Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ fcad7225

History | View | Annotate | Download (471.7 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-msg=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-msg=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-msg=C0103
135
    self.LogWarning = processor.LogWarning # pylint: disable-msg=C0103
136
    self.LogInfo = processor.LogInfo # pylint: disable-msg=C0103
137
    self.LogStep = processor.LogStep # pylint: disable-msg=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-msg=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-msg=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-msg=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-msg=W0142
1091

    
1092

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

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

    
1108

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

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

    
1119

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

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

    
1133

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

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

    
1142

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

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

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

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

    
1166

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

    
1170

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

1174
  """
1175

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

    
1178

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

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

    
1186

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

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

    
1194

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

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

    
1204
  return []
1205

    
1206

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

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

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

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

    
1221
  return faulty
1222

    
1223

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

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

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

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

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

    
1255

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

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

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

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

    
1276
  return iallocator
1277

    
1278

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

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

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

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

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

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

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

1303
    """
1304
    return True
1305

    
1306

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

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

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

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

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

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

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

1331
    This checks whether the cluster is empty.
1332

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

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

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

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

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

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

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

    
1361
    return master
1362

    
1363

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

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

1370
  """
1371
  try:
1372
    cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1373
                                           utils.ReadFile(filename))
1374
  except Exception, err: # pylint: disable-msg=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-msg=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-msg=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-msg=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-msg=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
    self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
1578
    self.all_node_info = self.cfg.GetAllNodesInfo()
1579
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1580
    self.needed_locks = {}
1581

    
1582
  def Exec(self, feedback_fn):
1583
    """Verify integrity of cluster, performing various test on nodes.
1584

1585
    """
1586
    self.bad = False
1587
    self._feedback_fn = feedback_fn
1588

    
1589
    feedback_fn("* Verifying cluster config")
1590

    
1591
    for msg in self.cfg.VerifyConfig():
1592
      self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
1593

    
1594
    feedback_fn("* Verifying cluster certificate files")
1595

    
1596
    for cert_filename in constants.ALL_CERT_FILES:
1597
      (errcode, msg) = _VerifyCertificate(cert_filename)
1598
      self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
1599

    
1600
    feedback_fn("* Verifying hypervisor parameters")
1601

    
1602
    self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
1603
                                                self.all_inst_info.values()))
1604

    
1605
    feedback_fn("* Verifying all nodes belong to an existing group")
1606

    
1607
    # We do this verification here because, should this bogus circumstance
1608
    # occur, it would never be caught by VerifyGroup, which only acts on
1609
    # nodes/instances reachable from existing node groups.
1610

    
1611
    dangling_nodes = set(node.name for node in self.all_node_info.values()
1612
                         if node.group not in self.all_group_info)
1613

    
1614
    dangling_instances = {}
1615
    no_node_instances = []
1616

    
1617
    for inst in self.all_inst_info.values():
1618
      if inst.primary_node in dangling_nodes:
1619
        dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
1620
      elif inst.primary_node not in self.all_node_info:
1621
        no_node_instances.append(inst.name)
1622

    
1623
    pretty_dangling = [
1624
        "%s (%s)" %
1625
        (node.name,
1626
         utils.CommaJoin(dangling_instances.get(node.name,
1627
                                                ["no instances"])))
1628
        for node in dangling_nodes]
1629

    
1630
    self._ErrorIf(bool(dangling_nodes), self.ECLUSTERDANGLINGNODES, None,
1631
                  "the following nodes (and their instances) belong to a non"
1632
                  " existing group: %s", utils.CommaJoin(pretty_dangling))
1633

    
1634
    self._ErrorIf(bool(no_node_instances), self.ECLUSTERDANGLINGINST, None,
1635
                  "the following instances have a non-existing primary-node:"
1636
                  " %s", utils.CommaJoin(no_node_instances))
1637

    
1638
    return (not self.bad, [g.name for g in self.all_group_info.values()])
1639

    
1640

    
1641
class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
1642
  """Verifies the status of a node group.
1643

1644
  """
1645
  HPATH = "cluster-verify"
1646
  HTYPE = constants.HTYPE_CLUSTER
1647
  REQ_BGL = False
1648

    
1649
  _HOOKS_INDENT_RE = re.compile("^", re.M)
1650

    
1651
  class NodeImage(object):
1652
    """A class representing the logical and physical status of a node.
1653

1654
    @type name: string
1655
    @ivar name: the node name to which this object refers
1656
    @ivar volumes: a structure as returned from
1657
        L{ganeti.backend.GetVolumeList} (runtime)
1658
    @ivar instances: a list of running instances (runtime)
1659
    @ivar pinst: list of configured primary instances (config)
1660
    @ivar sinst: list of configured secondary instances (config)
1661
    @ivar sbp: dictionary of {primary-node: list of instances} for all
1662
        instances for which this node is secondary (config)
1663
    @ivar mfree: free memory, as reported by hypervisor (runtime)
1664
    @ivar dfree: free disk, as reported by the node (runtime)
1665
    @ivar offline: the offline status (config)
1666
    @type rpc_fail: boolean
1667
    @ivar rpc_fail: whether the RPC verify call was successfull (overall,
1668
        not whether the individual keys were correct) (runtime)
1669
    @type lvm_fail: boolean
1670
    @ivar lvm_fail: whether the RPC call didn't return valid LVM data
1671
    @type hyp_fail: boolean
1672
    @ivar hyp_fail: whether the RPC call didn't return the instance list
1673
    @type ghost: boolean
1674
    @ivar ghost: whether this is a known node or not (config)
1675
    @type os_fail: boolean
1676
    @ivar os_fail: whether the RPC call didn't return valid OS data
1677
    @type oslist: list
1678
    @ivar oslist: list of OSes as diagnosed by DiagnoseOS
1679
    @type vm_capable: boolean
1680
    @ivar vm_capable: whether the node can host instances
1681

1682
    """
1683
    def __init__(self, offline=False, name=None, vm_capable=True):
1684
      self.name = name
1685
      self.volumes = {}
1686
      self.instances = []
1687
      self.pinst = []
1688
      self.sinst = []
1689
      self.sbp = {}
1690
      self.mfree = 0
1691
      self.dfree = 0
1692
      self.offline = offline
1693
      self.vm_capable = vm_capable
1694
      self.rpc_fail = False
1695
      self.lvm_fail = False
1696
      self.hyp_fail = False
1697
      self.ghost = False
1698
      self.os_fail = False
1699
      self.oslist = {}
1700

    
1701
  def ExpandNames(self):
1702
    # This raises errors.OpPrereqError on its own:
1703
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
1704

    
1705
    # Get instances in node group; this is unsafe and needs verification later
1706
    inst_names = self.cfg.GetNodeGroupInstances(self.group_uuid)
1707

    
1708
    self.needed_locks = {
1709
      locking.LEVEL_INSTANCE: inst_names,
1710
      locking.LEVEL_NODEGROUP: [self.group_uuid],
1711
      locking.LEVEL_NODE: [],
1712
      }
1713

    
1714
    self.share_locks = _ShareAll()
1715

    
1716
  def DeclareLocks(self, level):
1717
    if level == locking.LEVEL_NODE:
1718
      # Get members of node group; this is unsafe and needs verification later
1719
      nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
1720

    
1721
      all_inst_info = self.cfg.GetAllInstancesInfo()
1722

    
1723
      # In Exec(), we warn about mirrored instances that have primary and
1724
      # secondary living in separate node groups. To fully verify that
1725
      # volumes for these instances are healthy, we will need to do an
1726
      # extra call to their secondaries. We ensure here those nodes will
1727
      # be locked.
1728
      for inst in self.owned_locks(locking.LEVEL_INSTANCE):
1729
        # Important: access only the instances whose lock is owned
1730
        if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
1731
          nodes.update(all_inst_info[inst].secondary_nodes)
1732

    
1733
      self.needed_locks[locking.LEVEL_NODE] = nodes
1734

    
1735
  def CheckPrereq(self):
1736
    group_nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
1737
    group_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
1738

    
1739
    unlocked_nodes = \
1740
        group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1741

    
1742
    unlocked_instances = \
1743
        group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
1744

    
1745
    if unlocked_nodes:
1746
      raise errors.OpPrereqError("Missing lock for nodes: %s" %
1747
                                 utils.CommaJoin(unlocked_nodes))
1748

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

    
1753
    self.all_node_info = self.cfg.GetAllNodesInfo()
1754
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1755

    
1756
    self.my_node_names = utils.NiceSort(group_nodes)
1757
    self.my_inst_names = utils.NiceSort(group_instances)
1758

    
1759
    self.my_node_info = dict((name, self.all_node_info[name])
1760
                             for name in self.my_node_names)
1761

    
1762
    self.my_inst_info = dict((name, self.all_inst_info[name])
1763
                             for name in self.my_inst_names)
1764

    
1765
    # We detect here the nodes that will need the extra RPC calls for verifying
1766
    # split LV volumes; they should be locked.
1767
    extra_lv_nodes = set()
1768

    
1769
    for inst in self.my_inst_info.values():
1770
      if inst.disk_template in constants.DTS_INT_MIRROR:
1771
        group = self.my_node_info[inst.primary_node].group
1772
        for nname in inst.secondary_nodes:
1773
          if self.all_node_info[nname].group != group:
1774
            extra_lv_nodes.add(nname)
1775

    
1776
    unlocked_lv_nodes = \
1777
        extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
1778

    
1779
    if unlocked_lv_nodes:
1780
      raise errors.OpPrereqError("these nodes could be locked: %s" %
1781
                                 utils.CommaJoin(unlocked_lv_nodes))
1782
    self.extra_lv_nodes = list(extra_lv_nodes)
1783

    
1784
  def _VerifyNode(self, ninfo, nresult):
1785
    """Perform some basic validation on data returned from a node.
1786

1787
      - check the result data structure is well formed and has all the
1788
        mandatory fields
1789
      - check ganeti version
1790

1791
    @type ninfo: L{objects.Node}
1792
    @param ninfo: the node to check
1793
    @param nresult: the results from the node
1794
    @rtype: boolean
1795
    @return: whether overall this call was successful (and we can expect
1796
         reasonable values in the respose)
1797

1798
    """
1799
    node = ninfo.name
1800
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1801

    
1802
    # main result, nresult should be a non-empty dict
1803
    test = not nresult or not isinstance(nresult, dict)
1804
    _ErrorIf(test, self.ENODERPC, node,
1805
                  "unable to verify node: no data returned")
1806
    if test:
1807
      return False
1808

    
1809
    # compares ganeti version
1810
    local_version = constants.PROTOCOL_VERSION
1811
    remote_version = nresult.get("version", None)
1812
    test = not (remote_version and
1813
                isinstance(remote_version, (list, tuple)) and
1814
                len(remote_version) == 2)
1815
    _ErrorIf(test, self.ENODERPC, node,
1816
             "connection to node returned invalid data")
1817
    if test:
1818
      return False
1819

    
1820
    test = local_version != remote_version[0]
1821
    _ErrorIf(test, self.ENODEVERSION, node,
1822
             "incompatible protocol versions: master %s,"
1823
             " node %s", local_version, remote_version[0])
1824
    if test:
1825
      return False
1826

    
1827
    # node seems compatible, we can actually try to look into its results
1828

    
1829
    # full package version
1830
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1831
                  self.ENODEVERSION, node,
1832
                  "software version mismatch: master %s, node %s",
1833
                  constants.RELEASE_VERSION, remote_version[1],
1834
                  code=self.ETYPE_WARNING)
1835

    
1836
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1837
    if ninfo.vm_capable and isinstance(hyp_result, dict):
1838
      for hv_name, hv_result in hyp_result.iteritems():
1839
        test = hv_result is not None
1840
        _ErrorIf(test, self.ENODEHV, node,
1841
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1842

    
1843
    hvp_result = nresult.get(constants.NV_HVPARAMS, None)
1844
    if ninfo.vm_capable and isinstance(hvp_result, list):
1845
      for item, hv_name, hv_result in hvp_result:
1846
        _ErrorIf(True, self.ENODEHV, node,
1847
                 "hypervisor %s parameter verify failure (source %s): %s",
1848
                 hv_name, item, hv_result)
1849

    
1850
    test = nresult.get(constants.NV_NODESETUP,
1851
                       ["Missing NODESETUP results"])
1852
    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
1853
             "; ".join(test))
1854

    
1855
    return True
1856

    
1857
  def _VerifyNodeTime(self, ninfo, nresult,
1858
                      nvinfo_starttime, nvinfo_endtime):
1859
    """Check the node time.
1860

1861
    @type ninfo: L{objects.Node}
1862
    @param ninfo: the node to check
1863
    @param nresult: the remote results for the node
1864
    @param nvinfo_starttime: the start time of the RPC call
1865
    @param nvinfo_endtime: the end time of the RPC call
1866

1867
    """
1868
    node = ninfo.name
1869
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1870

    
1871
    ntime = nresult.get(constants.NV_TIME, None)
1872
    try:
1873
      ntime_merged = utils.MergeTime(ntime)
1874
    except (ValueError, TypeError):
1875
      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
1876
      return
1877

    
1878
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1879
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1880
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1881
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1882
    else:
1883
      ntime_diff = None
1884

    
1885
    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
1886
             "Node time diverges by at least %s from master node time",
1887
             ntime_diff)
1888

    
1889
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1890
    """Check the node LVM results.
1891

1892
    @type ninfo: L{objects.Node}
1893
    @param ninfo: the node to check
1894
    @param nresult: the remote results for the node
1895
    @param vg_name: the configured VG name
1896

1897
    """
1898
    if vg_name is None:
1899
      return
1900

    
1901
    node = ninfo.name
1902
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1903

    
1904
    # checks vg existence and size > 20G
1905
    vglist = nresult.get(constants.NV_VGLIST, None)
1906
    test = not vglist
1907
    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
1908
    if not test:
1909
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
1910
                                            constants.MIN_VG_SIZE)
1911
      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
1912

    
1913
    # check pv names
1914
    pvlist = nresult.get(constants.NV_PVLIST, None)
1915
    test = pvlist is None
1916
    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
1917
    if not test:
1918
      # check that ':' is not present in PV names, since it's a
1919
      # special character for lvcreate (denotes the range of PEs to
1920
      # use on the PV)
1921
      for _, pvname, owner_vg in pvlist:
1922
        test = ":" in pvname
1923
        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
1924
                 " '%s' of VG '%s'", pvname, owner_vg)
1925

    
1926
  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
1927
    """Check the node bridges.
1928

1929
    @type ninfo: L{objects.Node}
1930
    @param ninfo: the node to check
1931
    @param nresult: the remote results for the node
1932
    @param bridges: the expected list of bridges
1933

1934
    """
1935
    if not bridges:
1936
      return
1937

    
1938
    node = ninfo.name
1939
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1940

    
1941
    missing = nresult.get(constants.NV_BRIDGES, None)
1942
    test = not isinstance(missing, list)
1943
    _ErrorIf(test, self.ENODENET, node,
1944
             "did not return valid bridge information")
1945
    if not test:
1946
      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
1947
               utils.CommaJoin(sorted(missing)))
1948

    
1949
  def _VerifyNodeNetwork(self, ninfo, nresult):
1950
    """Check the node network connectivity results.
1951

1952
    @type ninfo: L{objects.Node}
1953
    @param ninfo: the node to check
1954
    @param nresult: the remote results for the node
1955

1956
    """
1957
    node = ninfo.name
1958
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1959

    
1960
    test = constants.NV_NODELIST not in nresult
1961
    _ErrorIf(test, self.ENODESSH, node,
1962
             "node hasn't returned node ssh connectivity data")
1963
    if not test:
1964
      if nresult[constants.NV_NODELIST]:
1965
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1966
          _ErrorIf(True, self.ENODESSH, node,
1967
                   "ssh communication with node '%s': %s", a_node, a_msg)
1968

    
1969
    test = constants.NV_NODENETTEST not in nresult
1970
    _ErrorIf(test, self.ENODENET, node,
1971
             "node hasn't returned node tcp connectivity data")
1972
    if not test:
1973
      if nresult[constants.NV_NODENETTEST]:
1974
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1975
        for anode in nlist:
1976
          _ErrorIf(True, self.ENODENET, node,
1977
                   "tcp communication with node '%s': %s",
1978
                   anode, nresult[constants.NV_NODENETTEST][anode])
1979

    
1980
    test = constants.NV_MASTERIP not in nresult
1981
    _ErrorIf(test, self.ENODENET, node,
1982
             "node hasn't returned node master IP reachability data")
1983
    if not test:
1984
      if not nresult[constants.NV_MASTERIP]:
1985
        if node == self.master_node:
1986
          msg = "the master node cannot reach the master IP (not configured?)"
1987
        else:
1988
          msg = "cannot reach the master IP"
1989
        _ErrorIf(True, self.ENODENET, node, msg)
1990

    
1991
  def _VerifyInstance(self, instance, instanceconfig, node_image,
1992
                      diskstatus):
1993
    """Verify an instance.
1994

1995
    This function checks to see if the required block devices are
1996
    available on the instance's node.
1997

1998
    """
1999
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2000
    node_current = instanceconfig.primary_node
2001

    
2002
    node_vol_should = {}
2003
    instanceconfig.MapLVsByNode(node_vol_should)
2004

    
2005
    for node in node_vol_should:
2006
      n_img = node_image[node]
2007
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2008
        # ignore missing volumes on offline or broken nodes
2009
        continue
2010
      for volume in node_vol_should[node]:
2011
        test = volume not in n_img.volumes
2012
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
2013
                 "volume %s missing on node %s", volume, node)
2014

    
2015
    if instanceconfig.admin_up:
2016
      pri_img = node_image[node_current]
2017
      test = instance not in pri_img.instances and not pri_img.offline
2018
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
2019
               "instance not running on its primary node %s",
2020
               node_current)
2021

    
2022
    diskdata = [(nname, success, status, idx)
2023
                for (nname, disks) in diskstatus.items()
2024
                for idx, (success, status) in enumerate(disks)]
2025

    
2026
    for nname, success, bdev_status, idx in diskdata:
2027
      # the 'ghost node' construction in Exec() ensures that we have a
2028
      # node here
2029
      snode = node_image[nname]
2030
      bad_snode = snode.ghost or snode.offline
2031
      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
2032
               self.EINSTANCEFAULTYDISK, instance,
2033
               "couldn't retrieve status for disk/%s on %s: %s",
2034
               idx, nname, bdev_status)
2035
      _ErrorIf((instanceconfig.admin_up and success and
2036
                bdev_status.ldisk_status == constants.LDS_FAULTY),
2037
               self.EINSTANCEFAULTYDISK, instance,
2038
               "disk/%s on %s is faulty", idx, nname)
2039

    
2040
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
2041
    """Verify if there are any unknown volumes in the cluster.
2042

2043
    The .os, .swap and backup volumes are ignored. All other volumes are
2044
    reported as unknown.
2045

2046
    @type reserved: L{ganeti.utils.FieldSet}
2047
    @param reserved: a FieldSet of reserved volume names
2048

2049
    """
2050
    for node, n_img in node_image.items():
2051
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
2052
        # skip non-healthy nodes
2053
        continue
2054
      for volume in n_img.volumes:
2055
        test = ((node not in node_vol_should or
2056
                volume not in node_vol_should[node]) and
2057
                not reserved.Matches(volume))
2058
        self._ErrorIf(test, self.ENODEORPHANLV, node,
2059
                      "volume %s is unknown", volume)
2060

    
2061
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
2062
    """Verify N+1 Memory Resilience.
2063

2064
    Check that if one single node dies we can still start all the
2065
    instances it was primary for.
2066

2067
    """
2068
    cluster_info = self.cfg.GetClusterInfo()
2069
    for node, n_img in node_image.items():
2070
      # This code checks that every node which is now listed as
2071
      # secondary has enough memory to host all instances it is
2072
      # supposed to should a single other node in the cluster fail.
2073
      # FIXME: not ready for failover to an arbitrary node
2074
      # FIXME: does not support file-backed instances
2075
      # WARNING: we currently take into account down instances as well
2076
      # as up ones, considering that even if they're down someone
2077
      # might want to start them even in the event of a node failure.
2078
      if n_img.offline:
2079
        # we're skipping offline nodes from the N+1 warning, since
2080
        # most likely we don't have good memory infromation from them;
2081
        # we already list instances living on such nodes, and that's
2082
        # enough warning
2083
        continue
2084
      for prinode, instances in n_img.sbp.items():
2085
        needed_mem = 0
2086
        for instance in instances:
2087
          bep = cluster_info.FillBE(instance_cfg[instance])
2088
          if bep[constants.BE_AUTO_BALANCE]:
2089
            needed_mem += bep[constants.BE_MEMORY]
2090
        test = n_img.mfree < needed_mem
2091
        self._ErrorIf(test, self.ENODEN1, node,
2092
                      "not enough memory to accomodate instance failovers"
2093
                      " should node %s fail (%dMiB needed, %dMiB available)",
2094
                      prinode, needed_mem, n_img.mfree)
2095

    
2096
  @classmethod
2097
  def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
2098
                   (files_all, files_all_opt, files_mc, files_vm)):
2099
    """Verifies file checksums collected from all nodes.
2100

2101
    @param errorif: Callback for reporting errors
2102
    @param nodeinfo: List of L{objects.Node} objects
2103
    @param master_node: Name of master node
2104
    @param all_nvinfo: RPC results
2105

2106
    """
2107
    node_names = frozenset(node.name for node in nodeinfo if not node.offline)
2108

    
2109
    assert master_node in node_names
2110
    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
2111
            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
2112
           "Found file listed in more than one file list"
2113

    
2114
    # Define functions determining which nodes to consider for a file
2115
    file2nodefn = dict([(filename, fn)
2116
      for (files, fn) in [(files_all, None),
2117
                          (files_all_opt, None),
2118
                          (files_mc, lambda node: (node.master_candidate or
2119
                                                   node.name == master_node)),
2120
                          (files_vm, lambda node: node.vm_capable)]
2121
      for filename in files])
2122

    
2123
    fileinfo = dict((filename, {}) for filename in file2nodefn.keys())
2124

    
2125
    for node in nodeinfo:
2126
      if node.offline:
2127
        continue
2128

    
2129
      nresult = all_nvinfo[node.name]
2130

    
2131
      if nresult.fail_msg or not nresult.payload:
2132
        node_files = None
2133
      else:
2134
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
2135

    
2136
      test = not (node_files and isinstance(node_files, dict))
2137
      errorif(test, cls.ENODEFILECHECK, node.name,
2138
              "Node did not return file checksum data")
2139
      if test:
2140
        continue
2141

    
2142
      for (filename, checksum) in node_files.items():
2143
        # Check if the file should be considered for a node
2144
        fn = file2nodefn[filename]
2145
        if fn is None or fn(node):
2146
          fileinfo[filename].setdefault(checksum, set()).add(node.name)
2147

    
2148
    for (filename, checksums) in fileinfo.items():
2149
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
2150

    
2151
      # Nodes having the file
2152
      with_file = frozenset(node_name
2153
                            for nodes in fileinfo[filename].values()
2154
                            for node_name in nodes)
2155

    
2156
      # Nodes missing file
2157
      missing_file = node_names - with_file
2158

    
2159
      if filename in files_all_opt:
2160
        # All or no nodes
2161
        errorif(missing_file and missing_file != node_names,
2162
                cls.ECLUSTERFILECHECK, None,
2163
                "File %s is optional, but it must exist on all or no"
2164
                " nodes (not found on %s)",
2165
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2166
      else:
2167
        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
2168
                "File %s is missing from node(s) %s", filename,
2169
                utils.CommaJoin(utils.NiceSort(missing_file)))
2170

    
2171
      # See if there are multiple versions of the file
2172
      test = len(checksums) > 1
2173
      if test:
2174
        variants = ["variant %s on %s" %
2175
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2176
                    for (idx, (checksum, nodes)) in
2177
                      enumerate(sorted(checksums.items()))]
2178
      else:
2179
        variants = []
2180

    
2181
      errorif(test, cls.ECLUSTERFILECHECK, None,
2182
              "File %s found with %s different checksums (%s)",
2183
              filename, len(checksums), "; ".join(variants))
2184

    
2185
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2186
                      drbd_map):
2187
    """Verifies and the node DRBD status.
2188

2189
    @type ninfo: L{objects.Node}
2190
    @param ninfo: the node to check
2191
    @param nresult: the remote results for the node
2192
    @param instanceinfo: the dict of instances
2193
    @param drbd_helper: the configured DRBD usermode helper
2194
    @param drbd_map: the DRBD map as returned by
2195
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2196

2197
    """
2198
    node = ninfo.name
2199
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2200

    
2201
    if drbd_helper:
2202
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2203
      test = (helper_result == None)
2204
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
2205
               "no drbd usermode helper returned")
2206
      if helper_result:
2207
        status, payload = helper_result
2208
        test = not status
2209
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2210
                 "drbd usermode helper check unsuccessful: %s", payload)
2211
        test = status and (payload != drbd_helper)
2212
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2213
                 "wrong drbd usermode helper: %s", payload)
2214

    
2215
    # compute the DRBD minors
2216
    node_drbd = {}
2217
    for minor, instance in drbd_map[node].items():
2218
      test = instance not in instanceinfo
2219
      _ErrorIf(test, self.ECLUSTERCFG, None,
2220
               "ghost instance '%s' in temporary DRBD map", instance)
2221
        # ghost instance should not be running, but otherwise we
2222
        # don't give double warnings (both ghost instance and
2223
        # unallocated minor in use)
2224
      if test:
2225
        node_drbd[minor] = (instance, False)
2226
      else:
2227
        instance = instanceinfo[instance]
2228
        node_drbd[minor] = (instance.name, instance.admin_up)
2229

    
2230
    # and now check them
2231
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2232
    test = not isinstance(used_minors, (tuple, list))
2233
    _ErrorIf(test, self.ENODEDRBD, node,
2234
             "cannot parse drbd status file: %s", str(used_minors))
2235
    if test:
2236
      # we cannot check drbd status
2237
      return
2238

    
2239
    for minor, (iname, must_exist) in node_drbd.items():
2240
      test = minor not in used_minors and must_exist
2241
      _ErrorIf(test, self.ENODEDRBD, node,
2242
               "drbd minor %d of instance %s is not active", minor, iname)
2243
    for minor in used_minors:
2244
      test = minor not in node_drbd
2245
      _ErrorIf(test, self.ENODEDRBD, node,
2246
               "unallocated drbd minor %d is in use", minor)
2247

    
2248
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2249
    """Builds the node OS structures.
2250

2251
    @type ninfo: L{objects.Node}
2252
    @param ninfo: the node to check
2253
    @param nresult: the remote results for the node
2254
    @param nimg: the node image object
2255

2256
    """
2257
    node = ninfo.name
2258
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2259

    
2260
    remote_os = nresult.get(constants.NV_OSLIST, None)
2261
    test = (not isinstance(remote_os, list) or
2262
            not compat.all(isinstance(v, list) and len(v) == 7
2263
                           for v in remote_os))
2264

    
2265
    _ErrorIf(test, self.ENODEOS, node,
2266
             "node hasn't returned valid OS data")
2267

    
2268
    nimg.os_fail = test
2269

    
2270
    if test:
2271
      return
2272

    
2273
    os_dict = {}
2274

    
2275
    for (name, os_path, status, diagnose,
2276
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2277

    
2278
      if name not in os_dict:
2279
        os_dict[name] = []
2280

    
2281
      # parameters is a list of lists instead of list of tuples due to
2282
      # JSON lacking a real tuple type, fix it:
2283
      parameters = [tuple(v) for v in parameters]
2284
      os_dict[name].append((os_path, status, diagnose,
2285
                            set(variants), set(parameters), set(api_ver)))
2286

    
2287
    nimg.oslist = os_dict
2288

    
2289
  def _VerifyNodeOS(self, ninfo, nimg, base):
2290
    """Verifies the node OS list.
2291

2292
    @type ninfo: L{objects.Node}
2293
    @param ninfo: the node to check
2294
    @param nimg: the node image object
2295
    @param base: the 'template' node we match against (e.g. from the master)
2296

2297
    """
2298
    node = ninfo.name
2299
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2300

    
2301
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2302

    
2303
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2304
    for os_name, os_data in nimg.oslist.items():
2305
      assert os_data, "Empty OS status for OS %s?!" % os_name
2306
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2307
      _ErrorIf(not f_status, self.ENODEOS, node,
2308
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2309
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
2310
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2311
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2312
      # comparisons with the 'base' image
2313
      test = os_name not in base.oslist
2314
      _ErrorIf(test, self.ENODEOS, node,
2315
               "Extra OS %s not present on reference node (%s)",
2316
               os_name, base.name)
2317
      if test:
2318
        continue
2319
      assert base.oslist[os_name], "Base node has empty OS status?"
2320
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2321
      if not b_status:
2322
        # base OS is invalid, skipping
2323
        continue
2324
      for kind, a, b in [("API version", f_api, b_api),
2325
                         ("variants list", f_var, b_var),
2326
                         ("parameters", beautify_params(f_param),
2327
                          beautify_params(b_param))]:
2328
        _ErrorIf(a != b, self.ENODEOS, node,
2329
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2330
                 kind, os_name, base.name,
2331
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2332

    
2333
    # check any missing OSes
2334
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2335
    _ErrorIf(missing, self.ENODEOS, node,
2336
             "OSes present on reference node %s but missing on this node: %s",
2337
             base.name, utils.CommaJoin(missing))
2338

    
2339
  def _VerifyOob(self, ninfo, nresult):
2340
    """Verifies out of band functionality of a node.
2341

2342
    @type ninfo: L{objects.Node}
2343
    @param ninfo: the node to check
2344
    @param nresult: the remote results for the node
2345

2346
    """
2347
    node = ninfo.name
2348
    # We just have to verify the paths on master and/or master candidates
2349
    # as the oob helper is invoked on the master
2350
    if ((ninfo.master_candidate or ninfo.master_capable) and
2351
        constants.NV_OOB_PATHS in nresult):
2352
      for path_result in nresult[constants.NV_OOB_PATHS]:
2353
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
2354

    
2355
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2356
    """Verifies and updates the node volume data.
2357

2358
    This function will update a L{NodeImage}'s internal structures
2359
    with data from the remote call.
2360

2361
    @type ninfo: L{objects.Node}
2362
    @param ninfo: the node to check
2363
    @param nresult: the remote results for the node
2364
    @param nimg: the node image object
2365
    @param vg_name: the configured VG name
2366

2367
    """
2368
    node = ninfo.name
2369
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2370

    
2371
    nimg.lvm_fail = True
2372
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2373
    if vg_name is None:
2374
      pass
2375
    elif isinstance(lvdata, basestring):
2376
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
2377
               utils.SafeEncode(lvdata))
2378
    elif not isinstance(lvdata, dict):
2379
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
2380
    else:
2381
      nimg.volumes = lvdata
2382
      nimg.lvm_fail = False
2383

    
2384
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2385
    """Verifies and updates the node instance list.
2386

2387
    If the listing was successful, then updates this node's instance
2388
    list. Otherwise, it marks the RPC call as failed for the instance
2389
    list key.
2390

2391
    @type ninfo: L{objects.Node}
2392
    @param ninfo: the node to check
2393
    @param nresult: the remote results for the node
2394
    @param nimg: the node image object
2395

2396
    """
2397
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2398
    test = not isinstance(idata, list)
2399
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
2400
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
2401
    if test:
2402
      nimg.hyp_fail = True
2403
    else:
2404
      nimg.instances = idata
2405

    
2406
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2407
    """Verifies and computes a node information map
2408

2409
    @type ninfo: L{objects.Node}
2410
    @param ninfo: the node to check
2411
    @param nresult: the remote results for the node
2412
    @param nimg: the node image object
2413
    @param vg_name: the configured VG name
2414

2415
    """
2416
    node = ninfo.name
2417
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2418

    
2419
    # try to read free memory (from the hypervisor)
2420
    hv_info = nresult.get(constants.NV_HVINFO, None)
2421
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2422
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
2423
    if not test:
2424
      try:
2425
        nimg.mfree = int(hv_info["memory_free"])
2426
      except (ValueError, TypeError):
2427
        _ErrorIf(True, self.ENODERPC, node,
2428
                 "node returned invalid nodeinfo, check hypervisor")
2429

    
2430
    # FIXME: devise a free space model for file based instances as well
2431
    if vg_name is not None:
2432
      test = (constants.NV_VGLIST not in nresult or
2433
              vg_name not in nresult[constants.NV_VGLIST])
2434
      _ErrorIf(test, self.ENODELVM, node,
2435
               "node didn't return data for the volume group '%s'"
2436
               " - it is either missing or broken", vg_name)
2437
      if not test:
2438
        try:
2439
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2440
        except (ValueError, TypeError):
2441
          _ErrorIf(True, self.ENODERPC, node,
2442
                   "node returned invalid LVM info, check LVM status")
2443

    
2444
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2445
    """Gets per-disk status information for all instances.
2446

2447
    @type nodelist: list of strings
2448
    @param nodelist: Node names
2449
    @type node_image: dict of (name, L{objects.Node})
2450
    @param node_image: Node objects
2451
    @type instanceinfo: dict of (name, L{objects.Instance})
2452
    @param instanceinfo: Instance objects
2453
    @rtype: {instance: {node: [(succes, payload)]}}
2454
    @return: a dictionary of per-instance dictionaries with nodes as
2455
        keys and disk information as values; the disk information is a
2456
        list of tuples (success, payload)
2457

2458
    """
2459
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2460

    
2461
    node_disks = {}
2462
    node_disks_devonly = {}
2463
    diskless_instances = set()
2464
    diskless = constants.DT_DISKLESS
2465

    
2466
    for nname in nodelist:
2467
      node_instances = list(itertools.chain(node_image[nname].pinst,
2468
                                            node_image[nname].sinst))
2469
      diskless_instances.update(inst for inst in node_instances
2470
                                if instanceinfo[inst].disk_template == diskless)
2471
      disks = [(inst, disk)
2472
               for inst in node_instances
2473
               for disk in instanceinfo[inst].disks]
2474

    
2475
      if not disks:
2476
        # No need to collect data
2477
        continue
2478

    
2479
      node_disks[nname] = disks
2480

    
2481
      # Creating copies as SetDiskID below will modify the objects and that can
2482
      # lead to incorrect data returned from nodes
2483
      devonly = [dev.Copy() for (_, dev) in disks]
2484

    
2485
      for dev in devonly:
2486
        self.cfg.SetDiskID(dev, nname)
2487

    
2488
      node_disks_devonly[nname] = devonly
2489

    
2490
    assert len(node_disks) == len(node_disks_devonly)
2491

    
2492
    # Collect data from all nodes with disks
2493
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2494
                                                          node_disks_devonly)
2495

    
2496
    assert len(result) == len(node_disks)
2497

    
2498
    instdisk = {}
2499

    
2500
    for (nname, nres) in result.items():
2501
      disks = node_disks[nname]
2502

    
2503
      if nres.offline:
2504
        # No data from this node
2505
        data = len(disks) * [(False, "node offline")]
2506
      else:
2507
        msg = nres.fail_msg
2508
        _ErrorIf(msg, self.ENODERPC, nname,
2509
                 "while getting disk information: %s", msg)
2510
        if msg:
2511
          # No data from this node
2512
          data = len(disks) * [(False, msg)]
2513
        else:
2514
          data = []
2515
          for idx, i in enumerate(nres.payload):
2516
            if isinstance(i, (tuple, list)) and len(i) == 2:
2517
              data.append(i)
2518
            else:
2519
              logging.warning("Invalid result from node %s, entry %d: %s",
2520
                              nname, idx, i)
2521
              data.append((False, "Invalid result from the remote node"))
2522

    
2523
      for ((inst, _), status) in zip(disks, data):
2524
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2525

    
2526
    # Add empty entries for diskless instances.
2527
    for inst in diskless_instances:
2528
      assert inst not in instdisk
2529
      instdisk[inst] = {}
2530

    
2531
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2532
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2533
                      compat.all(isinstance(s, (tuple, list)) and
2534
                                 len(s) == 2 for s in statuses)
2535
                      for inst, nnames in instdisk.items()
2536
                      for nname, statuses in nnames.items())
2537
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2538

    
2539
    return instdisk
2540

    
2541
  def BuildHooksEnv(self):
2542
    """Build hooks env.
2543

2544
    Cluster-Verify hooks just ran in the post phase and their failure makes
2545
    the output be logged in the verify output and the verification to fail.
2546

2547
    """
2548
    env = {
2549
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2550
      }
2551

    
2552
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2553
               for node in self.my_node_info.values())
2554

    
2555
    return env
2556

    
2557
  def BuildHooksNodes(self):
2558
    """Build hooks nodes.
2559

2560
    """
2561
    return ([], self.my_node_names)
2562

    
2563
  def Exec(self, feedback_fn):
2564
    """Verify integrity of the node group, performing various test on nodes.
2565

2566
    """
2567
    # This method has too many local variables. pylint: disable-msg=R0914
2568

    
2569
    if not self.my_node_names:
2570
      # empty node group
2571
      feedback_fn("* Empty node group, skipping verification")
2572
      return True
2573

    
2574
    self.bad = False
2575
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2576
    verbose = self.op.verbose
2577
    self._feedback_fn = feedback_fn
2578

    
2579
    vg_name = self.cfg.GetVGName()
2580
    drbd_helper = self.cfg.GetDRBDHelper()
2581
    cluster = self.cfg.GetClusterInfo()
2582
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2583
    hypervisors = cluster.enabled_hypervisors
2584
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2585

    
2586
    i_non_redundant = [] # Non redundant instances
2587
    i_non_a_balanced = [] # Non auto-balanced instances
2588
    n_offline = 0 # Count of offline nodes
2589
    n_drained = 0 # Count of nodes being drained
2590
    node_vol_should = {}
2591

    
2592
    # FIXME: verify OS list
2593

    
2594
    # File verification
2595
    filemap = _ComputeAncillaryFiles(cluster, False)
2596

    
2597
    # do local checksums
2598
    master_node = self.master_node = self.cfg.GetMasterNode()
2599
    master_ip = self.cfg.GetMasterIP()
2600

    
2601
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2602

    
2603
    # We will make nodes contact all nodes in their group, and one node from
2604
    # every other group.
2605
    # TODO: should it be a *random* node, different every time?
2606
    online_nodes = [node.name for node in node_data_list if not node.offline]
2607
    other_group_nodes = {}
2608

    
2609
    for name in sorted(self.all_node_info):
2610
      node = self.all_node_info[name]
2611
      if (node.group not in other_group_nodes
2612
          and node.group != self.group_uuid
2613
          and not node.offline):
2614
        other_group_nodes[node.group] = node.name
2615

    
2616
    node_verify_param = {
2617
      constants.NV_FILELIST:
2618
        utils.UniqueSequence(filename
2619
                             for files in filemap
2620
                             for filename in files),
2621
      constants.NV_NODELIST: online_nodes + other_group_nodes.values(),
2622
      constants.NV_HYPERVISOR: hypervisors,
2623
      constants.NV_HVPARAMS:
2624
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2625
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2626
                                 for node in node_data_list
2627
                                 if not node.offline],
2628
      constants.NV_INSTANCELIST: hypervisors,
2629
      constants.NV_VERSION: None,
2630
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2631
      constants.NV_NODESETUP: None,
2632
      constants.NV_TIME: None,
2633
      constants.NV_MASTERIP: (master_node, master_ip),
2634
      constants.NV_OSLIST: None,
2635
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2636
      }
2637

    
2638
    if vg_name is not None:
2639
      node_verify_param[constants.NV_VGLIST] = None
2640
      node_verify_param[constants.NV_LVLIST] = vg_name
2641
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2642
      node_verify_param[constants.NV_DRBDLIST] = None
2643

    
2644
    if drbd_helper:
2645
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2646

    
2647
    # bridge checks
2648
    # FIXME: this needs to be changed per node-group, not cluster-wide
2649
    bridges = set()
2650
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2651
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2652
      bridges.add(default_nicpp[constants.NIC_LINK])
2653
    for instance in self.my_inst_info.values():
2654
      for nic in instance.nics:
2655
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2656
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2657
          bridges.add(full_nic[constants.NIC_LINK])
2658

    
2659
    if bridges:
2660
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2661

    
2662
    # Build our expected cluster state
2663
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2664
                                                 name=node.name,
2665
                                                 vm_capable=node.vm_capable))
2666
                      for node in node_data_list)
2667

    
2668
    # Gather OOB paths
2669
    oob_paths = []
2670
    for node in self.all_node_info.values():
2671
      path = _SupportsOob(self.cfg, node)
2672
      if path and path not in oob_paths:
2673
        oob_paths.append(path)
2674

    
2675
    if oob_paths:
2676
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2677

    
2678
    for instance in self.my_inst_names:
2679
      inst_config = self.my_inst_info[instance]
2680

    
2681
      for nname in inst_config.all_nodes:
2682
        if nname not in node_image:
2683
          gnode = self.NodeImage(name=nname)
2684
          gnode.ghost = (nname not in self.all_node_info)
2685
          node_image[nname] = gnode
2686

    
2687
      inst_config.MapLVsByNode(node_vol_should)
2688

    
2689
      pnode = inst_config.primary_node
2690
      node_image[pnode].pinst.append(instance)
2691

    
2692
      for snode in inst_config.secondary_nodes:
2693
        nimg = node_image[snode]
2694
        nimg.sinst.append(instance)
2695
        if pnode not in nimg.sbp:
2696
          nimg.sbp[pnode] = []
2697
        nimg.sbp[pnode].append(instance)
2698

    
2699
    # At this point, we have the in-memory data structures complete,
2700
    # except for the runtime information, which we'll gather next
2701

    
2702
    # Due to the way our RPC system works, exact response times cannot be
2703
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2704
    # time before and after executing the request, we can at least have a time
2705
    # window.
2706
    nvinfo_starttime = time.time()
2707
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2708
                                           node_verify_param,
2709
                                           self.cfg.GetClusterName())
2710
    nvinfo_endtime = time.time()
2711

    
2712
    if self.extra_lv_nodes and vg_name is not None:
2713
      extra_lv_nvinfo = \
2714
          self.rpc.call_node_verify(self.extra_lv_nodes,
2715
                                    {constants.NV_LVLIST: vg_name},
2716
                                    self.cfg.GetClusterName())
2717
    else:
2718
      extra_lv_nvinfo = {}
2719

    
2720
    all_drbd_map = self.cfg.ComputeDRBDMap()
2721

    
2722
    feedback_fn("* Gathering disk information (%s nodes)" %
2723
                len(self.my_node_names))
2724
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2725
                                     self.my_inst_info)
2726

    
2727
    feedback_fn("* Verifying configuration file consistency")
2728

    
2729
    # If not all nodes are being checked, we need to make sure the master node
2730
    # and a non-checked vm_capable node are in the list.
2731
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2732
    if absent_nodes:
2733
      vf_nvinfo = all_nvinfo.copy()
2734
      vf_node_info = list(self.my_node_info.values())
2735
      additional_nodes = []
2736
      if master_node not in self.my_node_info:
2737
        additional_nodes.append(master_node)
2738
        vf_node_info.append(self.all_node_info[master_node])
2739
      # Add the first vm_capable node we find which is not included
2740
      for node in absent_nodes:
2741
        nodeinfo = self.all_node_info[node]
2742
        if nodeinfo.vm_capable and not nodeinfo.offline:
2743
          additional_nodes.append(node)
2744
          vf_node_info.append(self.all_node_info[node])
2745
          break
2746
      key = constants.NV_FILELIST
2747
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2748
                                                 {key: node_verify_param[key]},
2749
                                                 self.cfg.GetClusterName()))
2750
    else:
2751
      vf_nvinfo = all_nvinfo
2752
      vf_node_info = self.my_node_info.values()
2753

    
2754
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2755

    
2756
    feedback_fn("* Verifying node status")
2757

    
2758
    refos_img = None
2759

    
2760
    for node_i in node_data_list:
2761
      node = node_i.name
2762
      nimg = node_image[node]
2763

    
2764
      if node_i.offline:
2765
        if verbose:
2766
          feedback_fn("* Skipping offline node %s" % (node,))
2767
        n_offline += 1
2768
        continue
2769

    
2770
      if node == master_node:
2771
        ntype = "master"
2772
      elif node_i.master_candidate:
2773
        ntype = "master candidate"
2774
      elif node_i.drained:
2775
        ntype = "drained"
2776
        n_drained += 1
2777
      else:
2778
        ntype = "regular"
2779
      if verbose:
2780
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2781

    
2782
      msg = all_nvinfo[node].fail_msg
2783
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2784
      if msg:
2785
        nimg.rpc_fail = True
2786
        continue
2787

    
2788
      nresult = all_nvinfo[node].payload
2789

    
2790
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2791
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2792
      self._VerifyNodeNetwork(node_i, nresult)
2793
      self._VerifyOob(node_i, nresult)
2794

    
2795
      if nimg.vm_capable:
2796
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2797
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2798
                             all_drbd_map)
2799

    
2800
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2801
        self._UpdateNodeInstances(node_i, nresult, nimg)
2802
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2803
        self._UpdateNodeOS(node_i, nresult, nimg)
2804

    
2805
        if not nimg.os_fail:
2806
          if refos_img is None:
2807
            refos_img = nimg
2808
          self._VerifyNodeOS(node_i, nimg, refos_img)
2809
        self._VerifyNodeBridges(node_i, nresult, bridges)
2810

    
2811
        # Check whether all running instancies are primary for the node. (This
2812
        # can no longer be done from _VerifyInstance below, since some of the
2813
        # wrong instances could be from other node groups.)
2814
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2815

    
2816
        for inst in non_primary_inst:
2817
          test = inst in self.all_inst_info
2818
          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
2819
                   "instance should not run on node %s", node_i.name)
2820
          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
2821
                   "node is running unknown instance %s", inst)
2822

    
2823
    for node, result in extra_lv_nvinfo.items():
2824
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
2825
                              node_image[node], vg_name)
2826

    
2827
    feedback_fn("* Verifying instance status")
2828
    for instance in self.my_inst_names:
2829
      if verbose:
2830
        feedback_fn("* Verifying instance %s" % instance)
2831
      inst_config = self.my_inst_info[instance]
2832
      self._VerifyInstance(instance, inst_config, node_image,
2833
                           instdisk[instance])
2834
      inst_nodes_offline = []
2835

    
2836
      pnode = inst_config.primary_node
2837
      pnode_img = node_image[pnode]
2838
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2839
               self.ENODERPC, pnode, "instance %s, connection to"
2840
               " primary node failed", instance)
2841

    
2842
      _ErrorIf(inst_config.admin_up and pnode_img.offline,
2843
               self.EINSTANCEBADNODE, instance,
2844
               "instance is marked as running and lives on offline node %s",
2845
               inst_config.primary_node)
2846

    
2847
      # If the instance is non-redundant we cannot survive losing its primary
2848
      # node, so we are not N+1 compliant. On the other hand we have no disk
2849
      # templates with more than one secondary so that situation is not well
2850
      # supported either.
2851
      # FIXME: does not support file-backed instances
2852
      if not inst_config.secondary_nodes:
2853
        i_non_redundant.append(instance)
2854

    
2855
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2856
               instance, "instance has multiple secondary nodes: %s",
2857
               utils.CommaJoin(inst_config.secondary_nodes),
2858
               code=self.ETYPE_WARNING)
2859

    
2860
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2861
        pnode = inst_config.primary_node
2862
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2863
        instance_groups = {}
2864

    
2865
        for node in instance_nodes:
2866
          instance_groups.setdefault(self.all_node_info[node].group,
2867
                                     []).append(node)
2868

    
2869
        pretty_list = [
2870
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2871
          # Sort so that we always list the primary node first.
2872
          for group, nodes in sorted(instance_groups.items(),
2873
                                     key=lambda (_, nodes): pnode in nodes,
2874
                                     reverse=True)]
2875

    
2876
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2877
                      instance, "instance has primary and secondary nodes in"
2878
                      " different groups: %s", utils.CommaJoin(pretty_list),
2879
                      code=self.ETYPE_WARNING)
2880

    
2881
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2882
        i_non_a_balanced.append(instance)
2883

    
2884
      for snode in inst_config.secondary_nodes:
2885
        s_img = node_image[snode]
2886
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2887
                 "instance %s, connection to secondary node failed", instance)
2888

    
2889
        if s_img.offline:
2890
          inst_nodes_offline.append(snode)
2891

    
2892
      # warn that the instance lives on offline nodes
2893
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2894
               "instance has offline secondary node(s) %s",
2895
               utils.CommaJoin(inst_nodes_offline))
2896
      # ... or ghost/non-vm_capable nodes
2897
      for node in inst_config.all_nodes:
2898
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2899
                 "instance lives on ghost node %s", node)
2900
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2901
                 instance, "instance lives on non-vm_capable node %s", node)
2902

    
2903
    feedback_fn("* Verifying orphan volumes")
2904
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2905

    
2906
    # We will get spurious "unknown volume" warnings if any node of this group
2907
    # is secondary for an instance whose primary is in another group. To avoid
2908
    # them, we find these instances and add their volumes to node_vol_should.
2909
    for inst in self.all_inst_info.values():
2910
      for secondary in inst.secondary_nodes:
2911
        if (secondary in self.my_node_info
2912
            and inst.name not in self.my_inst_info):
2913
          inst.MapLVsByNode(node_vol_should)
2914
          break
2915

    
2916
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2917

    
2918
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2919
      feedback_fn("* Verifying N+1 Memory redundancy")
2920
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
2921

    
2922
    feedback_fn("* Other Notes")
2923
    if i_non_redundant:
2924
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2925
                  % len(i_non_redundant))
2926

    
2927
    if i_non_a_balanced:
2928
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2929
                  % len(i_non_a_balanced))
2930

    
2931
    if n_offline:
2932
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2933

    
2934
    if n_drained:
2935
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2936

    
2937
    return not self.bad
2938

    
2939
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2940
    """Analyze the post-hooks' result
2941

2942
    This method analyses the hook result, handles it, and sends some
2943
    nicely-formatted feedback back to the user.
2944

2945
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
2946
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
2947
    @param hooks_results: the results of the multi-node hooks rpc call
2948
    @param feedback_fn: function used send feedback back to the caller
2949
    @param lu_result: previous Exec result
2950
    @return: the new Exec result, based on the previous result
2951
        and hook results
2952

2953
    """
2954
    # We only really run POST phase hooks, only for non-empty groups,
2955
    # and are only interested in their results
2956
    if not self.my_node_names:
2957
      # empty node group
2958
      pass
2959
    elif phase == constants.HOOKS_PHASE_POST:
2960
      # Used to change hooks' output to proper indentation
2961
      feedback_fn("* Hooks Results")
2962
      assert hooks_results, "invalid result from hooks"
2963

    
2964
      for node_name in hooks_results:
2965
        res = hooks_results[node_name]
2966
        msg = res.fail_msg
2967
        test = msg and not res.offline
2968
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
2969
                      "Communication failure in hooks execution: %s", msg)
2970
        if res.offline or msg:
2971
          # No need to investigate payload if node is offline or gave an error.
2972
          # override manually lu_result here as _ErrorIf only
2973
          # overrides self.bad
2974
          lu_result = 1
2975
          continue
2976
        for script, hkr, output in res.payload:
2977
          test = hkr == constants.HKR_FAIL
2978
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
2979
                        "Script %s failed, output:", script)
2980
          if test:
2981
            output = self._HOOKS_INDENT_RE.sub("      ", output)
2982
            feedback_fn("%s" % output)
2983
            lu_result = 0
2984

    
2985
    return lu_result
2986

    
2987

    
2988
class LUClusterVerifyDisks(NoHooksLU):
2989
  """Verifies the cluster disks status.
2990

2991
  """
2992
  REQ_BGL = False
2993

    
2994
  def ExpandNames(self):
2995
    self.share_locks = _ShareAll()
2996
    self.needed_locks = {
2997
      locking.LEVEL_NODEGROUP: locking.ALL_SET,
2998
      }
2999

    
3000
  def Exec(self, feedback_fn):
3001
    group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
3002

    
3003
    # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
3004
    return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
3005
                           for group in group_names])
3006

    
3007

    
3008
class LUGroupVerifyDisks(NoHooksLU):
3009
  """Verifies the status of all disks in a node group.
3010

3011
  """
3012
  REQ_BGL = False
3013

    
3014
  def ExpandNames(self):
3015
    # Raises errors.OpPrereqError on its own if group can't be found
3016
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
3017

    
3018
    self.share_locks = _ShareAll()
3019
    self.needed_locks = {
3020
      locking.LEVEL_INSTANCE: [],
3021
      locking.LEVEL_NODEGROUP: [],
3022
      locking.LEVEL_NODE: [],
3023
      }
3024

    
3025
  def DeclareLocks(self, level):
3026
    if level == locking.LEVEL_INSTANCE:
3027
      assert not self.needed_locks[locking.LEVEL_INSTANCE]
3028

    
3029
      # Lock instances optimistically, needs verification once node and group
3030
      # locks have been acquired
3031
      self.needed_locks[locking.LEVEL_INSTANCE] = \
3032
        self.cfg.GetNodeGroupInstances(self.group_uuid)
3033

    
3034
    elif level == locking.LEVEL_NODEGROUP:
3035
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
3036

    
3037
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
3038
        set([self.group_uuid] +
3039
            # Lock all groups used by instances optimistically; this requires
3040
            # going via the node before it's locked, requiring verification
3041
            # later on
3042
            [group_uuid
3043
             for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
3044
             for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
3045

    
3046
    elif level == locking.LEVEL_NODE:
3047
      # This will only lock the nodes in the group to be verified which contain
3048
      # actual instances
3049
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
3050
      self._LockInstancesNodes()
3051

    
3052
      # Lock all nodes in group to be verified
3053
      assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
3054
      member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
3055
      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
3056

    
3057
  def CheckPrereq(self):
3058
    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
3059
    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
3060
    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
3061

    
3062
    assert self.group_uuid in owned_groups
3063

    
3064
    # Check if locked instances are still correct
3065
    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
3066

    
3067
    # Get instance information
3068
    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
3069

    
3070
    # Check if node groups for locked instances are still correct
3071
    for (instance_name, inst) in self.instances.items():
3072
      assert owned_nodes.issuperset(inst.all_nodes), \
3073
        "Instance %s's nodes changed while we kept the lock" % instance_name
3074

    
3075
      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
3076
                                             owned_groups)
3077

    
3078
      assert self.group_uuid in inst_groups, \
3079
        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
3080

    
3081
  def Exec(self, feedback_fn):
3082
    """Verify integrity of cluster disks.
3083

3084
    @rtype: tuple of three items
3085
    @return: a tuple of (dict of node-to-node_error, list of instances
3086
        which need activate-disks, dict of instance: (node, volume) for
3087
        missing volumes
3088

3089
    """
3090
    res_nodes = {}
3091
    res_instances = set()
3092
    res_missing = {}
3093

    
3094
    nv_dict = _MapInstanceDisksToNodes([inst
3095
                                        for inst in self.instances.values()
3096
                                        if inst.admin_up])
3097

    
3098
    if nv_dict:
3099
      nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
3100
                             set(self.cfg.GetVmCapableNodeList()))
3101

    
3102
      node_lvs = self.rpc.call_lv_list(nodes, [])
3103

    
3104
      for (node, node_res) in node_lvs.items():
3105
        if node_res.offline:
3106
          continue
3107

    
3108
        msg = node_res.fail_msg
3109
        if msg:
3110
          logging.warning("Error enumerating LVs on node %s: %s", node, msg)
3111
          res_nodes[node] = msg
3112
          continue
3113

    
3114
        for lv_name, (_, _, lv_online) in node_res.payload.items():
3115
          inst = nv_dict.pop((node, lv_name), None)
3116
          if not (lv_online or inst is None):
3117
            res_instances.add(inst)
3118

    
3119
      # any leftover items in nv_dict are missing LVs, let's arrange the data
3120
      # better
3121
      for key, inst in nv_dict.iteritems():
3122
        res_missing.setdefault(inst, []).append(key)
3123

    
3124
    return (res_nodes, list(res_instances), res_missing)
3125

    
3126

    
3127
class LUClusterRepairDiskSizes(NoHooksLU):
3128
  """Verifies the cluster disks sizes.
3129

3130
  """
3131
  REQ_BGL = False
3132

    
3133
  def ExpandNames(self):
3134
    if self.op.instances:
3135
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
3136
      self.needed_locks = {
3137
        locking.LEVEL_NODE: [],
3138
        locking.LEVEL_INSTANCE: self.wanted_names,
3139
        }
3140
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3141
    else:
3142
      self.wanted_names = None
3143
      self.needed_locks = {
3144
        locking.LEVEL_NODE: locking.ALL_SET,
3145
        locking.LEVEL_INSTANCE: locking.ALL_SET,
3146
        }
3147
    self.share_locks = _ShareAll()
3148

    
3149
  def DeclareLocks(self, level):
3150
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
3151
      self._LockInstancesNodes(primary_only=True)
3152

    
3153
  def CheckPrereq(self):
3154
    """Check prerequisites.
3155

3156
    This only checks the optional instance list against the existing names.
3157

3158
    """
3159
    if self.wanted_names is None:
3160
      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
3161

    
3162
    self.wanted_instances = \
3163
        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
3164

    
3165
  def _EnsureChildSizes(self, disk):
3166
    """Ensure children of the disk have the needed disk size.
3167

3168
    This is valid mainly for DRBD8 and fixes an issue where the
3169
    children have smaller disk size.
3170

3171
    @param disk: an L{ganeti.objects.Disk} object
3172

3173
    """
3174
    if disk.dev_type == constants.LD_DRBD8:
3175
      assert disk.children, "Empty children for DRBD8?"
3176
      fchild = disk.children[0]
3177
      mismatch = fchild.size < disk.size
3178
      if mismatch:
3179
        self.LogInfo("Child disk has size %d, parent %d, fixing",
3180
                     fchild.size, disk.size)
3181
        fchild.size = disk.size
3182

    
3183
      # and we recurse on this child only, not on the metadev
3184
      return self._EnsureChildSizes(fchild) or mismatch
3185
    else:
3186
      return False
3187

    
3188
  def Exec(self, feedback_fn):
3189
    """Verify the size of cluster disks.
3190

3191
    """
3192
    # TODO: check child disks too
3193
    # TODO: check differences in size between primary/secondary nodes
3194
    per_node_disks = {}
3195
    for instance in self.wanted_instances:
3196
      pnode = instance.primary_node
3197
      if pnode not in per_node_disks:
3198
        per_node_disks[pnode] = []
3199
      for idx, disk in enumerate(instance.disks):
3200
        per_node_disks[pnode].append((instance, idx, disk))
3201

    
3202
    changed = []
3203
    for node, dskl in per_node_disks.items():
3204
      newl = [v[2].Copy() for v in dskl]
3205
      for dsk in newl:
3206
        self.cfg.SetDiskID(dsk, node)
3207
      result = self.rpc.call_blockdev_getsize(node, newl)
3208
      if result.fail_msg:
3209
        self.LogWarning("Failure in blockdev_getsize call to node"
3210
                        " %s, ignoring", node)
3211
        continue
3212
      if len(result.payload) != len(dskl):
3213
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
3214
                        " result.payload=%s", node, len(dskl), result.payload)
3215
        self.LogWarning("Invalid result from node %s, ignoring node results",
3216
                        node)
3217
        continue
3218
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
3219
        if size is None:
3220
          self.LogWarning("Disk %d of instance %s did not return size"
3221
                          " information, ignoring", idx, instance.name)
3222
          continue
3223
        if not isinstance(size, (int, long)):
3224
          self.LogWarning("Disk %d of instance %s did not return valid"
3225
                          " size information, ignoring", idx, instance.name)
3226
          continue
3227
        size = size >> 20
3228
        if size != disk.size:
3229
          self.LogInfo("Disk %d of instance %s has mismatched size,"
3230
                       " correcting: recorded %d, actual %d", idx,
3231
                       instance.name, disk.size, size)
3232
          disk.size = size
3233
          self.cfg.Update(instance, feedback_fn)
3234
          changed.append((instance.name, idx, size))
3235
        if self._EnsureChildSizes(disk):
3236
          self.cfg.Update(instance, feedback_fn)
3237
          changed.append((instance.name, idx, disk.size))
3238
    return changed
3239

    
3240

    
3241
class LUClusterRename(LogicalUnit):
3242
  """Rename the cluster.
3243

3244
  """
3245
  HPATH = "cluster-rename"
3246
  HTYPE = constants.HTYPE_CLUSTER
3247

    
3248
  def BuildHooksEnv(self):
3249
    """Build hooks env.
3250

3251
    """
3252
    return {
3253
      "OP_TARGET": self.cfg.GetClusterName(),
3254
      "NEW_NAME": self.op.name,
3255
      }
3256

    
3257
  def BuildHooksNodes(self):
3258
    """Build hooks nodes.
3259

3260
    """
3261
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3262

    
3263
  def CheckPrereq(self):
3264
    """Verify that the passed name is a valid one.
3265

3266
    """
3267
    hostname = netutils.GetHostname(name=self.op.name,
3268
                                    family=self.cfg.GetPrimaryIPFamily())
3269

    
3270
    new_name = hostname.name
3271
    self.ip = new_ip = hostname.ip
3272
    old_name = self.cfg.GetClusterName()
3273
    old_ip = self.cfg.GetMasterIP()
3274
    if new_name == old_name and new_ip == old_ip:
3275
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3276
                                 " cluster has changed",
3277
                                 errors.ECODE_INVAL)
3278
    if new_ip != old_ip:
3279
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3280
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3281
                                   " reachable on the network" %
3282
                                   new_ip, errors.ECODE_NOTUNIQUE)
3283

    
3284
    self.op.name = new_name
3285

    
3286
  def Exec(self, feedback_fn):
3287
    """Rename the cluster.
3288

3289
    """
3290
    clustername = self.op.name
3291
    ip = self.ip
3292

    
3293
    # shutdown the master IP
3294
    master = self.cfg.GetMasterNode()
3295
    result = self.rpc.call_node_stop_master(master, False)
3296
    result.Raise("Could not disable the master role")
3297

    
3298
    try:
3299
      cluster = self.cfg.GetClusterInfo()
3300
      cluster.cluster_name = clustername
3301
      cluster.master_ip = ip
3302
      self.cfg.Update(cluster, feedback_fn)
3303

    
3304
      # update the known hosts file
3305
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3306
      node_list = self.cfg.GetOnlineNodeList()
3307
      try:
3308
        node_list.remove(master)
3309
      except ValueError:
3310
        pass
3311
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3312
    finally:
3313
      result = self.rpc.call_node_start_master(master, False, False)
3314
      msg = result.fail_msg
3315
      if msg:
3316
        self.LogWarning("Could not re-enable the master role on"
3317
                        " the master, please restart manually: %s", msg)
3318

    
3319
    return clustername
3320

    
3321

    
3322
class LUClusterSetParams(LogicalUnit):
3323
  """Change the parameters of the cluster.
3324

3325
  """
3326
  HPATH = "cluster-modify"
3327
  HTYPE = constants.HTYPE_CLUSTER
3328
  REQ_BGL = False
3329

    
3330
  def CheckArguments(self):
3331
    """Check parameters
3332

3333
    """
3334
    if self.op.uid_pool:
3335
      uidpool.CheckUidPool(self.op.uid_pool)
3336

    
3337
    if self.op.add_uids:
3338
      uidpool.CheckUidPool(self.op.add_uids)
3339

    
3340
    if self.op.remove_uids:
3341
      uidpool.CheckUidPool(self.op.remove_uids)
3342

    
3343
  def ExpandNames(self):
3344
    # FIXME: in the future maybe other cluster params won't require checking on
3345
    # all nodes to be modified.
3346
    self.needed_locks = {
3347
      locking.LEVEL_NODE: locking.ALL_SET,
3348
    }
3349
    self.share_locks[locking.LEVEL_NODE] = 1
3350

    
3351
  def BuildHooksEnv(self):
3352
    """Build hooks env.
3353

3354
    """
3355
    return {
3356
      "OP_TARGET": self.cfg.GetClusterName(),
3357
      "NEW_VG_NAME": self.op.vg_name,
3358
      }
3359

    
3360
  def BuildHooksNodes(self):
3361
    """Build hooks nodes.
3362

3363
    """
3364
    mn = self.cfg.GetMasterNode()
3365
    return ([mn], [mn])
3366

    
3367
  def CheckPrereq(self):
3368
    """Check prerequisites.
3369

3370
    This checks whether the given params don't conflict and
3371
    if the given volume group is valid.
3372

3373
    """
3374
    if self.op.vg_name is not None and not self.op.vg_name:
3375
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3376
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3377
                                   " instances exist", errors.ECODE_INVAL)
3378

    
3379
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3380
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3381
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3382
                                   " drbd-based instances exist",
3383
                                   errors.ECODE_INVAL)
3384

    
3385
    node_list = self.owned_locks(locking.LEVEL_NODE)
3386

    
3387
    # if vg_name not None, checks given volume group on all nodes
3388
    if self.op.vg_name:
3389
      vglist = self.rpc.call_vg_list(node_list)
3390
      for node in node_list:
3391
        msg = vglist[node].fail_msg
3392
        if msg:
3393
          # ignoring down node
3394
          self.LogWarning("Error while gathering data on node %s"
3395
                          " (ignoring node): %s", node, msg)
3396
          continue
3397
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3398
                                              self.op.vg_name,
3399
                                              constants.MIN_VG_SIZE)
3400
        if vgstatus:
3401
          raise errors.OpPrereqError("Error on node '%s': %s" %
3402
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3403

    
3404
    if self.op.drbd_helper:
3405
      # checks given drbd helper on all nodes
3406
      helpers = self.rpc.call_drbd_helper(node_list)
3407
      for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
3408
        if ninfo.offline:
3409
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3410
          continue
3411
        msg = helpers[node].fail_msg
3412
        if msg:
3413
          raise errors.OpPrereqError("Error checking drbd helper on node"
3414
                                     " '%s': %s" % (node, msg),
3415
                                     errors.ECODE_ENVIRON)
3416
        node_helper = helpers[node].payload
3417
        if node_helper != self.op.drbd_helper:
3418
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3419
                                     (node, node_helper), errors.ECODE_ENVIRON)
3420

    
3421
    self.cluster = cluster = self.cfg.GetClusterInfo()
3422
    # validate params changes
3423
    if self.op.beparams:
3424
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3425
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3426

    
3427
    if self.op.ndparams:
3428
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3429
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3430

    
3431
      # TODO: we need a more general way to handle resetting
3432
      # cluster-level parameters to default values
3433
      if self.new_ndparams["oob_program"] == "":
3434
        self.new_ndparams["oob_program"] = \
3435
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3436

    
3437
    if self.op.nicparams:
3438
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3439
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3440
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3441
      nic_errors = []
3442

    
3443
      # check all instances for consistency
3444
      for instance in self.cfg.GetAllInstancesInfo().values():
3445
        for nic_idx, nic in enumerate(instance.nics):
3446
          params_copy = copy.deepcopy(nic.nicparams)
3447
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3448

    
3449
          # check parameter syntax
3450
          try:
3451
            objects.NIC.CheckParameterSyntax(params_filled)
3452
          except errors.ConfigurationError, err:
3453
            nic_errors.append("Instance %s, nic/%d: %s" %
3454
                              (instance.name, nic_idx, err))
3455

    
3456
          # if we're moving instances to routed, check that they have an ip
3457
          target_mode = params_filled[constants.NIC_MODE]
3458
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3459
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3460
                              " address" % (instance.name, nic_idx))
3461
      if nic_errors:
3462
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3463
                                   "\n".join(nic_errors))
3464

    
3465
    # hypervisor list/parameters
3466
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3467
    if self.op.hvparams:
3468
      for hv_name, hv_dict in self.op.hvparams.items():
3469
        if hv_name not in self.new_hvparams:
3470
          self.new_hvparams[hv_name] = hv_dict
3471
        else:
3472
          self.new_hvparams[hv_name].update(hv_dict)
3473

    
3474
    # os hypervisor parameters
3475
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3476
    if self.op.os_hvp:
3477
      for os_name, hvs in self.op.os_hvp.items():
3478
        if os_name not in self.new_os_hvp:
3479
          self.new_os_hvp[os_name] = hvs
3480
        else:
3481
          for hv_name, hv_dict in hvs.items():
3482
            if hv_name not in self.new_os_hvp[os_name]:
3483
              self.new_os_hvp[os_name][hv_name] = hv_dict
3484
            else:
3485
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3486

    
3487
    # os parameters
3488
    self.new_osp = objects.FillDict(cluster.osparams, {})
3489
    if self.op.osparams:
3490
      for os_name, osp in self.op.osparams.items():
3491
        if os_name not in self.new_osp:
3492
          self.new_osp[os_name] = {}
3493

    
3494
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3495
                                                  use_none=True)
3496

    
3497
        if not self.new_osp[os_name]:
3498
          # we removed all parameters
3499
          del self.new_osp[os_name]
3500
        else:
3501
          # check the parameter validity (remote check)
3502
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3503
                         os_name, self.new_osp[os_name])
3504

    
3505
    # changes to the hypervisor list
3506
    if self.op.enabled_hypervisors is not None:
3507
      self.hv_list = self.op.enabled_hypervisors
3508
      for hv in self.hv_list:
3509
        # if the hypervisor doesn't already exist in the cluster
3510
        # hvparams, we initialize it to empty, and then (in both
3511
        # cases) we make sure to fill the defaults, as we might not
3512
        # have a complete defaults list if the hypervisor wasn't
3513
        # enabled before
3514
        if hv not in new_hvp:
3515
          new_hvp[hv] = {}
3516
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3517
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3518
    else:
3519
      self.hv_list = cluster.enabled_hypervisors
3520

    
3521
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3522
      # either the enabled list has changed, or the parameters have, validate
3523
      for hv_name, hv_params in self.new_hvparams.items():
3524
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3525
            (self.op.enabled_hypervisors and
3526
             hv_name in self.op.enabled_hypervisors)):
3527
          # either this is a new hypervisor, or its parameters have changed
3528
          hv_class = hypervisor.GetHypervisor(hv_name)
3529
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3530
          hv_class.CheckParameterSyntax(hv_params)
3531
          _CheckHVParams(self, node_list, hv_name, hv_params)
3532

    
3533
    if self.op.os_hvp:
3534
      # no need to check any newly-enabled hypervisors, since the
3535
      # defaults have already been checked in the above code-block
3536
      for os_name, os_hvp in self.new_os_hvp.items():
3537
        for hv_name, hv_params in os_hvp.items():
3538
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3539
          # we need to fill in the new os_hvp on top of the actual hv_p
3540
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3541
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3542
          hv_class = hypervisor.GetHypervisor(hv_name)
3543
          hv_class.CheckParameterSyntax(new_osp)
3544
          _CheckHVParams(self, node_list, hv_name, new_osp)
3545

    
3546
    if self.op.default_iallocator:
3547
      alloc_script = utils.FindFile(self.op.default_iallocator,
3548
                                    constants.IALLOCATOR_SEARCH_PATH,
3549
                                    os.path.isfile)
3550
      if alloc_script is None:
3551
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3552
                                   " specified" % self.op.default_iallocator,
3553
                                   errors.ECODE_INVAL)
3554

    
3555
  def Exec(self, feedback_fn):
3556
    """Change the parameters of the cluster.
3557

3558
    """
3559
    if self.op.vg_name is not None:
3560
      new_volume = self.op.vg_name
3561
      if not new_volume:
3562
        new_volume = None
3563
      if new_volume != self.cfg.GetVGName():
3564
        self.cfg.SetVGName(new_volume)
3565
      else:
3566
        feedback_fn("Cluster LVM configuration already in desired"
3567
                    " state, not changing")
3568
    if self.op.drbd_helper is not None:
3569
      new_helper = self.op.drbd_helper
3570
      if not new_helper:
3571
        new_helper = None
3572
      if new_helper != self.cfg.GetDRBDHelper():
3573
        self.cfg.SetDRBDHelper(new_helper)
3574
      else:
3575
        feedback_fn("Cluster DRBD helper already in desired state,"
3576
                    " not changing")
3577
    if self.op.hvparams:
3578
      self.cluster.hvparams = self.new_hvparams
3579
    if self.op.os_hvp:
3580
      self.cluster.os_hvp = self.new_os_hvp
3581
    if self.op.enabled_hypervisors is not None:
3582
      self.cluster.hvparams = self.new_hvparams
3583
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3584
    if self.op.beparams:
3585
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3586
    if self.op.nicparams:
3587
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3588
    if self.op.osparams:
3589
      self.cluster.osparams = self.new_osp
3590
    if self.op.ndparams:
3591
      self.cluster.ndparams = self.new_ndparams
3592

    
3593
    if self.op.candidate_pool_size is not None:
3594
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3595
      # we need to update the pool size here, otherwise the save will fail
3596
      _AdjustCandidatePool(self, [])
3597

    
3598
    if self.op.maintain_node_health is not None:
3599
      self.cluster.maintain_node_health = self.op.maintain_node_health
3600

    
3601
    if self.op.prealloc_wipe_disks is not None:
3602
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3603

    
3604
    if self.op.add_uids is not None:
3605
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3606

    
3607
    if self.op.remove_uids is not None:
3608
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3609

    
3610
    if self.op.uid_pool is not None:
3611
      self.cluster.uid_pool = self.op.uid_pool
3612

    
3613
    if self.op.default_iallocator is not None:
3614
      self.cluster.default_iallocator = self.op.default_iallocator
3615

    
3616
    if self.op.reserved_lvs is not None:
3617
      self.cluster.reserved_lvs = self.op.reserved_lvs
3618

    
3619
    def helper_os(aname, mods, desc):
3620
      desc += " OS list"
3621
      lst = getattr(self.cluster, aname)
3622
      for key, val in mods:
3623
        if key == constants.DDM_ADD:
3624
          if val in lst:
3625
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3626
          else:
3627
            lst.append(val)
3628
        elif key == constants.DDM_REMOVE:
3629
          if val in lst:
3630
            lst.remove(val)
3631
          else:
3632
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3633
        else:
3634
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3635

    
3636
    if self.op.hidden_os:
3637
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3638

    
3639
    if self.op.blacklisted_os:
3640
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3641

    
3642
    if self.op.master_netdev:
3643
      master = self.cfg.GetMasterNode()
3644
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3645
                  self.cluster.master_netdev)
3646
      result = self.rpc.call_node_stop_master(master, False)
3647
      result.Raise("Could not disable the master ip")
3648
      feedback_fn("Changing master_netdev from %s to %s" %
3649
                  (self.cluster.master_netdev, self.op.master_netdev))
3650
      self.cluster.master_netdev = self.op.master_netdev
3651

    
3652
    self.cfg.Update(self.cluster, feedback_fn)
3653

    
3654
    if self.op.master_netdev:
3655
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3656
                  self.op.master_netdev)
3657
      result = self.rpc.call_node_start_master(master, False, False)
3658
      if result.fail_msg:
3659
        self.LogWarning("Could not re-enable the master ip on"
3660
                        " the master, please restart manually: %s",
3661
                        result.fail_msg)
3662

    
3663

    
3664
def _UploadHelper(lu, nodes, fname):
3665
  """Helper for uploading a file and showing warnings.
3666

3667
  """
3668
  if os.path.exists(fname):
3669
    result = lu.rpc.call_upload_file(nodes, fname)
3670
    for to_node, to_result in result.items():
3671
      msg = to_result.fail_msg
3672
      if msg:
3673
        msg = ("Copy of file %s to node %s failed: %s" %
3674
               (fname, to_node, msg))
3675
        lu.proc.LogWarning(msg)
3676

    
3677

    
3678
def _ComputeAncillaryFiles(cluster, redist):
3679
  """Compute files external to Ganeti which need to be consistent.
3680

3681
  @type redist: boolean
3682
  @param redist: Whether to include files which need to be redistributed
3683

3684
  """
3685
  # Compute files for all nodes
3686
  files_all = set([
3687
    constants.SSH_KNOWN_HOSTS_FILE,
3688
    constants.CONFD_HMAC_KEY,
3689
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3690
    ])
3691

    
3692
  if not redist:
3693
    files_all.update(constants.ALL_CERT_FILES)
3694
    files_all.update(ssconf.SimpleStore().GetFileList())
3695

    
3696
  if cluster.modify_etc_hosts:
3697
    files_all.add(constants.ETC_HOSTS)
3698

    
3699
  # Files which must either exist on all nodes or on none
3700
  files_all_opt = set([
3701
    constants.RAPI_USERS_FILE,
3702
    ])
3703

    
3704
  # Files which should only be on master candidates
3705
  files_mc = set()
3706
  if not redist:
3707
    files_mc.add(constants.CLUSTER_CONF_FILE)
3708

    
3709
  # Files which should only be on VM-capable nodes
3710
  files_vm = set(filename
3711
    for hv_name in cluster.enabled_hypervisors
3712
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
3713

    
3714
  # Filenames must be unique
3715
  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
3716
          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
3717
         "Found file listed in more than one file list"
3718

    
3719
  return (files_all, files_all_opt, files_mc, files_vm)
3720

    
3721

    
3722
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3723
  """Distribute additional files which are part of the cluster configuration.
3724

3725
  ConfigWriter takes care of distributing the config and ssconf files, but
3726
  there are more files which should be distributed to all nodes. This function
3727
  makes sure those are copied.
3728

3729
  @param lu: calling logical unit
3730
  @param additional_nodes: list of nodes not in the config to distribute to
3731
  @type additional_vm: boolean
3732
  @param additional_vm: whether the additional nodes are vm-capable or not
3733

3734
  """
3735
  # Gather target nodes
3736
  cluster = lu.cfg.GetClusterInfo()
3737
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3738

    
3739
  online_nodes = lu.cfg.GetOnlineNodeList()
3740
  vm_nodes = lu.cfg.GetVmCapableNodeList()
3741

    
3742
  if additional_nodes is not None:
3743
    online_nodes.extend(additional_nodes)
3744
    if additional_vm:
3745
      vm_nodes.extend(additional_nodes)
3746

    
3747
  # Never distribute to master node
3748
  for nodelist in [online_nodes, vm_nodes]:
3749
    if master_info.name in nodelist:
3750
      nodelist.remove(master_info.name)
3751

    
3752
  # Gather file lists
3753
  (files_all, files_all_opt, files_mc, files_vm) = \
3754
    _ComputeAncillaryFiles(cluster, True)
3755

    
3756
  # Never re-distribute configuration file from here
3757
  assert not (constants.CLUSTER_CONF_FILE in files_all or
3758
              constants.CLUSTER_CONF_FILE in files_vm)
3759
  assert not files_mc, "Master candidates not handled in this function"
3760

    
3761
  filemap = [
3762
    (online_nodes, files_all),
3763
    (online_nodes, files_all_opt),
3764
    (vm_nodes, files_vm),
3765
    ]
3766

    
3767
  # Upload the files
3768
  for (node_list, files) in filemap:
3769
    for fname in files:
3770
      _UploadHelper(lu, node_list, fname)
3771

    
3772

    
3773
class LUClusterRedistConf(NoHooksLU):
3774
  """Force the redistribution of cluster configuration.
3775

3776
  This is a very simple LU.
3777

3778
  """
3779
  REQ_BGL = False
3780

    
3781
  def ExpandNames(self):
3782
    self.needed_locks = {
3783
      locking.LEVEL_NODE: locking.ALL_SET,
3784
    }
3785
    self.share_locks[locking.LEVEL_NODE] = 1
3786

    
3787
  def Exec(self, feedback_fn):
3788
    """Redistribute the configuration.
3789

3790
    """
3791
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3792
    _RedistributeAncillaryFiles(self)
3793

    
3794

    
3795
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3796
  """Sleep and poll for an instance's disk to sync.
3797

3798
  """
3799
  if not instance.disks or disks is not None and not disks:
3800
    return True
3801

    
3802
  disks = _ExpandCheckDisks(instance, disks)
3803

    
3804
  if not oneshot:
3805
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3806

    
3807
  node = instance.primary_node
3808

    
3809
  for dev in disks:
3810
    lu.cfg.SetDiskID(dev, node)
3811

    
3812
  # TODO: Convert to utils.Retry
3813

    
3814
  retries = 0
3815
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3816
  while True:
3817
    max_time = 0
3818
    done = True
3819
    cumul_degraded = False
3820
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3821
    msg = rstats.fail_msg
3822
    if msg:
3823
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3824
      retries += 1
3825
      if retries >= 10:
3826
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3827
                                 " aborting." % node)
3828
      time.sleep(6)
3829
      continue
3830
    rstats = rstats.payload
3831
    retries = 0
3832
    for i, mstat in enumerate(rstats):
3833
      if mstat is None:
3834
        lu.LogWarning("Can't compute data for node %s/%s",
3835
                           node, disks[i].iv_name)
3836
        continue
3837

    
3838
      cumul_degraded = (cumul_degraded or
3839
                        (mstat.is_degraded and mstat.sync_percent is None))
3840
      if mstat.sync_percent is not None:
3841
        done = False
3842
        if mstat.estimated_time is not None:
3843
          rem_time = ("%s remaining (estimated)" %
3844
                      utils.FormatSeconds(mstat.estimated_time))
3845
          max_time = mstat.estimated_time
3846
        else:
3847
          rem_time = "no time estimate"
3848
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3849
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3850

    
3851
    # if we're done but degraded, let's do a few small retries, to
3852
    # make sure we see a stable and not transient situation; therefore
3853
    # we force restart of the loop
3854
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3855
      logging.info("Degraded disks found, %d retries left", degr_retries)
3856
      degr_retries -= 1
3857
      time.sleep(1)
3858
      continue
3859

    
3860
    if done or oneshot:
3861
      break
3862

    
3863
    time.sleep(min(60, max_time))
3864

    
3865
  if done:
3866
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3867
  return not cumul_degraded
3868

    
3869

    
3870
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3871
  """Check that mirrors are not degraded.
3872

3873
  The ldisk parameter, if True, will change the test from the
3874
  is_degraded attribute (which represents overall non-ok status for
3875
  the device(s)) to the ldisk (representing the local storage status).
3876

3877
  """
3878
  lu.cfg.SetDiskID(dev, node)
3879

    
3880
  result = True
3881

    
3882
  if on_primary or dev.AssembleOnSecondary():
3883
    rstats = lu.rpc.call_blockdev_find(node, dev)
3884
    msg = rstats.fail_msg
3885
    if msg:
3886
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3887
      result = False
3888
    elif not rstats.payload:
3889
      lu.LogWarning("Can't find disk on node %s", node)
3890
      result = False
3891
    else:
3892
      if ldisk:
3893
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3894
      else:
3895
        result = result and not rstats.payload.is_degraded
3896

    
3897
  if dev.children:
3898
    for child in dev.children:
3899
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3900

    
3901
  return result
3902

    
3903

    
3904
class LUOobCommand(NoHooksLU):
3905
  """Logical unit for OOB handling.
3906

3907
  """
3908
  REG_BGL = False
3909
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
3910

    
3911
  def ExpandNames(self):
3912
    """Gather locks we need.
3913

3914
    """
3915
    if self.op.node_names:
3916
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
3917
      lock_names = self.op.node_names
3918
    else:
3919
      lock_names = locking.ALL_SET
3920

    
3921
    self.needed_locks = {
3922
      locking.LEVEL_NODE: lock_names,
3923
      }
3924

    
3925
  def CheckPrereq(self):
3926
    """Check prerequisites.
3927

3928
    This checks:
3929
     - the node exists in the configuration
3930
     - OOB is supported
3931

3932
    Any errors are signaled by raising errors.OpPrereqError.
3933

3934
    """
3935
    self.nodes = []
3936
    self.master_node = self.cfg.GetMasterNode()
3937

    
3938
    assert self.op.power_delay >= 0.0
3939

    
3940
    if self.op.node_names:
3941
      if (self.op.command in self._SKIP_MASTER and
3942
          self.master_node in self.op.node_names):
3943
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
3944
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
3945

    
3946
        if master_oob_handler:
3947
          additional_text = ("run '%s %s %s' if you want to operate on the"
3948
                             " master regardless") % (master_oob_handler,
3949
                                                      self.op.command,
3950
                                                      self.master_node)
3951
        else:
3952
          additional_text = "it does not support out-of-band operations"
3953

    
3954
        raise errors.OpPrereqError(("Operating on the master node %s is not"
3955
                                    " allowed for %s; %s") %
3956
                                   (self.master_node, self.op.command,
3957
                                    additional_text), errors.ECODE_INVAL)
3958
    else:
3959
      self.op.node_names = self.cfg.GetNodeList()
3960
      if self.op.command in self._SKIP_MASTER:
3961
        self.op.node_names.remove(self.master_node)
3962

    
3963
    if self.op.command in self._SKIP_MASTER:
3964
      assert self.master_node not in self.op.node_names
3965

    
3966
    for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
3967
      if node is None:
3968
        raise errors.OpPrereqError("Node %s not found" % node_name,
3969
                                   errors.ECODE_NOENT)
3970
      else:
3971
        self.nodes.append(node)
3972

    
3973
      if (not self.op.ignore_status and
3974
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
3975
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
3976
                                    " not marked offline") % node_name,
3977
                                   errors.ECODE_STATE)
3978

    
3979
  def Exec(self, feedback_fn):
3980
    """Execute OOB and return result if we expect any.
3981

3982
    """
3983
    master_node = self.master_node
3984
    ret = []
3985

    
3986
    for idx, node in enumerate(utils.NiceSort(self.nodes,
3987
                                              key=lambda node: node.name)):
3988
      node_entry = [(constants.RS_NORMAL, node.name)]
3989
      ret.append(node_entry)
3990

    
3991
      oob_program = _SupportsOob(self.cfg, node)
3992

    
3993
      if not oob_program:
3994
        node_entry.append((constants.RS_UNAVAIL, None))
3995
        continue
3996

    
3997
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
3998
                   self.op.command, oob_program, node.name)
3999
      result = self.rpc.call_run_oob(master_node, oob_program,
4000
                                     self.op.command, node.name,
4001
                                     self.op.timeout)
4002

    
4003
      if result.fail_msg:
4004
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
4005
                        node.name, result.fail_msg)
4006
        node_entry.append((constants.RS_NODATA, None))
4007
      else:
4008
        try:
4009
          self._CheckPayload(result)
4010
        except errors.OpExecError, err:
4011
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
4012
                          node.name, err)
4013
          node_entry.append((constants.RS_NODATA, None))
4014
        else:
4015
          if self.op.command == constants.OOB_HEALTH:
4016
            # For health we should log important events
4017
            for item, status in result.payload:
4018
              if status in [constants.OOB_STATUS_WARNING,
4019
                            constants.OOB_STATUS_CRITICAL]:
4020
                self.LogWarning("Item '%s' on node '%s' has status '%s'",
4021
                                item, node.name, status)
4022

    
4023
          if self.op.command == constants.OOB_POWER_ON:
4024
            node.powered = True
4025
          elif self.op.command == constants.OOB_POWER_OFF:
4026
            node.powered = False
4027
          elif self.op.command == constants.OOB_POWER_STATUS:
4028
            powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
4029
            if powered != node.powered:
4030
              logging.warning(("Recorded power state (%s) of node '%s' does not"
4031
                               " match actual power state (%s)"), node.powered,
4032
                              node.name, powered)
4033

    
4034
          # For configuration changing commands we should update the node
4035
          if self.op.command in (constants.OOB_POWER_ON,
4036
                                 constants.OOB_POWER_OFF):
4037
            self.cfg.Update(node, feedback_fn)
4038

    
4039
          node_entry.append((constants.RS_NORMAL, result.payload))
4040

    
4041
          if (self.op.command == constants.OOB_POWER_ON and
4042
              idx < len(self.nodes) - 1):
4043
            time.sleep(self.op.power_delay)
4044

    
4045
    return ret
4046

    
4047
  def _CheckPayload(self, result):
4048
    """Checks if the payload is valid.
4049

4050
    @param result: RPC result
4051
    @raises errors.OpExecError: If payload is not valid
4052

4053
    """
4054
    errs = []
4055
    if self.op.command == constants.OOB_HEALTH:
4056
      if not isinstance(result.payload, list):
4057
        errs.append("command 'health' is expected to return a list but got %s" %
4058
                    type(result.payload))
4059
      else:
4060
        for item, status in result.payload:
4061
          if status not in constants.OOB_STATUSES:
4062
            errs.append("health item '%s' has invalid status '%s'" %
4063
                        (item, status))
4064

    
4065
    if self.op.command == constants.OOB_POWER_STATUS:
4066
      if not isinstance(result.payload, dict):
4067
        errs.append("power-status is expected to return a dict but got %s" %
4068
                    type(result.payload))
4069