Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 0ad1ea40

History | View | Annotate | Download (474.6 kB)

1
#
2
#
3

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

    
21

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

    
24
# pylint: disable=W0201,C0302
25

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

    
29
# C0302: since we have waaaay 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_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
    # Define functions determining which nodes to consider for a file
2112
    files2nodefn = [
2113
      (files_all, None),
2114
      (files_mc, lambda node: (node.master_candidate or
2115
                               node.name == master_node)),
2116
      (files_vm, lambda node: node.vm_capable),
2117
      ]
2118

    
2119
    # Build mapping from filename to list of nodes which should have the file
2120
    nodefiles = {}
2121
    for (files, fn) in files2nodefn:
2122
      if fn is None:
2123
        filenodes = nodeinfo
2124
      else:
2125
        filenodes = filter(fn, nodeinfo)
2126
      nodefiles.update((filename,
2127
                        frozenset(map(operator.attrgetter("name"), filenodes)))
2128
                       for filename in files)
2129

    
2130
    assert set(nodefiles) == (files_all | files_mc | files_vm)
2131

    
2132
    fileinfo = dict((filename, {}) for filename in nodefiles)
2133
    ignore_nodes = set()
2134

    
2135
    for node in nodeinfo:
2136
      if node.offline:
2137
        ignore_nodes.add(node.name)
2138
        continue
2139

    
2140
      nresult = all_nvinfo[node.name]
2141

    
2142
      if nresult.fail_msg or not nresult.payload:
2143
        node_files = None
2144
      else:
2145
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
2146

    
2147
      test = not (node_files and isinstance(node_files, dict))
2148
      errorif(test, cls.ENODEFILECHECK, node.name,
2149
              "Node did not return file checksum data")
2150
      if test:
2151
        ignore_nodes.add(node.name)
2152
        continue
2153

    
2154
      # Build per-checksum mapping from filename to nodes having it
2155
      for (filename, checksum) in node_files.items():
2156
        assert filename in nodefiles
2157
        fileinfo[filename].setdefault(checksum, set()).add(node.name)
2158

    
2159
    for (filename, checksums) in fileinfo.items():
2160
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2161

    
2162
      # Nodes having the file
2163
      with_file = frozenset(node_name
2164
                            for nodes in fileinfo[filename].values()
2165
                            for node_name in nodes) - ignore_nodes
2166

    
2167
      expected_nodes = nodefiles[filename] - ignore_nodes
2168

    
2169
      # Nodes missing file
2170
      missing_file = expected_nodes - with_file
2171

    
2172
      if filename in files_opt:
2173
        # All or no nodes
2174
        errorif(missing_file and missing_file != expected_nodes,
2175
                cls.ECLUSTERFILECHECK, None,
2176
                "File %s is optional, but it must exist on all or no"
2177
                " nodes (not found on %s)",
2178
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2179
      else:
2180
        # Non-optional files
2181
        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
2182
                "File %s is missing from node(s) %s", filename,
2183
                utils.CommaJoin(utils.NiceSort(missing_file)))
2184

    
2185
        # Warn if a node has a file it shouldn't
2186
        unexpected = with_file - expected_nodes
2187
        errorif(unexpected,
2188
                cls.ECLUSTERFILECHECK, None,
2189
                "File %s should not exist on node(s) %s",
2190
                filename, utils.CommaJoin(utils.NiceSort(unexpected)))
2191

    
2192
      # See if there are multiple versions of the file
2193
      test = len(checksums) > 1
2194
      if test:
2195
        variants = ["variant %s on %s" %
2196
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2197
                    for (idx, (checksum, nodes)) in
2198
                      enumerate(sorted(checksums.items()))]
2199
      else:
2200
        variants = []
2201

    
2202
      errorif(test, cls.ECLUSTERFILECHECK, None,
2203
              "File %s found with %s different checksums (%s)",
2204
              filename, len(checksums), "; ".join(variants))
2205

    
2206
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2207
                      drbd_map):
2208
    """Verifies and the node DRBD status.
2209

2210
    @type ninfo: L{objects.Node}
2211
    @param ninfo: the node to check
2212
    @param nresult: the remote results for the node
2213
    @param instanceinfo: the dict of instances
2214
    @param drbd_helper: the configured DRBD usermode helper
2215
    @param drbd_map: the DRBD map as returned by
2216
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2217

2218
    """
2219
    node = ninfo.name
2220
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2221

    
2222
    if drbd_helper:
2223
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2224
      test = (helper_result == None)
2225
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
2226
               "no drbd usermode helper returned")
2227
      if helper_result:
2228
        status, payload = helper_result
2229
        test = not status
2230
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2231
                 "drbd usermode helper check unsuccessful: %s", payload)
2232
        test = status and (payload != drbd_helper)
2233
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2234
                 "wrong drbd usermode helper: %s", payload)
2235

    
2236
    # compute the DRBD minors
2237
    node_drbd = {}
2238
    for minor, instance in drbd_map[node].items():
2239
      test = instance not in instanceinfo
2240
      _ErrorIf(test, self.ECLUSTERCFG, None,
2241
               "ghost instance '%s' in temporary DRBD map", instance)
2242
        # ghost instance should not be running, but otherwise we
2243
        # don't give double warnings (both ghost instance and
2244
        # unallocated minor in use)
2245
      if test:
2246
        node_drbd[minor] = (instance, False)
2247
      else:
2248
        instance = instanceinfo[instance]
2249
        node_drbd[minor] = (instance.name, instance.admin_up)
2250

    
2251
    # and now check them
2252
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2253
    test = not isinstance(used_minors, (tuple, list))
2254
    _ErrorIf(test, self.ENODEDRBD, node,
2255
             "cannot parse drbd status file: %s", str(used_minors))
2256
    if test:
2257
      # we cannot check drbd status
2258
      return
2259

    
2260
    for minor, (iname, must_exist) in node_drbd.items():
2261
      test = minor not in used_minors and must_exist
2262
      _ErrorIf(test, self.ENODEDRBD, node,
2263
               "drbd minor %d of instance %s is not active", minor, iname)
2264
    for minor in used_minors:
2265
      test = minor not in node_drbd
2266
      _ErrorIf(test, self.ENODEDRBD, node,
2267
               "unallocated drbd minor %d is in use", minor)
2268

    
2269
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2270
    """Builds the node OS structures.
2271

2272
    @type ninfo: L{objects.Node}
2273
    @param ninfo: the node to check
2274
    @param nresult: the remote results for the node
2275
    @param nimg: the node image object
2276

2277
    """
2278
    node = ninfo.name
2279
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2280

    
2281
    remote_os = nresult.get(constants.NV_OSLIST, None)
2282
    test = (not isinstance(remote_os, list) or
2283
            not compat.all(isinstance(v, list) and len(v) == 7
2284
                           for v in remote_os))
2285

    
2286
    _ErrorIf(test, self.ENODEOS, node,
2287
             "node hasn't returned valid OS data")
2288

    
2289
    nimg.os_fail = test
2290

    
2291
    if test:
2292
      return
2293

    
2294
    os_dict = {}
2295

    
2296
    for (name, os_path, status, diagnose,
2297
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2298

    
2299
      if name not in os_dict:
2300
        os_dict[name] = []
2301

    
2302
      # parameters is a list of lists instead of list of tuples due to
2303
      # JSON lacking a real tuple type, fix it:
2304
      parameters = [tuple(v) for v in parameters]
2305
      os_dict[name].append((os_path, status, diagnose,
2306
                            set(variants), set(parameters), set(api_ver)))
2307

    
2308
    nimg.oslist = os_dict
2309

    
2310
  def _VerifyNodeOS(self, ninfo, nimg, base):
2311
    """Verifies the node OS list.
2312

2313
    @type ninfo: L{objects.Node}
2314
    @param ninfo: the node to check
2315
    @param nimg: the node image object
2316
    @param base: the 'template' node we match against (e.g. from the master)
2317

2318
    """
2319
    node = ninfo.name
2320
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2321

    
2322
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2323

    
2324
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2325
    for os_name, os_data in nimg.oslist.items():
2326
      assert os_data, "Empty OS status for OS %s?!" % os_name
2327
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2328
      _ErrorIf(not f_status, self.ENODEOS, node,
2329
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2330
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
2331
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2332
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2333
      # comparisons with the 'base' image
2334
      test = os_name not in base.oslist
2335
      _ErrorIf(test, self.ENODEOS, node,
2336
               "Extra OS %s not present on reference node (%s)",
2337
               os_name, base.name)
2338
      if test:
2339
        continue
2340
      assert base.oslist[os_name], "Base node has empty OS status?"
2341
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2342
      if not b_status:
2343
        # base OS is invalid, skipping
2344
        continue
2345
      for kind, a, b in [("API version", f_api, b_api),
2346
                         ("variants list", f_var, b_var),
2347
                         ("parameters", beautify_params(f_param),
2348
                          beautify_params(b_param))]:
2349
        _ErrorIf(a != b, self.ENODEOS, node,
2350
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2351
                 kind, os_name, base.name,
2352
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2353

    
2354
    # check any missing OSes
2355
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2356
    _ErrorIf(missing, self.ENODEOS, node,
2357
             "OSes present on reference node %s but missing on this node: %s",
2358
             base.name, utils.CommaJoin(missing))
2359

    
2360
  def _VerifyOob(self, ninfo, nresult):
2361
    """Verifies out of band functionality of a node.
2362

2363
    @type ninfo: L{objects.Node}
2364
    @param ninfo: the node to check
2365
    @param nresult: the remote results for the node
2366

2367
    """
2368
    node = ninfo.name
2369
    # We just have to verify the paths on master and/or master candidates
2370
    # as the oob helper is invoked on the master
2371
    if ((ninfo.master_candidate or ninfo.master_capable) and
2372
        constants.NV_OOB_PATHS in nresult):
2373
      for path_result in nresult[constants.NV_OOB_PATHS]:
2374
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
2375

    
2376
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2377
    """Verifies and updates the node volume data.
2378

2379
    This function will update a L{NodeImage}'s internal structures
2380
    with data from the remote call.
2381

2382
    @type ninfo: L{objects.Node}
2383
    @param ninfo: the node to check
2384
    @param nresult: the remote results for the node
2385
    @param nimg: the node image object
2386
    @param vg_name: the configured VG name
2387

2388
    """
2389
    node = ninfo.name
2390
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2391

    
2392
    nimg.lvm_fail = True
2393
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2394
    if vg_name is None:
2395
      pass
2396
    elif isinstance(lvdata, basestring):
2397
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
2398
               utils.SafeEncode(lvdata))
2399
    elif not isinstance(lvdata, dict):
2400
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
2401
    else:
2402
      nimg.volumes = lvdata
2403
      nimg.lvm_fail = False
2404

    
2405
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2406
    """Verifies and updates the node instance list.
2407

2408
    If the listing was successful, then updates this node's instance
2409
    list. Otherwise, it marks the RPC call as failed for the instance
2410
    list key.
2411

2412
    @type ninfo: L{objects.Node}
2413
    @param ninfo: the node to check
2414
    @param nresult: the remote results for the node
2415
    @param nimg: the node image object
2416

2417
    """
2418
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2419
    test = not isinstance(idata, list)
2420
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
2421
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
2422
    if test:
2423
      nimg.hyp_fail = True
2424
    else:
2425
      nimg.instances = idata
2426

    
2427
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2428
    """Verifies and computes a node information map
2429

2430
    @type ninfo: L{objects.Node}
2431
    @param ninfo: the node to check
2432
    @param nresult: the remote results for the node
2433
    @param nimg: the node image object
2434
    @param vg_name: the configured VG name
2435

2436
    """
2437
    node = ninfo.name
2438
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2439

    
2440
    # try to read free memory (from the hypervisor)
2441
    hv_info = nresult.get(constants.NV_HVINFO, None)
2442
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2443
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
2444
    if not test:
2445
      try:
2446
        nimg.mfree = int(hv_info["memory_free"])
2447
      except (ValueError, TypeError):
2448
        _ErrorIf(True, self.ENODERPC, node,
2449
                 "node returned invalid nodeinfo, check hypervisor")
2450

    
2451
    # FIXME: devise a free space model for file based instances as well
2452
    if vg_name is not None:
2453
      test = (constants.NV_VGLIST not in nresult or
2454
              vg_name not in nresult[constants.NV_VGLIST])
2455
      _ErrorIf(test, self.ENODELVM, node,
2456
               "node didn't return data for the volume group '%s'"
2457
               " - it is either missing or broken", vg_name)
2458
      if not test:
2459
        try:
2460
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2461
        except (ValueError, TypeError):
2462
          _ErrorIf(True, self.ENODERPC, node,
2463
                   "node returned invalid LVM info, check LVM status")
2464

    
2465
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2466
    """Gets per-disk status information for all instances.
2467

2468
    @type nodelist: list of strings
2469
    @param nodelist: Node names
2470
    @type node_image: dict of (name, L{objects.Node})
2471
    @param node_image: Node objects
2472
    @type instanceinfo: dict of (name, L{objects.Instance})
2473
    @param instanceinfo: Instance objects
2474
    @rtype: {instance: {node: [(succes, payload)]}}
2475
    @return: a dictionary of per-instance dictionaries with nodes as
2476
        keys and disk information as values; the disk information is a
2477
        list of tuples (success, payload)
2478

2479
    """
2480
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2481

    
2482
    node_disks = {}
2483
    node_disks_devonly = {}
2484
    diskless_instances = set()
2485
    diskless = constants.DT_DISKLESS
2486

    
2487
    for nname in nodelist:
2488
      node_instances = list(itertools.chain(node_image[nname].pinst,
2489
                                            node_image[nname].sinst))
2490
      diskless_instances.update(inst for inst in node_instances
2491
                                if instanceinfo[inst].disk_template == diskless)
2492
      disks = [(inst, disk)
2493
               for inst in node_instances
2494
               for disk in instanceinfo[inst].disks]
2495

    
2496
      if not disks:
2497
        # No need to collect data
2498
        continue
2499

    
2500
      node_disks[nname] = disks
2501

    
2502
      # Creating copies as SetDiskID below will modify the objects and that can
2503
      # lead to incorrect data returned from nodes
2504
      devonly = [dev.Copy() for (_, dev) in disks]
2505

    
2506
      for dev in devonly:
2507
        self.cfg.SetDiskID(dev, nname)
2508

    
2509
      node_disks_devonly[nname] = devonly
2510

    
2511
    assert len(node_disks) == len(node_disks_devonly)
2512

    
2513
    # Collect data from all nodes with disks
2514
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2515
                                                          node_disks_devonly)
2516

    
2517
    assert len(result) == len(node_disks)
2518

    
2519
    instdisk = {}
2520

    
2521
    for (nname, nres) in result.items():
2522
      disks = node_disks[nname]
2523

    
2524
      if nres.offline:
2525
        # No data from this node
2526
        data = len(disks) * [(False, "node offline")]
2527
      else:
2528
        msg = nres.fail_msg
2529
        _ErrorIf(msg, self.ENODERPC, nname,
2530
                 "while getting disk information: %s", msg)
2531
        if msg:
2532
          # No data from this node
2533
          data = len(disks) * [(False, msg)]
2534
        else:
2535
          data = []
2536
          for idx, i in enumerate(nres.payload):
2537
            if isinstance(i, (tuple, list)) and len(i) == 2:
2538
              data.append(i)
2539
            else:
2540
              logging.warning("Invalid result from node %s, entry %d: %s",
2541
                              nname, idx, i)
2542
              data.append((False, "Invalid result from the remote node"))
2543

    
2544
      for ((inst, _), status) in zip(disks, data):
2545
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2546

    
2547
    # Add empty entries for diskless instances.
2548
    for inst in diskless_instances:
2549
      assert inst not in instdisk
2550
      instdisk[inst] = {}
2551

    
2552
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2553
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2554
                      compat.all(isinstance(s, (tuple, list)) and
2555
                                 len(s) == 2 for s in statuses)
2556
                      for inst, nnames in instdisk.items()
2557
                      for nname, statuses in nnames.items())
2558
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2559

    
2560
    return instdisk
2561

    
2562
  @staticmethod
2563
  def _SshNodeSelector(group_uuid, all_nodes):
2564
    """Create endless iterators for all potential SSH check hosts.
2565

2566
    """
2567
    nodes = [node for node in all_nodes
2568
             if (node.group != group_uuid and
2569
                 not node.offline)]
2570
    keyfunc = operator.attrgetter("group")
2571

    
2572
    return map(itertools.cycle,
2573
               [sorted(map(operator.attrgetter("name"), names))
2574
                for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
2575
                                                  keyfunc)])
2576

    
2577
  @classmethod
2578
  def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
2579
    """Choose which nodes should talk to which other nodes.
2580

2581
    We will make nodes contact all nodes in their group, and one node from
2582
    every other group.
2583

2584
    @warning: This algorithm has a known issue if one node group is much
2585
      smaller than others (e.g. just one node). In such a case all other
2586
      nodes will talk to the single node.
2587

2588
    """
2589
    online_nodes = sorted(node.name for node in group_nodes if not node.offline)
2590
    sel = cls._SshNodeSelector(group_uuid, all_nodes)
2591

    
2592
    return (online_nodes,
2593
            dict((name, sorted([i.next() for i in sel]))
2594
                 for name in online_nodes))
2595

    
2596
  def BuildHooksEnv(self):
2597
    """Build hooks env.
2598

2599
    Cluster-Verify hooks just ran in the post phase and their failure makes
2600
    the output be logged in the verify output and the verification to fail.
2601

2602
    """
2603
    env = {
2604
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2605
      }
2606

    
2607
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2608
               for node in self.my_node_info.values())
2609

    
2610
    return env
2611

    
2612
  def BuildHooksNodes(self):
2613
    """Build hooks nodes.
2614

2615
    """
2616
    return ([], self.my_node_names)
2617

    
2618
  def Exec(self, feedback_fn):
2619
    """Verify integrity of the node group, performing various test on nodes.
2620

2621
    """
2622
    # This method has too many local variables. pylint: disable=R0914
2623
    feedback_fn("* Verifying group '%s'" % self.group_info.name)
2624

    
2625
    if not self.my_node_names:
2626
      # empty node group
2627
      feedback_fn("* Empty node group, skipping verification")
2628
      return True
2629

    
2630
    self.bad = False
2631
    _ErrorIf = self._ErrorIf # pylint: disable=C0103
2632
    verbose = self.op.verbose
2633
    self._feedback_fn = feedback_fn
2634

    
2635
    vg_name = self.cfg.GetVGName()
2636
    drbd_helper = self.cfg.GetDRBDHelper()
2637
    cluster = self.cfg.GetClusterInfo()
2638
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2639
    hypervisors = cluster.enabled_hypervisors
2640
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2641

    
2642
    i_non_redundant = [] # Non redundant instances
2643
    i_non_a_balanced = [] # Non auto-balanced instances
2644
    n_offline = 0 # Count of offline nodes
2645
    n_drained = 0 # Count of nodes being drained
2646
    node_vol_should = {}
2647

    
2648
    # FIXME: verify OS list
2649

    
2650
    # File verification
2651
    filemap = _ComputeAncillaryFiles(cluster, False)
2652

    
2653
    # do local checksums
2654
    master_node = self.master_node = self.cfg.GetMasterNode()
2655
    master_ip = self.cfg.GetMasterIP()
2656

    
2657
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2658

    
2659
    node_verify_param = {
2660
      constants.NV_FILELIST:
2661
        utils.UniqueSequence(filename
2662
                             for files in filemap
2663
                             for filename in files),
2664
      constants.NV_NODELIST:
2665
        self._SelectSshCheckNodes(node_data_list, self.group_uuid,
2666
                                  self.all_node_info.values()),
2667
      constants.NV_HYPERVISOR: hypervisors,
2668
      constants.NV_HVPARAMS:
2669
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2670
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2671
                                 for node in node_data_list
2672
                                 if not node.offline],
2673
      constants.NV_INSTANCELIST: hypervisors,
2674
      constants.NV_VERSION: None,
2675
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2676
      constants.NV_NODESETUP: None,
2677
      constants.NV_TIME: None,
2678
      constants.NV_MASTERIP: (master_node, master_ip),
2679
      constants.NV_OSLIST: None,
2680
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2681
      }
2682

    
2683
    if vg_name is not None:
2684
      node_verify_param[constants.NV_VGLIST] = None
2685
      node_verify_param[constants.NV_LVLIST] = vg_name
2686
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2687
      node_verify_param[constants.NV_DRBDLIST] = None
2688

    
2689
    if drbd_helper:
2690
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2691

    
2692
    # bridge checks
2693
    # FIXME: this needs to be changed per node-group, not cluster-wide
2694
    bridges = set()
2695
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2696
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2697
      bridges.add(default_nicpp[constants.NIC_LINK])
2698
    for instance in self.my_inst_info.values():
2699
      for nic in instance.nics:
2700
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2701
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2702
          bridges.add(full_nic[constants.NIC_LINK])
2703

    
2704
    if bridges:
2705
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2706

    
2707
    # Build our expected cluster state
2708
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2709
                                                 name=node.name,
2710
                                                 vm_capable=node.vm_capable))
2711
                      for node in node_data_list)
2712

    
2713
    # Gather OOB paths
2714
    oob_paths = []
2715
    for node in self.all_node_info.values():
2716
      path = _SupportsOob(self.cfg, node)
2717
      if path and path not in oob_paths:
2718
        oob_paths.append(path)
2719

    
2720
    if oob_paths:
2721
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2722

    
2723
    for instance in self.my_inst_names:
2724
      inst_config = self.my_inst_info[instance]
2725

    
2726
      for nname in inst_config.all_nodes:
2727
        if nname not in node_image:
2728
          gnode = self.NodeImage(name=nname)
2729
          gnode.ghost = (nname not in self.all_node_info)
2730
          node_image[nname] = gnode
2731

    
2732
      inst_config.MapLVsByNode(node_vol_should)
2733

    
2734
      pnode = inst_config.primary_node
2735
      node_image[pnode].pinst.append(instance)
2736

    
2737
      for snode in inst_config.secondary_nodes:
2738
        nimg = node_image[snode]
2739
        nimg.sinst.append(instance)
2740
        if pnode not in nimg.sbp:
2741
          nimg.sbp[pnode] = []
2742
        nimg.sbp[pnode].append(instance)
2743

    
2744
    # At this point, we have the in-memory data structures complete,
2745
    # except for the runtime information, which we'll gather next
2746

    
2747
    # Due to the way our RPC system works, exact response times cannot be
2748
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2749
    # time before and after executing the request, we can at least have a time
2750
    # window.
2751
    nvinfo_starttime = time.time()
2752
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2753
                                           node_verify_param,
2754
                                           self.cfg.GetClusterName())
2755
    nvinfo_endtime = time.time()
2756

    
2757
    if self.extra_lv_nodes and vg_name is not None:
2758
      extra_lv_nvinfo = \
2759
          self.rpc.call_node_verify(self.extra_lv_nodes,
2760
                                    {constants.NV_LVLIST: vg_name},
2761
                                    self.cfg.GetClusterName())
2762
    else:
2763
      extra_lv_nvinfo = {}
2764

    
2765
    all_drbd_map = self.cfg.ComputeDRBDMap()
2766

    
2767
    feedback_fn("* Gathering disk information (%s nodes)" %
2768
                len(self.my_node_names))
2769
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2770
                                     self.my_inst_info)
2771

    
2772
    feedback_fn("* Verifying configuration file consistency")
2773

    
2774
    # If not all nodes are being checked, we need to make sure the master node
2775
    # and a non-checked vm_capable node are in the list.
2776
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2777
    if absent_nodes:
2778
      vf_nvinfo = all_nvinfo.copy()
2779
      vf_node_info = list(self.my_node_info.values())
2780
      additional_nodes = []
2781
      if master_node not in self.my_node_info:
2782
        additional_nodes.append(master_node)
2783
        vf_node_info.append(self.all_node_info[master_node])
2784
      # Add the first vm_capable node we find which is not included
2785
      for node in absent_nodes:
2786
        nodeinfo = self.all_node_info[node]
2787
        if nodeinfo.vm_capable and not nodeinfo.offline:
2788
          additional_nodes.append(node)
2789
          vf_node_info.append(self.all_node_info[node])
2790
          break
2791
      key = constants.NV_FILELIST
2792
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2793
                                                 {key: node_verify_param[key]},
2794
                                                 self.cfg.GetClusterName()))
2795
    else:
2796
      vf_nvinfo = all_nvinfo
2797
      vf_node_info = self.my_node_info.values()
2798

    
2799
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2800

    
2801
    feedback_fn("* Verifying node status")
2802

    
2803
    refos_img = None
2804

    
2805
    for node_i in node_data_list:
2806
      node = node_i.name
2807
      nimg = node_image[node]
2808

    
2809
      if node_i.offline:
2810
        if verbose:
2811
          feedback_fn("* Skipping offline node %s" % (node,))
2812
        n_offline += 1
2813
        continue
2814

    
2815
      if node == master_node:
2816
        ntype = "master"
2817
      elif node_i.master_candidate:
2818
        ntype = "master candidate"
2819
      elif node_i.drained:
2820
        ntype = "drained"
2821
        n_drained += 1
2822
      else:
2823
        ntype = "regular"
2824
      if verbose:
2825
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2826

    
2827
      msg = all_nvinfo[node].fail_msg
2828
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2829
      if msg:
2830
        nimg.rpc_fail = True
2831
        continue
2832

    
2833
      nresult = all_nvinfo[node].payload
2834

    
2835
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2836
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2837
      self._VerifyNodeNetwork(node_i, nresult)
2838
      self._VerifyOob(node_i, nresult)
2839

    
2840
      if nimg.vm_capable:
2841
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2842
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2843
                             all_drbd_map)
2844

    
2845
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2846
        self._UpdateNodeInstances(node_i, nresult, nimg)
2847
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2848
        self._UpdateNodeOS(node_i, nresult, nimg)
2849

    
2850
        if not nimg.os_fail:
2851
          if refos_img is None:
2852
            refos_img = nimg
2853
          self._VerifyNodeOS(node_i, nimg, refos_img)
2854
        self._VerifyNodeBridges(node_i, nresult, bridges)
2855

    
2856
        # Check whether all running instancies are primary for the node. (This
2857
        # can no longer be done from _VerifyInstance below, since some of the
2858
        # wrong instances could be from other node groups.)
2859
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2860

    
2861
        for inst in non_primary_inst:
2862
          test = inst in self.all_inst_info
2863
          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
2864
                   "instance should not run on node %s", node_i.name)
2865
          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
2866
                   "node is running unknown instance %s", inst)
2867

    
2868
    for node, result in extra_lv_nvinfo.items():
2869
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
2870
                              node_image[node], vg_name)
2871

    
2872
    feedback_fn("* Verifying instance status")
2873
    for instance in self.my_inst_names:
2874
      if verbose:
2875
        feedback_fn("* Verifying instance %s" % instance)
2876
      inst_config = self.my_inst_info[instance]
2877
      self._VerifyInstance(instance, inst_config, node_image,
2878
                           instdisk[instance])
2879
      inst_nodes_offline = []
2880

    
2881
      pnode = inst_config.primary_node
2882
      pnode_img = node_image[pnode]
2883
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2884
               self.ENODERPC, pnode, "instance %s, connection to"
2885
               " primary node failed", instance)
2886

    
2887
      _ErrorIf(inst_config.admin_up and pnode_img.offline,
2888
               self.EINSTANCEBADNODE, instance,
2889
               "instance is marked as running and lives on offline node %s",
2890
               inst_config.primary_node)
2891

    
2892
      # If the instance is non-redundant we cannot survive losing its primary
2893
      # node, so we are not N+1 compliant. On the other hand we have no disk
2894
      # templates with more than one secondary so that situation is not well
2895
      # supported either.
2896
      # FIXME: does not support file-backed instances
2897
      if not inst_config.secondary_nodes:
2898
        i_non_redundant.append(instance)
2899

    
2900
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2901
               instance, "instance has multiple secondary nodes: %s",
2902
               utils.CommaJoin(inst_config.secondary_nodes),
2903
               code=self.ETYPE_WARNING)
2904

    
2905
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2906
        pnode = inst_config.primary_node
2907
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2908
        instance_groups = {}
2909

    
2910
        for node in instance_nodes:
2911
          instance_groups.setdefault(self.all_node_info[node].group,
2912
                                     []).append(node)
2913

    
2914
        pretty_list = [
2915
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2916
          # Sort so that we always list the primary node first.
2917
          for group, nodes in sorted(instance_groups.items(),
2918
                                     key=lambda (_, nodes): pnode in nodes,
2919
                                     reverse=True)]
2920

    
2921
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2922
                      instance, "instance has primary and secondary nodes in"
2923
                      " different groups: %s", utils.CommaJoin(pretty_list),
2924
                      code=self.ETYPE_WARNING)
2925

    
2926
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2927
        i_non_a_balanced.append(instance)
2928

    
2929
      for snode in inst_config.secondary_nodes:
2930
        s_img = node_image[snode]
2931
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2932
                 "instance %s, connection to secondary node failed", instance)
2933

    
2934
        if s_img.offline:
2935
          inst_nodes_offline.append(snode)
2936

    
2937
      # warn that the instance lives on offline nodes
2938
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2939
               "instance has offline secondary node(s) %s",
2940
               utils.CommaJoin(inst_nodes_offline))
2941
      # ... or ghost/non-vm_capable nodes
2942
      for node in inst_config.all_nodes:
2943
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2944
                 "instance lives on ghost node %s", node)
2945
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2946
                 instance, "instance lives on non-vm_capable node %s", node)
2947

    
2948
    feedback_fn("* Verifying orphan volumes")
2949
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2950

    
2951
    # We will get spurious "unknown volume" warnings if any node of this group
2952
    # is secondary for an instance whose primary is in another group. To avoid
2953
    # them, we find these instances and add their volumes to node_vol_should.
2954
    for inst in self.all_inst_info.values():
2955
      for secondary in inst.secondary_nodes:
2956
        if (secondary in self.my_node_info
2957
            and inst.name not in self.my_inst_info):
2958
          inst.MapLVsByNode(node_vol_should)
2959
          break
2960

    
2961
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2962

    
2963
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2964
      feedback_fn("* Verifying N+1 Memory redundancy")
2965
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
2966

    
2967
    feedback_fn("* Other Notes")
2968
    if i_non_redundant:
2969
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2970
                  % len(i_non_redundant))
2971

    
2972
    if i_non_a_balanced:
2973
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2974
                  % len(i_non_a_balanced))
2975

    
2976
    if n_offline:
2977
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2978

    
2979
    if n_drained:
2980
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2981

    
2982
    return not self.bad
2983

    
2984
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2985
    """Analyze the post-hooks' result
2986

2987
    This method analyses the hook result, handles it, and sends some
2988
    nicely-formatted feedback back to the user.
2989

2990
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
2991
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
2992
    @param hooks_results: the results of the multi-node hooks rpc call
2993
    @param feedback_fn: function used send feedback back to the caller
2994
    @param lu_result: previous Exec result
2995
    @return: the new Exec result, based on the previous result
2996
        and hook results
2997

2998
    """
2999
    # We only really run POST phase hooks, only for non-empty groups,
3000
    # and are only interested in their results
3001
    if not self.my_node_names:
3002
      # empty node group
3003
      pass
3004
    elif phase == constants.HOOKS_PHASE_POST:
3005
      # Used to change hooks' output to proper indentation
3006
      feedback_fn("* Hooks Results")
3007
      assert hooks_results, "invalid result from hooks"
3008

    
3009
      for node_name in hooks_results:
3010
        res = hooks_results[node_name]
3011
        msg = res.fail_msg
3012
        test = msg and not res.offline
3013
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
3014
                      "Communication failure in hooks execution: %s", msg)
3015
        if res.offline or msg:
3016
          # No need to investigate payload if node is offline or gave
3017
          # an error.
3018
          continue
3019
        for script, hkr, output in res.payload:
3020
          test = hkr == constants.HKR_FAIL
3021
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
3022
                        "Script %s failed, output:", script)
3023
          if test:
3024
            output = self._HOOKS_INDENT_RE.sub("      ", output)
3025
            feedback_fn("%s" % output)
3026
            lu_result = False
3027

    
3028
    return lu_result
3029

    
3030

    
3031
class LUClusterVerifyDisks(NoHooksLU):
3032
  """Verifies the cluster disks status.
3033

3034
  """
3035
  REQ_BGL = False
3036

    
3037
  def ExpandNames(self):
3038
    self.share_locks = _ShareAll()
3039
    self.needed_locks = {
3040
      locking.LEVEL_NODEGROUP: locking.ALL_SET,
3041
      }
3042

    
3043
  def Exec(self, feedback_fn):
3044
    group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3045

    
3046
    # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3047
    return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3048
                           for group in group_names])
3049

    
3050

    
3051
class LUGroupVerifyDisks(NoHooksLU):
3052
  """Verifies the status of all disks in a node group.
3053

3054
  """
3055
  REQ_BGL = False
3056

    
3057
  def ExpandNames(self):
3058
    # Raises errors.OpPrereqError on its own if group can't be found
3059
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3060

    
3061
    self.share_locks = _ShareAll()
3062
    self.needed_locks = {
3063
      locking.LEVEL_INSTANCE: [],
3064
      locking.LEVEL_NODEGROUP: [],
3065
      locking.LEVEL_NODE: [],
3066
      }
3067

    
3068
  def DeclareLocks(self, level):
3069
    if level == locking.LEVEL_INSTANCE:
3070
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
3071

    
3072
      # Lock instances optimistically, needs verification once node and group
3073
      # locks have been acquired
3074
      self.needed_locks[locking.LEVEL_INSTANCE] = \
3075
        self.cfg.GetNodeGroupInstances(self.group_uuid)
3076

    
3077
    elif level == locking.LEVEL_NODEGROUP:
3078
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3079

    
3080
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
3081
        set([self.group_uuid] +
3082
            # Lock all groups used by instances optimistically; this requires
3083
            # going via the node before it's locked, requiring verification
3084
            # later on
3085
            [group_uuid
3086
             for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3087
             for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3088

    
3089
    elif level == locking.LEVEL_NODE:
3090
      # This will only lock the nodes in the group to be verified which contain
3091
      # actual instances
3092
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3093
      self._LockInstancesNodes()
3094

    
3095
      # Lock all nodes in group to be verified
3096
      assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3097
      member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3098
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3099

    
3100
  def CheckPrereq(self):
3101
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3102
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3103
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3104

    
3105
    assert self.group_uuid in owned_groups
3106

    
3107
    # Check if locked instances are still correct
3108
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3109

    
3110
    # Get instance information
3111
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3112

    
3113
    # Check if node groups for locked instances are still correct
3114
    for (instance_name, inst) in self.instances.items():
3115
      assert owned_nodes.issuperset(inst.all_nodes), \
3116
        "Instance %s's nodes changed while we kept the lock" % instance_name
3117

    
3118
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
3119
                                             owned_groups)
3120

    
3121
      assert self.group_uuid in inst_groups, \
3122
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
3123

    
3124
  def Exec(self, feedback_fn):
3125
    """Verify integrity of cluster disks.
3126

3127
    @rtype: tuple of three items
3128
    @return: a tuple of (dict of node-to-node_error, list of instances
3129
        which need activate-disks, dict of instance: (node, volume) for
3130
        missing volumes
3131

3132
    """
3133
    res_nodes = {}
3134
    res_instances = set()
3135
    res_missing = {}
3136

    
3137
    nv_dict = _MapInstanceDisksToNodes([inst
3138
                                        for inst in self.instances.values()
3139
                                        if inst.admin_up])
3140

    
3141
    if nv_dict:
3142
      nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3143
                             set(self.cfg.GetVmCapableNodeList()))
3144

    
3145
      node_lvs = self.rpc.call_lv_list(nodes, [])
3146

    
3147
      for (node, node_res) in node_lvs.items():
3148
        if node_res.offline:
3149
          continue
3150

    
3151
        msg = node_res.fail_msg
3152
        if msg:
3153
          logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3154
          res_nodes[node] = msg
3155
          continue
3156

    
3157
        for lv_name, (_, _, lv_online) in node_res.payload.items():
3158
          inst = nv_dict.pop((node, lv_name), None)
3159
          if not (lv_online or inst is None):
3160
            res_instances.add(inst)
3161

    
3162
      # any leftover items in nv_dict are missing LVs, let's arrange the data
3163
      # better
3164
      for key, inst in nv_dict.iteritems():
3165
        res_missing.setdefault(inst, []).append(key)
3166

    
3167
    return (res_nodes, list(res_instances), res_missing)
3168

    
3169

    
3170
class LUClusterRepairDiskSizes(NoHooksLU):
3171
  """Verifies the cluster disks sizes.
3172

3173
  """
3174
  REQ_BGL = False
3175

    
3176
  def ExpandNames(self):
3177
    if self.op.instances:
3178
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
3179
      self.needed_locks = {
3180
        locking.LEVEL_NODE: [],
3181
        locking.LEVEL_INSTANCE: self.wanted_names,
3182
        }
3183
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3184
    else:
3185
      self.wanted_names = None
3186
      self.needed_locks = {
3187
        locking.LEVEL_NODE: locking.ALL_SET,
3188
        locking.LEVEL_INSTANCE: locking.ALL_SET,
3189
        }
3190
    self.share_locks = _ShareAll()
3191

    
3192
  def DeclareLocks(self, level):
3193
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
3194
      self._LockInstancesNodes(primary_only=True)
3195

    
3196
  def CheckPrereq(self):
3197
    """Check prerequisites.
3198

3199
    This only checks the optional instance list against the existing names.
3200

3201
    """
3202
    if self.wanted_names is None:
3203
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3204

    
3205
    self.wanted_instances = \
3206
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3207

    
3208
  def _EnsureChildSizes(self, disk):
3209
    """Ensure children of the disk have the needed disk size.
3210

3211
    This is valid mainly for DRBD8 and fixes an issue where the
3212
    children have smaller disk size.
3213

3214
    @param disk: an L{ganeti.objects.Disk} object
3215

3216
    """
3217
    if disk.dev_type == constants.LD_DRBD8:
3218
      assert disk.children, "Empty children for DRBD8?"
3219
      fchild = disk.children[0]
3220
      mismatch = fchild.size < disk.size
3221
      if mismatch:
3222
        self.LogInfo("Child disk has size %d, parent %d, fixing",
3223
                     fchild.size, disk.size)
3224
        fchild.size = disk.size
3225

    
3226
      # and we recurse on this child only, not on the metadev
3227
      return self._EnsureChildSizes(fchild) or mismatch
3228
    else:
3229
      return False
3230

    
3231
  def Exec(self, feedback_fn):
3232
    """Verify the size of cluster disks.
3233

3234
    """
3235
    # TODO: check child disks too
3236
    # TODO: check differences in size between primary/secondary nodes
3237
    per_node_disks = {}
3238
    for instance in self.wanted_instances:
3239
      pnode = instance.primary_node
3240
      if pnode not in per_node_disks:
3241
        per_node_disks[pnode] = []
3242
      for idx, disk in enumerate(instance.disks):
3243
        per_node_disks[pnode].append((instance, idx, disk))
3244

    
3245
    changed = []
3246
    for node, dskl in per_node_disks.items():
3247
      newl = [v[2].Copy() for v in dskl]
3248
      for dsk in newl:
3249
        self.cfg.SetDiskID(dsk, node)
3250
      result = self.rpc.call_blockdev_getsize(node, newl)
3251
      if result.fail_msg:
3252
        self.LogWarning("Failure in blockdev_getsize call to node"
3253
                        " %s, ignoring", node)
3254
        continue
3255
      if len(result.payload) != len(dskl):
3256
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
3257
                        " result.payload=%s", node, len(dskl), result.payload)
3258
        self.LogWarning("Invalid result from node %s, ignoring node results",
3259
                        node)
3260
        continue
3261
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
3262
        if size is None:
3263
          self.LogWarning("Disk %d of instance %s did not return size"
3264
                          " information, ignoring", idx, instance.name)
3265
          continue
3266
        if not isinstance(size, (int, long)):
3267
          self.LogWarning("Disk %d of instance %s did not return valid"
3268
                          " size information, ignoring", idx, instance.name)
3269
          continue
3270
        size = size >> 20
3271
        if size != disk.size:
3272
          self.LogInfo("Disk %d of instance %s has mismatched size,"
3273
                       " correcting: recorded %d, actual %d", idx,
3274
                       instance.name, disk.size, size)
3275
          disk.size = size
3276
          self.cfg.Update(instance, feedback_fn)
3277
          changed.append((instance.name, idx, size))
3278
        if self._EnsureChildSizes(disk):
3279
          self.cfg.Update(instance, feedback_fn)
3280
          changed.append((instance.name, idx, disk.size))
3281
    return changed
3282

    
3283

    
3284
class LUClusterRename(LogicalUnit):
3285
  """Rename the cluster.
3286

3287
  """
3288
  HPATH = "cluster-rename"
3289
  HTYPE = constants.HTYPE_CLUSTER
3290

    
3291
  def BuildHooksEnv(self):
3292
    """Build hooks env.
3293

3294
    """
3295
    return {
3296
      "OP_TARGET": self.cfg.GetClusterName(),
3297
      "NEW_NAME": self.op.name,
3298
      }
3299

    
3300
  def BuildHooksNodes(self):
3301
    """Build hooks nodes.
3302

3303
    """
3304
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3305

    
3306
  def CheckPrereq(self):
3307
    """Verify that the passed name is a valid one.
3308

3309
    """
3310
    hostname = netutils.GetHostname(name=self.op.name,
3311
                                    family=self.cfg.GetPrimaryIPFamily())
3312

    
3313
    new_name = hostname.name
3314
    self.ip = new_ip = hostname.ip
3315
    old_name = self.cfg.GetClusterName()
3316
    old_ip = self.cfg.GetMasterIP()
3317
    if new_name == old_name and new_ip == old_ip:
3318
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3319
                                 " cluster has changed",
3320
                                 errors.ECODE_INVAL)
3321
    if new_ip != old_ip:
3322
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3323
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3324
                                   " reachable on the network" %
3325
                                   new_ip, errors.ECODE_NOTUNIQUE)
3326

    
3327
    self.op.name = new_name
3328

    
3329
  def Exec(self, feedback_fn):
3330
    """Rename the cluster.
3331

3332
    """
3333
    clustername = self.op.name
3334
    ip = self.ip
3335

    
3336
    # shutdown the master IP
3337
    master = self.cfg.GetMasterNode()
3338
    result = self.rpc.call_node_deactivate_master_ip(master)
3339
    result.Raise("Could not disable the master role")
3340

    
3341
    try:
3342
      cluster = self.cfg.GetClusterInfo()
3343
      cluster.cluster_name = clustername
3344
      cluster.master_ip = ip
3345
      self.cfg.Update(cluster, feedback_fn)
3346

    
3347
      # update the known hosts file
3348
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3349
      node_list = self.cfg.GetOnlineNodeList()
3350
      try:
3351
        node_list.remove(master)
3352
      except ValueError:
3353
        pass
3354
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3355
    finally:
3356
      result = self.rpc.call_node_activate_master_ip(master)
3357
      msg = result.fail_msg
3358
      if msg:
3359
        self.LogWarning("Could not re-enable the master role on"
3360
                        " the master, please restart manually: %s", msg)
3361

    
3362
    return clustername
3363

    
3364

    
3365
class LUClusterSetParams(LogicalUnit):
3366
  """Change the parameters of the cluster.
3367

3368
  """
3369
  HPATH = "cluster-modify"
3370
  HTYPE = constants.HTYPE_CLUSTER
3371
  REQ_BGL = False
3372

    
3373
  def CheckArguments(self):
3374
    """Check parameters
3375

3376
    """
3377
    if self.op.uid_pool:
3378
      uidpool.CheckUidPool(self.op.uid_pool)
3379

    
3380
    if self.op.add_uids:
3381
      uidpool.CheckUidPool(self.op.add_uids)
3382

    
3383
    if self.op.remove_uids:
3384
      uidpool.CheckUidPool(self.op.remove_uids)
3385

    
3386
  def ExpandNames(self):
3387
    # FIXME: in the future maybe other cluster params won't require checking on
3388
    # all nodes to be modified.
3389
    self.needed_locks = {
3390
      locking.LEVEL_NODE: locking.ALL_SET,
3391
    }
3392
    self.share_locks[locking.LEVEL_NODE] = 1
3393

    
3394
  def BuildHooksEnv(self):
3395
    """Build hooks env.
3396

3397
    """
3398
    return {
3399
      "OP_TARGET": self.cfg.GetClusterName(),
3400
      "NEW_VG_NAME": self.op.vg_name,
3401
      }
3402

    
3403
  def BuildHooksNodes(self):
3404
    """Build hooks nodes.
3405

3406
    """
3407
    mn = self.cfg.GetMasterNode()
3408
    return ([mn], [mn])
3409

    
3410
  def CheckPrereq(self):
3411
    """Check prerequisites.
3412

3413
    This checks whether the given params don't conflict and
3414
    if the given volume group is valid.
3415

3416
    """
3417
    if self.op.vg_name is not None and not self.op.vg_name:
3418
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3419
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3420
                                   " instances exist", errors.ECODE_INVAL)
3421

    
3422
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3423
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3424
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3425
                                   " drbd-based instances exist",
3426
                                   errors.ECODE_INVAL)
3427

    
3428
    node_list = self.owned_locks(locking.LEVEL_NODE)
3429

    
3430
    # if vg_name not None, checks given volume group on all nodes
3431
    if self.op.vg_name:
3432
      vglist = self.rpc.call_vg_list(node_list)
3433
      for node in node_list:
3434
        msg = vglist[node].fail_msg
3435
        if msg:
3436
          # ignoring down node
3437
          self.LogWarning("Error while gathering data on node %s"
3438
                          " (ignoring node): %s", node, msg)
3439
          continue
3440
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3441
                                              self.op.vg_name,
3442
                                              constants.MIN_VG_SIZE)
3443
        if vgstatus:
3444
          raise errors.OpPrereqError("Error on node '%s': %s" %
3445
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3446

    
3447
    if self.op.drbd_helper:
3448
      # checks given drbd helper on all nodes
3449
      helpers = self.rpc.call_drbd_helper(node_list)
3450
      for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3451
        if ninfo.offline:
3452
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3453
          continue
3454
        msg = helpers[node].fail_msg
3455
        if msg:
3456
          raise errors.OpPrereqError("Error checking drbd helper on node"
3457
                                     " '%s': %s" % (node, msg),
3458
                                     errors.ECODE_ENVIRON)
3459
        node_helper = helpers[node].payload
3460
        if node_helper != self.op.drbd_helper:
3461
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3462
                                     (node, node_helper), errors.ECODE_ENVIRON)
3463

    
3464
    self.cluster = cluster = self.cfg.GetClusterInfo()
3465
    # validate params changes
3466
    if self.op.beparams:
3467
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3468
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3469

    
3470
    if self.op.ndparams:
3471
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3472
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3473

    
3474
      # TODO: we need a more general way to handle resetting
3475
      # cluster-level parameters to default values
3476
      if self.new_ndparams["oob_program"] == "":
3477
        self.new_ndparams["oob_program"] = \
3478
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3479

    
3480
    if self.op.nicparams:
3481
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3482
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3483
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3484
      nic_errors = []
3485

    
3486
      # check all instances for consistency
3487
      for instance in self.cfg.GetAllInstancesInfo().values():
3488
        for nic_idx, nic in enumerate(instance.nics):
3489
          params_copy = copy.deepcopy(nic.nicparams)
3490
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3491

    
3492
          # check parameter syntax
3493
          try:
3494
            objects.NIC.CheckParameterSyntax(params_filled)
3495
          except errors.ConfigurationError, err:
3496
            nic_errors.append("Instance %s, nic/%d: %s" %
3497
                              (instance.name, nic_idx, err))
3498

    
3499
          # if we're moving instances to routed, check that they have an ip
3500
          target_mode = params_filled[constants.NIC_MODE]
3501
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3502
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3503
                              " address" % (instance.name, nic_idx))
3504
      if nic_errors:
3505
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3506
                                   "\n".join(nic_errors))
3507

    
3508
    # hypervisor list/parameters
3509
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3510
    if self.op.hvparams:
3511
      for hv_name, hv_dict in self.op.hvparams.items():
3512
        if hv_name not in self.new_hvparams:
3513
          self.new_hvparams[hv_name] = hv_dict
3514
        else:
3515
          self.new_hvparams[hv_name].update(hv_dict)
3516

    
3517
    # os hypervisor parameters
3518
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3519
    if self.op.os_hvp:
3520
      for os_name, hvs in self.op.os_hvp.items():
3521
        if os_name not in self.new_os_hvp:
3522
          self.new_os_hvp[os_name] = hvs
3523
        else:
3524
          for hv_name, hv_dict in hvs.items():
3525
            if hv_name not in self.new_os_hvp[os_name]:
3526
              self.new_os_hvp[os_name][hv_name] = hv_dict
3527
            else:
3528
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3529

    
3530
    # os parameters
3531
    self.new_osp = objects.FillDict(cluster.osparams, {})
3532
    if self.op.osparams:
3533
      for os_name, osp in self.op.osparams.items():
3534
        if os_name not in self.new_osp:
3535
          self.new_osp[os_name] = {}
3536

    
3537
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3538
                                                  use_none=True)
3539

    
3540
        if not self.new_osp[os_name]:
3541
          # we removed all parameters
3542
          del self.new_osp[os_name]
3543
        else:
3544
          # check the parameter validity (remote check)
3545
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3546
                         os_name, self.new_osp[os_name])
3547

    
3548
    # changes to the hypervisor list
3549
    if self.op.enabled_hypervisors is not None:
3550
      self.hv_list = self.op.enabled_hypervisors
3551
      for hv in self.hv_list:
3552
        # if the hypervisor doesn't already exist in the cluster
3553
        # hvparams, we initialize it to empty, and then (in both
3554
        # cases) we make sure to fill the defaults, as we might not
3555
        # have a complete defaults list if the hypervisor wasn't
3556
        # enabled before
3557
        if hv not in new_hvp:
3558
          new_hvp[hv] = {}
3559
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3560
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3561
    else:
3562
      self.hv_list = cluster.enabled_hypervisors
3563

    
3564
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3565
      # either the enabled list has changed, or the parameters have, validate
3566
      for hv_name, hv_params in self.new_hvparams.items():
3567
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3568
            (self.op.enabled_hypervisors and
3569
             hv_name in self.op.enabled_hypervisors)):
3570
          # either this is a new hypervisor, or its parameters have changed
3571
          hv_class = hypervisor.GetHypervisor(hv_name)
3572
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3573
          hv_class.CheckParameterSyntax(hv_params)
3574
          _CheckHVParams(self, node_list, hv_name, hv_params)
3575

    
3576
    if self.op.os_hvp:
3577
      # no need to check any newly-enabled hypervisors, since the
3578
      # defaults have already been checked in the above code-block
3579
      for os_name, os_hvp in self.new_os_hvp.items():
3580
        for hv_name, hv_params in os_hvp.items():
3581
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3582
          # we need to fill in the new os_hvp on top of the actual hv_p
3583
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3584
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3585
          hv_class = hypervisor.GetHypervisor(hv_name)
3586
          hv_class.CheckParameterSyntax(new_osp)
3587
          _CheckHVParams(self, node_list, hv_name, new_osp)
3588

    
3589
    if self.op.default_iallocator:
3590
      alloc_script = utils.FindFile(self.op.default_iallocator,
3591
                                    constants.IALLOCATOR_SEARCH_PATH,
3592
                                    os.path.isfile)
3593
      if alloc_script is None:
3594
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3595
                                   " specified" % self.op.default_iallocator,
3596
                                   errors.ECODE_INVAL)
3597

    
3598
  def Exec(self, feedback_fn):
3599
    """Change the parameters of the cluster.
3600

3601
    """
3602
    if self.op.vg_name is not None:
3603
      new_volume = self.op.vg_name
3604
      if not new_volume:
3605
        new_volume = None
3606
      if new_volume != self.cfg.GetVGName():
3607
        self.cfg.SetVGName(new_volume)
3608
      else:
3609
        feedback_fn("Cluster LVM configuration already in desired"
3610
                    " state, not changing")
3611
    if self.op.drbd_helper is not None:
3612
      new_helper = self.op.drbd_helper
3613
      if not new_helper:
3614
        new_helper = None
3615
      if new_helper != self.cfg.GetDRBDHelper():
3616
        self.cfg.SetDRBDHelper(new_helper)
3617
      else:
3618
        feedback_fn("Cluster DRBD helper already in desired state,"
3619
                    " not changing")
3620
    if self.op.hvparams:
3621
      self.cluster.hvparams = self.new_hvparams
3622
    if self.op.os_hvp:
3623
      self.cluster.os_hvp = self.new_os_hvp
3624
    if self.op.enabled_hypervisors is not None:
3625
      self.cluster.hvparams = self.new_hvparams
3626
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3627
    if self.op.beparams:
3628
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3629
    if self.op.nicparams:
3630
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3631
    if self.op.osparams:
3632
      self.cluster.osparams = self.new_osp
3633
    if self.op.ndparams:
3634
      self.cluster.ndparams = self.new_ndparams
3635

    
3636
    if self.op.candidate_pool_size is not None:
3637
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3638
      # we need to update the pool size here, otherwise the save will fail
3639
      _AdjustCandidatePool(self, [])
3640

    
3641
    if self.op.maintain_node_health is not None:
3642
      self.cluster.maintain_node_health = self.op.maintain_node_health
3643

    
3644
    if self.op.prealloc_wipe_disks is not None:
3645
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3646

    
3647
    if self.op.add_uids is not None:
3648
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3649

    
3650
    if self.op.remove_uids is not None:
3651
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3652

    
3653
    if self.op.uid_pool is not None:
3654
      self.cluster.uid_pool = self.op.uid_pool
3655

    
3656
    if self.op.default_iallocator is not None:
3657
      self.cluster.default_iallocator = self.op.default_iallocator
3658

    
3659
    if self.op.reserved_lvs is not None:
3660
      self.cluster.reserved_lvs = self.op.reserved_lvs
3661

    
3662
    def helper_os(aname, mods, desc):
3663
      desc += " OS list"
3664
      lst = getattr(self.cluster, aname)
3665
      for key, val in mods:
3666
        if key == constants.DDM_ADD:
3667
          if val in lst:
3668
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3669
          else:
3670
            lst.append(val)
3671
        elif key == constants.DDM_REMOVE:
3672
          if val in lst:
3673
            lst.remove(val)
3674
          else:
3675
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3676
        else:
3677
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3678

    
3679
    if self.op.hidden_os:
3680
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3681

    
3682
    if self.op.blacklisted_os:
3683
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3684

    
3685
    if self.op.master_netdev:
3686
      master = self.cfg.GetMasterNode()
3687
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3688
                  self.cluster.master_netdev)
3689
      result = self.rpc.call_node_deactivate_master_ip(master)
3690
      result.Raise("Could not disable the master ip")
3691
      feedback_fn("Changing master_netdev from %s to %s" %
3692
                  (self.cluster.master_netdev, self.op.master_netdev))
3693
      self.cluster.master_netdev = self.op.master_netdev
3694

    
3695
    self.cfg.Update(self.cluster, feedback_fn)
3696

    
3697
    if self.op.master_netdev:
3698
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3699
                  self.op.master_netdev)
3700
      result = self.rpc.call_node_activate_master_ip(master)
3701
      if result.fail_msg:
3702
        self.LogWarning("Could not re-enable the master ip on"
3703
                        " the master, please restart manually: %s",
3704
                        result.fail_msg)
3705

    
3706

    
3707
def _UploadHelper(lu, nodes, fname):
3708
  """Helper for uploading a file and showing warnings.
3709

3710
  """
3711
  if os.path.exists(fname):
3712
    result = lu.rpc.call_upload_file(nodes, fname)
3713
    for to_node, to_result in result.items():
3714
      msg = to_result.fail_msg
3715
      if msg:
3716
        msg = ("Copy of file %s to node %s failed: %s" %
3717
               (fname, to_node, msg))
3718
        lu.proc.LogWarning(msg)
3719

    
3720

    
3721
def _ComputeAncillaryFiles(cluster, redist):
3722
  """Compute files external to Ganeti which need to be consistent.
3723

3724
  @type redist: boolean
3725
  @param redist: Whether to include files which need to be redistributed
3726

3727
  """
3728
  # Compute files for all nodes
3729
  files_all = set([
3730
    constants.SSH_KNOWN_HOSTS_FILE,
3731
    constants.CONFD_HMAC_KEY,
3732
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3733
    constants.RAPI_USERS_FILE,
3734
    ])
3735

    
3736
  if not redist:
3737
    files_all.update(constants.ALL_CERT_FILES)
3738
    files_all.update(ssconf.SimpleStore().GetFileList())
3739
  else:
3740
    # we need to ship at least the RAPI certificate
3741
    files_all.add(constants.RAPI_CERT_FILE)
3742

    
3743
  if cluster.modify_etc_hosts:
3744
    files_all.add(constants.ETC_HOSTS)
3745

    
3746
  # Files which are optional, these must:
3747
  # - be present in one other category as well
3748
  # - either exist or not exist on all nodes of that category (mc, vm all)
3749
  files_opt = set([
3750
    constants.RAPI_USERS_FILE,
3751
    ])
3752

    
3753
  # Files which should only be on master candidates
3754
  files_mc = set()
3755
  if not redist:
3756
    files_mc.add(constants.CLUSTER_CONF_FILE)
3757

    
3758
  # Files which should only be on VM-capable nodes
3759
  files_vm = set(filename
3760
    for hv_name in cluster.enabled_hypervisors
3761
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
3762

    
3763
  files_opt |= set(filename
3764
    for hv_name in cluster.enabled_hypervisors
3765
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
3766

    
3767
  # Filenames in each category must be unique
3768
  all_files_set = files_all | files_mc | files_vm
3769
  assert (len(all_files_set) ==
3770
          sum(map(len, [files_all, files_mc, files_vm]))), \
3771
         "Found file listed in more than one file list"
3772

    
3773
  # Optional files must be present in one other category
3774
  assert all_files_set.issuperset(files_opt), \
3775
         "Optional file not in a different required list"
3776

    
3777
  return (files_all, files_opt, files_mc, files_vm)
3778

    
3779

    
3780
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3781
  """Distribute additional files which are part of the cluster configuration.
3782

3783
  ConfigWriter takes care of distributing the config and ssconf files, but
3784
  there are more files which should be distributed to all nodes. This function
3785
  makes sure those are copied.
3786

3787
  @param lu: calling logical unit
3788
  @param additional_nodes: list of nodes not in the config to distribute to
3789
  @type additional_vm: boolean
3790
  @param additional_vm: whether the additional nodes are vm-capable or not
3791

3792
  """
3793
  # Gather target nodes
3794
  cluster = lu.cfg.GetClusterInfo()
3795
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3796

    
3797
  online_nodes = lu.cfg.GetOnlineNodeList()
3798
  vm_nodes = lu.cfg.GetVmCapableNodeList()
3799

    
3800
  if additional_nodes is not None:
3801
    online_nodes.extend(additional_nodes)
3802
    if additional_vm:
3803
      vm_nodes.extend(additional_nodes)
3804

    
3805
  # Never distribute to master node
3806
  for nodelist in [online_nodes, vm_nodes]:
3807
    if master_info.name in nodelist:
3808
      nodelist.remove(master_info.name)
3809

    
3810
  # Gather file lists
3811
  (files_all, _, files_mc, files_vm) = \
3812
    _ComputeAncillaryFiles(cluster, True)
3813

    
3814
  # Never re-distribute configuration file from here
3815
  assert not (constants.CLUSTER_CONF_FILE in files_all or
3816
              constants.CLUSTER_CONF_FILE in files_vm)
3817
  assert not files_mc, "Master candidates not handled in this function"
3818

    
3819
  filemap = [
3820
    (online_nodes, files_all),
3821
    (vm_nodes, files_vm),
3822
    ]
3823

    
3824
  # Upload the files
3825
  for (node_list, files) in filemap:
3826
    for fname in files:
3827
      _UploadHelper(lu, node_list, fname)
3828

    
3829

    
3830
class LUClusterRedistConf(NoHooksLU):
3831
  """Force the redistribution of cluster configuration.
3832

3833
  This is a very simple LU.
3834

3835
  """
3836
  REQ_BGL = False
3837

    
3838
  def ExpandNames(self):
3839
    self.needed_locks = {
3840
      locking.LEVEL_NODE: locking.ALL_SET,
3841
    }
3842
    self.share_locks[locking.LEVEL_NODE] = 1
3843

    
3844
  def Exec(self, feedback_fn):
3845
    """Redistribute the configuration.
3846

3847
    """
3848
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3849
    _RedistributeAncillaryFiles(self)
3850

    
3851

    
3852
class LUClusterActivateMasterIp(NoHooksLU):
3853
  """Activate the master IP on the master node.
3854

3855
  """
3856
  def Exec(self, feedback_fn):
3857
    """Activate the master IP.
3858

3859
    """
3860
    master = self.cfg.GetMasterNode()
3861
    self.rpc.call_node_activate_master_ip(master)
3862

    
3863

    
3864
class LUClusterDeactivateMasterIp(NoHooksLU):
3865
  """Deactivate the master IP on the master node.
3866

3867
  """
3868
  def Exec(self, feedback_fn):
3869
    """Deactivate the master IP.
3870

3871
    """
3872
    master = self.cfg.GetMasterNode()
3873
    self.rpc.call_node_deactivate_master_ip(master)
3874

    
3875

    
3876
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3877
  """Sleep and poll for an instance's disk to sync.
3878

3879
  """
3880
  if not instance.disks or disks is not None and not disks:
3881
    return True
3882

    
3883
  disks = _ExpandCheckDisks(instance, disks)
3884

    
3885
  if not oneshot:
3886
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3887

    
3888
  node = instance.primary_node
3889

    
3890
  for dev in disks:
3891
    lu.cfg.SetDiskID(dev, node)
3892

    
3893
  # TODO: Convert to utils.Retry
3894

    
3895
  retries = 0
3896
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3897
  while True:
3898
    max_time = 0
3899
    done = True
3900
    cumul_degraded = False
3901
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3902
    msg = rstats.fail_msg
3903
    if msg:
3904
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3905
      retries += 1
3906
      if retries >= 10:
3907
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3908
                                 " aborting." % node)
3909
      time.sleep(6)
3910
      continue
3911
    rstats = rstats.payload
3912
    retries = 0
3913
    for i, mstat in enumerate(rstats):
3914
      if mstat is None:
3915
        lu.LogWarning("Can't compute data for node %s/%s",
3916
                           node, disks[i].iv_name)
3917
        continue
3918

    
3919
      cumul_degraded = (cumul_degraded or
3920
                        (mstat.is_degraded and mstat.sync_percent is None))
3921
      if mstat.sync_percent is not None:
3922
        done = False
3923
        if mstat.estimated_time is not None:
3924
          rem_time = ("%s remaining (estimated)" %
3925
                      utils.FormatSeconds(mstat.estimated_time))
3926
          max_time = mstat.estimated_time
3927
        else:
3928
          rem_time = "no time estimate"
3929
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3930
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3931

    
3932
    # if we're done but degraded, let's do a few small retries, to
3933
    # make sure we see a stable and not transient situation; therefore
3934
    # we force restart of the loop
3935
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3936
      logging.info("Degraded disks found, %d retries left", degr_retries)
3937
      degr_retries -= 1
3938
      time.sleep(1)
3939
      continue
3940

    
3941
    if done or oneshot:
3942
      break
3943

    
3944
    time.sleep(min(60, max_time))
3945

    
3946
  if done:
3947
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3948
  return not cumul_degraded
3949

    
3950

    
3951
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3952
  """Check that mirrors are not degraded.
3953

3954
  The ldisk parameter, if True, will change the test from the
3955
  is_degraded attribute (which represents overall non-ok status for
3956
  the device(s)) to the ldisk (representing the local storage status).
3957

3958
  """
3959
  lu.cfg.SetDiskID(dev, node)
3960

    
3961
  result = True
3962

    
3963
  if on_primary or dev.AssembleOnSecondary():
3964
    rstats = lu.rpc.call_blockdev_find(node, dev)
3965
    msg = rstats.fail_msg
3966
    if msg:
3967
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3968
      result = False
3969
    elif not rstats.payload:
3970
      lu.LogWarning("Can't find disk on node %s", node)
3971
      result = False
3972
    else:
3973
      if ldisk:
3974
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3975
      else:
3976
        result = result and not rstats.payload.is_degraded
3977

    
3978
  if dev.children:
3979
    for child in dev.children:
3980
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3981

    
3982
  return result
3983

    
3984

    
3985
class LUOobCommand(NoHooksLU):
3986
  """Logical unit for OOB handling.
3987

3988
  """
3989
  REG_BGL = False
3990
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
3991

    
3992
  def ExpandNames(self):
3993
    """Gather locks we need.
3994

3995
    """
3996
    if self.op.node_names:
3997
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
3998
      lock_names = self.op.node_names
3999
    else:
4000
      lock_names = locking.ALL_SET
4001

    
4002
    self.needed_locks = {
4003
      locking.LEVEL_NODE: lock_names,
4004
      }
4005

    
4006
  def CheckPrereq(self):
4007
    """Check prerequisites.
4008

4009
    This checks:
4010
     - the node exists in the configuration
4011
     - OOB is supported
4012

4013
    Any errors are signaled by raising errors.OpPrereqError.
4014

4015
    """
4016
    self.nodes = []
4017
    self.master_node = self.cfg.GetMasterNode()
4018

    
4019
    assert self.op.power_delay >= 0.0
4020

    
4021
    if self.op.node_names:
4022
      if (self.op.command in self._SKIP_MASTER and
4023
          self.master_node in self.op.node_names):
4024
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
4025
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
4026

    
4027
        if master_oob_handler:
4028
          additional_text = ("run '%s %s %s' if you want to operate on the"
4029
                             " master regardless") % (master_oob_handler,
4030
                                                      self.op.command,
4031
                                                      self.master_node)
4032
        else:
4033
          additional_text = "it does not support out-of-band operations"
4034

    
4035
        raise errors.OpPrereqError(("Operating on the master node %s is not"
4036
                                    " allowed for %s; %s") %
4037
                                   (self.master_node, self.op.command,
4038
                                    additional_text), errors.ECODE_INVAL)
4039
    else:
4040
      self.op.node_names = self.cfg.GetNodeList()
4041
      if self.op.command in self._SKIP_MASTER:
4042
        self.op.node_names.remove(self.master_node)
4043

    
4044
    if self.op.command in self._SKIP_MASTER:
4045
      assert self.master_node not in self.op.node_names
4046

    
4047
    for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
4048
      if node is None:
4049
        raise errors.OpPrereqError("Node %s not found" % node_name,
4050
                                   errors.ECODE_NOENT)
4051
      else:
4052
        self.nodes.append(node)
4053

    
4054
      if (not self.op.ignore_status and
4055
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
4056
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
4057
                                    " not marked offline") % node_name,
4058
                                   errors.ECODE_STATE)
4059

    
4060
  def Exec(self, feedback_fn):
4061
    """Execute OOB and return result if we expect any.
4062

4063
    """
4064
    master_node = self.master_node
4065
    ret = []
4066

    
4067
    for idx, node in enumerate(utils.NiceSort(self.nodes,
4068
                                              key=lambda node: node.name)):
4069
      node_entry = [(constants.RS_NORMAL, node.name)]
4070
      ret.append(node_entry)
4071

    
4072
      oob_program = _SupportsOob(self.cfg, node)
4073

    
4074
      if not oob_program:
4075
        node_entry.append((constants.RS_UNAVAIL, None))
4076
        continue
4077

    
4078
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
4079
                   self.op.command, oob_program, node.name)
4080
      result = self.rpc.call_run_oob(master_node, oob_program,
4081
                                     self.op.command, node.name,
4082
                                     self.op.timeout)
4083

    
4084
      if result.fail_msg:
4085
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4086
                        node.name, result.fail_msg)
4087
        node_entry.append((constants.RS_NODATA, None))
4088
      else:
4089
        try:
4090
          self._CheckPayload(result)
4091
        except errors.OpExecError, err:
4092
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
4093
                          node.name, err)
4094
          node_entry.append((constants.RS_NODATA, None))
4095
        else:
409