [],
3071
      }
3072

    
3073
  def DeclareLocks(self, level):
3074
    if level == locking.LEVEL_INSTANCE:
3075
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
3076

    
3077
      # Lock instances optimistically, needs verification once node and group
3078
      # locks have been acquired
3079
      self.needed_locks[locking.LEVEL_INSTANCE] = \
3080
        self.cfg.GetNodeGroupInstances(self.group_uuid)
3081

    
3082
    elif level == locking.LEVEL_NODEGROUP:
3083
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3084

    
3085
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
3086
        set([self.group_uuid] +
3087
            # Lock all groups used by instances optimistically; this requires
3088
            # going via the node before it's locked, requiring verification
3089
            # later on
3090
            [group_uuid
3091
             for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3092
             for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3093

    
3094
    elif level == locking.LEVEL_NODE:
3095
      # This will only lock the nodes in the group to be verified which contain
3096
      # actual instances
3097
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3098
      self._LockInstancesNodes()
3099

    
3100
      # Lock all nodes in group to be verified
3101
      assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3102
      member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3103
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3104

    
3105
  def CheckPrereq(self):
3106
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3107
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3108
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3109

    
3110
    assert self.group_uuid in owned_groups
3111

    
3112
    # Check if locked instances are still correct
3113
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3114

    
3115
    # Get instance information
3116
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3117

    
3118
    # Check if node groups for locked instances are still correct
3119
    for (instance_name, inst) in self.instances.items():
3120
      assert owned_nodes.issuperset(inst.all_nodes), \
3121
        "Instance %s's nodes changed while we kept the lock" % instance_name
3122

    
3123
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
3124
                                             owned_groups)
3125

    
3126
      assert self.group_uuid in inst_groups, \
3127
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
3128

    
3129
  def Exec(self, feedback_fn):
3130
    """Verify integrity of cluster disks.
3131

3132
    @rtype: tuple of three items
3133
    @return: a tuple of (dict of node-to-node_error, list of instances
3134
        which need activate-disks, dict of instance: (node, volume) for
3135
        missing volumes
3136

3137
    """
3138
    res_nodes = {}
3139
    res_instances = set()
3140
    res_missing = {}
3141

    
3142
    nv_dict = _MapInstanceDisksToNodes([inst
3143
                                        for inst in self.instances.values()
3144
                                        if inst.admin_up])
3145

    
3146
    if nv_dict:
3147
      nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3148
                             set(self.cfg.GetVmCapableNodeList()))
3149

    
3150
      node_lvs = self.rpc.call_lv_list(nodes, [])
3151

    
3152
      for (node, node_res) in node_lvs.items():
3153
        if node_res.offline:
3154
          continue
3155

    
3156
        msg = node_res.fail_msg
3157
        if msg:
3158
          logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3159
          res_nodes[node] = msg
3160
          continue
3161

    
3162
        for lv_name, (_, _, lv_online) in node_res.payload.items():
3163
          inst = nv_dict.pop((node, lv_name), None)
3164
          if not (lv_online or inst is None):
3165
            res_instances.add(inst)
3166

    
3167
      # any leftover items in nv_dict are missing LVs, let's arrange the data
3168
      # better
3169
      for key, inst in nv_dict.iteritems():
3170
        res_missing.setdefault(inst, []).append(key)
3171

    
3172
    return (res_nodes, list(res_instances), res_missing)
3173

    
3174

    
3175
class LUClusterRepairDiskSizes(NoHooksLU):
3176
  """Verifies the cluster disks sizes.
3177

3178
  """
3179
  REQ_BGL = False
3180

    
3181
  def ExpandNames(self):
3182
    if self.op.instances:
3183
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
3184
      self.needed_locks = {
3185
        locking.LEVEL_NODE: [],
3186
        locking.LEVEL_INSTANCE: self.wanted_names,
3187
        }
3188
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3189
    else:
3190
      self.wanted_names = None
3191
      self.needed_locks = {
3192
        locking.LEVEL_NODE: locking.ALL_SET,
3193
        locking.LEVEL_INSTANCE: locking.ALL_SET,
3194
        }
3195
    self.share_locks = _ShareAll()
3196

    
3197
  def DeclareLocks(self, level):
3198
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
3199
      self._LockInstancesNodes(primary_only=True)
3200

    
3201
  def CheckPrereq(self):
3202
    """Check prerequisites.
3203

3204
    This only checks the optional instance list against the existing names.
3205

3206
    """
3207
    if self.wanted_names is None:
3208
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3209

    
3210
    self.wanted_instances = \
3211
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3212

    
3213
  def _EnsureChildSizes(self, disk):
3214
    """Ensure children of the disk have the needed disk size.
3215

3216
    This is valid mainly for DRBD8 and fixes an issue where the
3217
    children have smaller disk size.
3218

3219
    @param disk: an L{ganeti.objects.Disk} object
3220

3221
    """
3222
    if disk.dev_type == constants.LD_DRBD8:
3223
      assert disk.children, "Empty children for DRBD8?"
3224
      fchild = disk.children[0]
3225
      mismatch = fchild.size < disk.size
3226
      if mismatch:
3227
        self.LogInfo("Child disk has size %d, parent %d, fixing",
3228
                     fchild.size, disk.size)
3229
        fchild.size = disk.size
3230

    
3231
      # and we recurse on this child only, not on the metadev
3232
      return self._EnsureChildSizes(fchild) or mismatch
3233
    else:
3234
      return False
3235

    
3236
  def Exec(self, feedback_fn):
3237
    """Verify the size of cluster disks.
3238

3239
    """
3240
    # TODO: check child disks too
3241
    # TODO: check differences in size between primary/secondary nodes
3242
    per_node_disks = {}
3243
    for instance in self.wanted_instances:
3244
      pnode = instance.primary_node
3245
      if pnode not in per_node_disks:
3246
        per_node_disks[pnode] = []
3247
      for idx, disk in enumerate(instance.disks):
3248
        per_node_disks[pnode].append((instance, idx, disk))
3249

    
3250
    changed = []
3251
    for node, dskl in per_node_disks.items():
3252
      newl = [v[2].Copy() for v in dskl]
3253
      for dsk in newl:
3254
        self.cfg.SetDiskID(dsk, node)
3255
      result = self.rpc.call_blockdev_getsize(node, newl)
3256
      if result.fail_msg:
3257
        self.LogWarning("Failure in blockdev_getsize call to node"
3258
                        " %s, ignoring", node)
3259
        continue
3260
      if len(result.payload) != len(dskl):
3261
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
3262
                        " result.payload=%s", node, len(dskl), result.payload)
3263
        self.LogWarning("Invalid result from node %s, ignoring node results",
3264
                        node)
3265
        continue
3266
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
3267
        if size is None:
3268
          self.LogWarning("Disk %d of instance %s did not return size"
3269
                          " information, ignoring", idx, instance.name)
3270
          continue
3271
        if not isinstance(size, (int, long)):
3272
          self.LogWarning("Disk %d of instance %s did not return valid"
3273
                          " size information, ignoring", idx, instance.name)
3274
          continue
3275
        size = size >> 20
3276
        if size != disk.size:
3277
          self.LogInfo("Disk %d of instance %s has mismatched size,"
3278
                       " correcting: recorded %d, actual %d", idx,
3279
                       instance.name, disk.size, size)
3280
          disk.size = size
3281
          self.cfg.Update(instance, feedback_fn)
3282
          changed.append((instance.name, idx, size))
3283
        if self._EnsureChildSizes(disk):
3284
          self.cfg.Update(instance, feedback_fn)
3285
          changed.append((instance.name, idx, disk.size))
3286
    return changed
3287

    
3288

    
3289
class LUClusterRename(LogicalUnit):
3290
  """Rename the cluster.
3291

3292
  """
3293
  HPATH = "cluster-rename"
3294
  HTYPE = constants.HTYPE_CLUSTER
3295

    
3296
  def BuildHooksEnv(self):
3297
    """Build hooks env.
3298

3299
    """
3300
    return {
3301
      "OP_TARGET": self.cfg.GetClusterName(),
3302
      "NEW_NAME": self.op.name,
3303
      }
3304

    
3305
  def BuildHooksNodes(self):
3306
    """Build hooks nodes.
3307

3308
    """
3309
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3310

    
3311
  def CheckPrereq(self):
3312
    """Verify that the passed name is a valid one.
3313

3314
    """
3315
    hostname = netutils.GetHostname(name=self.op.name,
3316
                                    family=self.cfg.GetPrimaryIPFamily())
3317

    
3318
    new_name = hostname.name
3319
    self.ip = new_ip = hostname.ip
3320
    old_name = self.cfg.GetClusterName()
3321
    old_ip = self.cfg.GetMasterIP()
3322
    if new_name == old_name and new_ip == old_ip:
3323
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3324
                                 " cluster has changed",
3325
                                 errors.ECODE_INVAL)
3326
    if new_ip != old_ip:
3327
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3328
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3329
                                   " reachable on the network" %
3330
                                   new_ip, errors.ECODE_NOTUNIQUE)
3331

    
3332
    self.op.name = new_name
3333

    
3334
  def Exec(self, feedback_fn):
3335
    """Rename the cluster.
3336

3337
    """
3338
    clustername = self.op.name
3339
    ip = self.ip
3340

    
3341
    # shutdown the master IP
3342
    master = self.cfg.GetMasterNode()
3343
    result = self.rpc.call_node_deactivate_master_ip(master)
3344
    result.Raise("Could not disable the master role")
3345

    
3346
    try:
3347
      cluster = self.cfg.GetClusterInfo()
3348
      cluster.cluster_name = clustername
3349
      cluster.master_ip = ip
3350
      self.cfg.Update(cluster, feedback_fn)
3351

    
3352
      # update the known hosts file
3353
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3354
      node_list = self.cfg.GetOnlineNodeList()
3355
      try:
3356
        node_list.remove(master)
3357
      except ValueError:
3358
        pass
3359
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3360
    finally:
3361
      result = self.rpc.call_node_activate_master_ip(master)
3362
      msg = result.fail_msg
3363
      if msg:
3364
        self.LogWarning("Could not re-enable the master role on"
3365
                        " the master, please restart manually: %s", msg)
3366

    
3367
    return clustername
3368

    
3369

    
3370
class LUClusterSetParams(LogicalUnit):
3371
  """Change the parameters of the cluster.
3372

3373
  """
3374
  HPATH = "cluster-modify"
3375
  HTYPE = constants.HTYPE_CLUSTER
3376
  REQ_BGL = False
3377

    
3378
  def CheckArguments(self):
3379
    """Check parameters
3380

3381
    """
3382
    if self.op.uid_pool:
3383
      uidpool.CheckUidPool(self.op.uid_pool)
3384

    
3385
    if self.op.add_uids:
3386
      uidpool.CheckUidPool(self.op.add_uids)
3387

    
3388
    if self.op.remove_uids:
3389
      uidpool.CheckUidPool(self.op.remove_uids)
3390

    
3391
  def ExpandNames(self):
3392
    # FIXME: in the future maybe other cluster params won't require checking on
3393
    # all nodes to be modified.
3394
    self.needed_locks = {
3395
      locking.LEVEL_NODE: locking.ALL_SET,
3396
    }
3397
    self.share_locks[locking.LEVEL_NODE] = 1
3398

    
3399
  def BuildHooksEnv(self):
3400
    """Build hooks env.
3401

3402
    """
3403
    return {
3404
      "OP_TARGET": self.cfg.GetClusterName(),
3405
      "NEW_VG_NAME": self.op.vg_name,
3406
      }
3407

    
3408
  def BuildHooksNodes(self):
3409
    """Build hooks nodes.
3410

3411
    """
3412
    mn = self.cfg.GetMasterNode()
3413
    return ([mn], [mn])
3414

    
3415
  def CheckPrereq(self):
3416
    """Check prerequisites.
3417

3418
    This checks whether the given params don't conflict and
3419
    if the given volume group is valid.
3420

3421
    """
3422
    if self.op.vg_name is not None and not self.op.vg_name:
3423
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3424
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3425
                                   " instances exist", errors.ECODE_INVAL)
3426

    
3427
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3428
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3429
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3430
                                   " drbd-based instances exist",
3431
                                   errors.ECODE_INVAL)
3432

    
3433
    node_list = self.owned_locks(locking.LEVEL_NODE)
3434

    
3435
    # if vg_name not None, checks given volume group on all nodes
3436
    if self.op.vg_name:
3437
      vglist = self.rpc.call_vg_list(node_list)
3438
      for node in node_list:
3439
        msg = vglist[node].fail_msg
3440
        if msg:
3441
          # ignoring down node
3442
          self.LogWarning("Error while gathering data on node %s"
3443
                          " (ignoring node): %s", node, msg)
3444
          continue
3445
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3446
                                              self.op.vg_name,
3447
                                              constants.MIN_VG_SIZE)
3448
        if vgstatus:
3449
          raise errors.OpPrereqError("Error on node '%s': %s" %
3450
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3451

    
3452
    if self.op.drbd_helper:
3453
      # checks given drbd helper on all nodes
3454
      helpers = self.rpc.call_drbd_helper(node_list)
3455
      for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3456
        if ninfo.offline:
3457
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3458
          continue
3459
        msg = helpers[node].fail_msg
3460
        if msg:
3461
          raise errors.OpPrereqError("Error checking drbd helper on node"
3462
                                     " '%s': %s" % (node, msg),
3463
                                     errors.ECODE_ENVIRON)
3464
        node_helper = helpers[node].payload
3465
        if node_helper != self.op.drbd_helper:
3466
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3467
                                     (node, node_helper), errors.ECODE_ENVIRON)
3468

    
3469
    self.cluster = cluster = self.cfg.GetClusterInfo()
3470
    # validate params changes
3471
    if self.op.beparams:
3472
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3473
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3474

    
3475
    if self.op.ndparams:
3476
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3477
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3478

    
3479
      # TODO: we need a more general way to handle resetting
3480
      # cluster-level parameters to default values
3481
      if self.new_ndparams["oob_program"] == "":
3482
        self.new_ndparams["oob_program"] = \
3483
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3484

    
3485
    if self.op.nicparams:
3486
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3487
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3488
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3489
      nic_errors = []
3490

    
3491
      # check all instances for consistency
3492
      for instance in self.cfg.GetAllInstancesInfo().values():
3493
        for nic_idx, nic in enumerate(instance.nics):
3494
          params_copy = copy.deepcopy(nic.nicparams)
3495
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3496

    
3497
          # check parameter syntax
3498
          try:
3499
            objects.NIC.CheckParameterSyntax(params_filled)
3500
          except errors.ConfigurationError, err:
3501
            nic_errors.append("Instance %s, nic/%d: %s" %
3502
                              (instance.name, nic_idx, err))
3503

    
3504
          # if we're moving instances to routed, check that they have an ip
3505
          target_mode = params_filled[constants.NIC_MODE]
3506
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3507
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3508
                              " address" % (instance.name, nic_idx))
3509
      if nic_errors:
3510
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3511
                                   "\n".join(nic_errors))
3512

    
3513
    # hypervisor list/parameters
3514
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3515
    if self.op.hvparams:
3516
      for hv_name, hv_dict in self.op.hvparams.items():
3517
        if hv_name not in self.new_hvparams:
3518
          self.new_hvparams[hv_name] = hv_dict
3519
        else:
3520
          self.new_hvparams[hv_name].update(hv_dict)
3521

    
3522
    # os hypervisor parameters
3523
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3524
    if self.op.os_hvp:
3525
      for os_name, hvs in self.op.os_hvp.items():
3526
        if os_name not in self.new_os_hvp:
3527
          self.new_os_hvp[os_name] = hvs
3528
        else:
3529
          for hv_name, hv_dict in hvs.items():
3530
            if hv_name not in self.new_os_hvp[os_name]:
3531
              self.new_os_hvp[os_name][hv_name] = hv_dict
3532
            else:
3533
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3534

    
3535
    # os parameters
3536
    self.new_osp = objects.FillDict(cluster.osparams, {})
3537
    if self.op.osparams:
3538
      for os_name, osp in self.op.osparams.items():
3539
        if os_name not in self.new_osp:
3540
          self.new_osp[os_name] = {}
3541

    
3542
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3543
                                                  use_none=True)
3544

    
3545
        if not self.new_osp[os_name]:
3546
          # we removed all parameters
3547
          del self.new_osp[os_name]
3548
        else:
3549
          # check the parameter validity (remote check)
3550
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3551
                         os_name, self.new_osp[os_name])
3552

    
3553
    # changes to the hypervisor list
3554
    if self.op.enabled_hypervisors is not None:
3555
      self.hv_list = self.op.enabled_hypervisors
3556
      for hv in self.hv_list:
3557
        # if the hypervisor doesn't already exist in the cluster
3558
        # hvparams, we initialize it to empty, and then (in both
3559
        # cases) we make sure to fill the defaults, as we might not
3560
        # have a complete defaults list if the hypervisor wasn't
3561
        # enabled before
3562
        if hv not in new_hvp:
3563
          new_hvp[hv] = {}
3564
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3565
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3566
    else:
3567
      self.hv_list = cluster.enabled_hypervisors
3568

    
3569
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3570
      # either the enabled list has changed, or the parameters have, validate
3571
      for hv_name, hv_params in self.new_hvparams.items():
3572
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3573
            (self.op.enabled_hypervisors and
3574
             hv_name in self.op.enabled_hypervisors)):
3575
          # either this is a new hypervisor, or its parameters have changed
3576
          hv_class = hypervisor.GetHypervisor(hv_name)
3577
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3578
          hv_class.CheckParameterSyntax(hv_params)
3579
          _CheckHVParams(self, node_list, hv_name, hv_params)
3580

    
3581
    if self.op.os_hvp:
3582
      # no need to check any newly-enabled hypervisors, since the
3583
      # defaults have already been checked in the above code-block
3584
      for os_name, os_hvp in self.new_os_hvp.items():
3585
        for hv_name, hv_params in os_hvp.items():
3586
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3587
          # we need to fill in the new os_hvp on top of the actual hv_p
3588
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3589
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3590
          hv_class = hypervisor.GetHypervisor(hv_name)
3591
          hv_class.CheckParameterSyntax(new_osp)
3592
          _CheckHVParams(self, node_list, hv_name, new_osp)
3593

    
3594
    if self.op.default_iallocator:
3595
      alloc_script = utils.FindFile(self.op.default_iallocator,
3596
                                    constants.IALLOCATOR_SEARCH_PATH,
3597
                                    os.path.isfile)
3598
      if alloc_script is None:
3599
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3600
                                   " specified" % self.op.default_iallocator,
3601
                                   errors.ECODE_INVAL)
3602

    
3603
  def Exec(self, feedback_fn):
3604
    """Change the parameters of the cluster.
3605

3606
    """
3607
    if self.op.vg_name is not None:
3608
      new_volume = self.op.vg_name
3609
      if not new_volume:
3610
        new_volume = None
3611
      if new_volume != self.cfg.GetVGName():
3612
        self.cfg.SetVGName(new_volume)
3613
      else:
3614
        feedback_fn("Cluster LVM configuration already in desired"
3615
                    " state, not changing")
3616
    if self.op.drbd_helper is not None:
3617
      new_helper = self.op.drbd_helper
3618
      if not new_helper:
3619
        new_helper = None
3620
      if new_helper != self.cfg.GetDRBDHelper():
3621
        self.cfg.SetDRBDHelper(new_helper)
3622
      else:
3623
        feedback_fn("Cluster DRBD helper already in desired state,"
3624
                    " not changing")
3625
    if self.op.hvparams:
3626
      self.cluster.hvparams = self.new_hvparams
3627
    if self.op.os_hvp:
3628
      self.cluster.os_hvp = self.new_os_hvp
3629
    if self.op.enabled_hypervisors is not None:
3630
      self.cluster.hvparams = self.new_hvparams
3631
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3632
    if self.op.beparams:
3633
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3634
    if self.op.nicparams:
3635
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3636
    if self.op.osparams:
3637
      self.cluster.osparams = self.new_osp
3638
    if self.op.ndparams:
3639
      self.cluster.ndparams = self.new_ndparams
3640

    
3641
    if self.op.candidate_pool_size is not None:
3642
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3643
      # we need to update the pool size here, otherwise the save will fail
3644
      _AdjustCandidatePool(self, [])
3645

    
3646
    if self.op.maintain_node_health is not None:
3647
      self.cluster.maintain_node_health = self.op.maintain_node_health
3648

    
3649
    if self.op.prealloc_wipe_disks is not None:
3650
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3651

    
3652
    if self.op.add_uids is not None:
3653
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3654

    
3655
    if self.op.remove_uids is not None:
3656
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3657

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

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

    
3664
    if self.op.reserved_lvs is not None:
3665
      self.cluster.reserved_lvs = self.op.reserved_lvs
3666

    
3667
    def helper_os(aname, mods, desc):
3668
      desc += " OS list"
3669
      lst = getattr(self.cluster, aname)
3670
      for key, val in mods:
3671
        if key == constants.DDM_ADD:
3672
          if val in lst:
3673
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3674
          else:
3675
            lst.append(val)
3676
        elif key == constants.DDM_REMOVE:
3677
          if val in lst:
3678
            lst.remove(val)
3679
          else:
3680
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3681
        else:
3682
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3683

    
3684
    if self.op.hidden_os:
3685
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3686

    
3687
    if self.op.blacklisted_os:
3688
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3689

    
3690
    if self.op.master_netdev:
3691
      master = self.cfg.GetMasterNode()
3692
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3693
                  self.cluster.master_netdev)
3694
      result = self.rpc.call_node_deactivate_master_ip(master)
3695
      result.Raise("Could not disable the master ip")
3696
      feedback_fn("Changing master_netdev from %s to %s" %
3697
                  (self.cluster.master_netdev, self.op.master_netdev))
3698
      self.cluster.master_netdev = self.op.master_netdev
3699

    
3700
    self.cfg.Update(self.cluster, feedback_fn)
3701

    
3702
    if self.op.master_netdev:
3703
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3704
                  self.op.master_netdev)
3705
      result = self.rpc.call_node_activate_master_ip(master)
3706
      if result.fail_msg:
3707
        self.LogWarning("Could not re-enable the master ip on"
3708
                        " the master, please restart manually: %s",
3709
                        result.fail_msg)
3710

    
3711

    
3712
def _UploadHelper(lu, nodes, fname):
3713
  """Helper for uploading a file and showing warnings.
3714

3715
  """
3716
  if os.path.exists(fname):
3717
    result = lu.rpc.call_upload_file(nodes, fname)
3718
    for to_node, to_result in result.items():
3719
      msg = to_result.fail_msg
3720
      if msg:
3721
        msg = ("Copy of file %s to node %s failed: %s" %
3722
               (fname, to_node, msg))
3723
        lu.proc.LogWarning(msg)
3724

    
3725

    
3726
def _ComputeAncillaryFiles(cluster, redist):
3727
  """Compute files external to Ganeti which need to be consistent.
3728

3729
  @type redist: boolean
3730
  @param redist: Whether to include files which need to be redistributed
3731

3732
  """
3733
  # Compute files for all nodes
3734
  files_all = set([
3735
    constants.SSH_KNOWN_HOSTS_FILE,
3736
    constants.CONFD_HMAC_KEY,
3737
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3738
    ])
3739

    
3740
  if not redist:
3741
    files_all.update(constants.ALL_CERT_FILES)
3742
    files_all.update(ssconf.SimpleStore().GetFileList())
3743
  else:
3744
    # we need to ship at least the RAPI certificate
3745
    files_all.add(constants.RAPI_CERT_FILE)
3746

    
3747
  if cluster.modify_etc_hosts:
3748
    files_all.add(constants.ETC_HOSTS)
3749

    
3750
  # Files which must either exist on all nodes or on none
3751
  files_all_opt = set([
3752
    constants.RAPI_USERS_FILE,
3753
    ])
3754

    
3755
  # Files which should only be on master candidates
3756
  files_mc = set()
3757
  if not redist:
3758
    files_mc.add(constants.CLUSTER_CONF_FILE)
3759

    
3760
  # Files which should only be on VM-capable nodes
3761
  files_vm = set(filename
3762
    for hv_name in cluster.enabled_hypervisors
3763
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
3764

    
3765
  # Filenames must be unique
3766
  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
3767
          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
3768
         "Found file listed in more than one file list"
3769

    
3770
  return (files_all, files_all_opt, files_mc, files_vm)
3771

    
3772

    
3773
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3774
  """Distribute additional files which are part of the cluster configuration.
3775

3776
  ConfigWriter takes care of distributing the config and ssconf files, but
3777
  there are more files which should be distributed to all nodes. This function
3778
  makes sure those are copied.
3779

3780
  @param lu: calling logical unit
3781
  @param additional_nodes: list of nodes not in the config to distribute to
3782
  @type additional_vm: boolean
3783
  @param additional_vm: whether the additional nodes are vm-capable or not
3784

3785
  """
3786
  # Gather target nodes
3787
  cluster = lu.cfg.GetClusterInfo()
3788
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3789

    
3790
  online_nodes = lu.cfg.GetOnlineNodeList()
3791
  vm_nodes = lu.cfg.GetVmCapableNodeList()
3792

    
3793
  if additional_nodes is not None:
3794
    online_nodes.extend(additional_nodes)
3795
    if additional_vm:
3796
      vm_nodes.extend(additional_nodes)
3797

    
3798
  # Never distribute to master node
3799
  for nodelist in [online_nodes, vm_nodes]:
3800
    if master_info.name in nodelist:
3801
      nodelist.remove(master_info.name)
3802

    
3803
  # Gather file lists
3804
  (files_all, files_all_opt, files_mc, files_vm) = \
3805
    _ComputeAncillaryFiles(cluster, True)
3806

    
3807
  # Never re-distribute configuration file from here
3808
  assert not (constants.CLUSTER_CONF_FILE in files_all or
3809
              constants.CLUSTER_CONF_FILE in files_vm)
3810
  assert not files_mc, "Master candidates not handled in this function"
3811

    
3812
  filemap = [
3813
    (online_nodes, files_all),
3814
    (online_nodes, files_all_opt),
3815
    (vm_nodes, files_vm),
3816
    ]
3817

    
3818
  # Upload the files
3819
  for (node_list, files) in filemap:
3820
    for fname in files:
3821
      _UploadHelper(lu, node_list, fname)
3822

    
3823

    
3824
class LUClusterRedistConf(NoHooksLU):
3825
  """Force the redistribution of cluster configuration.
3826

3827
  This is a very simple LU.
3828

3829
  """
3830
  REQ_BGL = False
3831

    
3832
  def ExpandNames(self):
3833
    self.needed_locks = {
3834
      locking.LEVEL_NODE: locking.ALL_SET,
3835
    }
3836
    self.share_locks[locking.LEVEL_NODE] = 1
3837

    
3838
  def Exec(self, feedback_fn):
3839
    """Redistribute the configuration.
3840

3841
    """
3842
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3843
    _RedistributeAncillaryFiles(self)
3844

    
3845

    
3846
class LUClusterActivateMasterIp(NoHooksLU):
3847
  """Activate the master IP on the master node.
3848

3849
  """
3850
  def Exec(self, feedback_fn):
3851
    """Activate the master IP.
3852

3853
    """
3854
    master = self.cfg.GetMasterNode()
3855
    self.rpc.call_node_activate_master_ip(master)
3856

    
3857

    
3858
class LUClusterDeactivateMasterIp(NoHooksLU):
3859
  """Deactivate the master IP on the master node.
3860

3861
  """
3862
  def Exec(self, feedback_fn):
3863
    """Deactivate the master IP.
3864

3865
    """
3866
    master = self.cfg.GetMasterNode()
3867
    self.rpc.call_node_deactivate_master_ip(master)
3868

    
3869

    
3870
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3871
  """Sleep and poll for an instance's disk to sync.
3872

3873
  """
3874
  if not instance.disks or disks is not None and not disks:
3875
    return True
3876

    
3877
  disks = _ExpandCheckDisks(instance, disks)
3878

    
3879
  if not oneshot:
3880
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3881

    
3882
  node = instance.primary_node
3883

    
3884
  for dev in disks:
3885
    lu.cfg.SetDiskID(dev, node)
3886

    
3887
  # TODO: Convert to utils.Retry
3888

    
3889
  retries = 0
3890
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3891
  while True:
3892
    max_time = 0
3893
    done = True
3894
    cumul_degraded = False
3895
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3896
    msg = rstats.fail_msg
3897
    if msg:
3898
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3899
      retries += 1
3900
      if retries >= 10:
3901
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3902
                                 " aborting." % node)
3903
      time.sleep(6)
3904
      continue
3905
    rstats = rstats.payload
3906
    retries = 0
3907
    for i, mstat in enumerate(rstats):
3908
      if mstat is None:
3909
        lu.LogWarning("Can't compute data for node %s/%s",
3910
                           node, disks[i].iv_name)
3911
        continue
3912

    
3913
      cumul_degraded = (cumul_degraded or
3914
                        (mstat.is_degraded and mstat.sync_percent is None))
3915
      if mstat.sync_percent is not None:
3916
        done = False
3917
        if mstat.estimated_time is not None:
3918
          rem_time = ("%s remaining (estimated)" %
3919
                      utils.FormatSeconds(mstat.estimated_time))
3920
          max_time = mstat.estimated_time
3921
        else:
3922
          rem_time = "no time estimate"
3923
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3924
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3925

    
3926
    # if we're done but degraded, let's do a few small retries, to
3927
    # make sure we see a stable and not transient situation; therefore
3928
    # we force restart of the loop
3929
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3930
      logging.info("Degraded disks found, %d retries left", degr_retries)
3931
      degr_retries -= 1
3932
      time.sleep(1)
3933
      continue
3934

    
3935
    if done or oneshot:
3936
      break
3937

    
3938
    time.sleep(min(60, max_time))
3939

    
3940
  if done:
3941
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3942
  return not cumul_degraded
3943

    
3944

    
3945
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3946
  """Check that mirrors are not degraded.
3947

3948
  The ldisk parameter, if True, will change the test from the
3949
  is_degraded attribute (which represents overall non-ok status for
3950
  the device(s)) to the ldisk (representing the local storage status).
3951

3952
  """
3953
  lu.cfg.SetDiskID(dev, node)
3954

    
3955
  result = True
3956

    
3957
  if on_primary or dev.AssembleOnSecondary():
3958
    rstats = lu.rpc.call_blockdev_find(node, dev)
3959
    msg = rstats.fail_msg
3960
    if msg:
3961
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3962
      result = False
3963
    elif not rstats.payload:
3964
      lu.LogWarning("Can't find disk on node %s", node)
3965
      result = False
3966
    else:
3967
      if ldisk:
3968
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3969
      else:
3970
        result = result and not rstats.payload.is_degraded
3971

    
3972
  if dev.children:
3973
    for child in dev.children:
3974
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3975

    
3976
  return result
3977

    
3978

    
3979
class LUOobCommand(NoHooksLU):
3980
  """Logical unit for OOB handling.
3981

3982
  """
3983
  REG_BGL = False
3984
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
3985

    
3986
  def ExpandNames(self):
3987
    """Gather locks we need.
3988

3989
    """
3990
    if self.op.node_names:
3991
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
3992
      lock_names = self.op.node_names
3993
    else:
3994
      lock_names = locking.ALL_SET
3995

    
3996
    self.needed_locks = {
3997
      locking.LEVEL_NODE: lock_names,
3998
      }
3999

    
4000
  def CheckPrereq(self):
4001
    """Check prerequisites.
4002

4003
    This checks:
4004
     - the node exists in the configuration
4005
     - OOB is supported
4006

4007
    Any errors are signaled by raising errors.OpPrereqError.
4008

4009
    """
4010
    self.nodes = []
4011
    self.master_node = self.cfg.GetMasterNode()
4012

    
4013
    assert self.op.power_delay >= 0.0
4014

    
4015
    if self.op.node_names:
4016
      if (self.op.command in self._SKIP_MASTER and
4017
          self.master_node in self.op.node_names):
4018
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
4019
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
4020

    
4021
        if master_oob_handler:
4022
          additional_text = ("run '%s %s %s' if you want to operate on the"
4023
                             " master regardless") % (master_oob_handler,
4024
                                                      self.op.command,
4025
                                                      self.master_node)
4026
        else:
4027
          additional_text = "it does not support out-of-band operations"
4028

    
4029
        raise errors.OpPrereqError(("Operating on the master node %s is not"
4030
                                    " allowed for %s; %s") %
4031
                                   (self.master_node, self.op.command,
4032
                                    additional_text), errors.ECODE_INVAL)
4033
    else:
4034
      self.op.node_names = self.cfg.GetNodeList()
4035
      if self.op.command in self._SKIP_MASTER:
4036
        self.op.node_names.remove(self.master_node)
4037

    
4038
    if self.op.command in self._SKIP_MASTER:
4039
      assert self.master_node not in self.op.node_names
4040

    
4041
    for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
4042
      if node is None:
4043
        raise errors.OpPrereqError("Node %s not found" % node_name,
4044
                                   errors.ECODE_NOENT)
4045
      else:
4046
        self.nodes.append(node)
4047

    
4048
      if (not self.op.ignore_status and
4049
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
4050
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
4051
                                    " not marked offline") % node_name,
4052
                                   errors.ECODE_STATE)
4053

    
4054
  def Exec(self, feedback_fn):
4055
    """Execute OOB and return result if we expect any.
4056

4057
    """
4058
    master_node = self.master_node
4059
    ret = []
4060

    
4061
    for idx, node in enumerate(utils.NiceSort(self.nodes,
4062
                                              key=lambda node: node.name)):
4063
      node_entry = [(constants.RS_NORMAL, node.name)]
4064
      ret.append(node_entry)
4065

    
4066
      oob_program = _SupportsOob(self.cfg, node)
4067

    
4068
      if not oob_program:
4069
        node_entry.append((constants.RS_UNAVAIL, None))
4070
        continue
4071

    
4072
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
4073
                   self.op.command, oob_program, node.name)
4074
      result = self.rpc.call_run_oob(master_node, oob_program,
4075
                                     self.op.command, node.name,
4076
                                     self.op.timeout)
4077

    
4078
      if result.fail_msg:
4079
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4080
                        node.name, result.fail_msg)
4081
        node_entry.append((constants.RS_NODATA, None))
4082
      else:
4083
        try:
4084
          self._CheckPayload(result)
4085
        except errors.OpExecError, err:
4086
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
4087
                          node.name, err)
4088
          node_entry.append((constants.RS_NODATA, None))
4089
        else:
4090
          if self.op.command == constants.OOB_HEALTH:
4091
            # For health we should log important events
4092
            for item, status in result.payload:
4093
              if status in [constants.OOB_STATUS_WARNING,
4094
                            constants.OOB_STATUS_CRITICAL]: