Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ f8fa4175

History | View | Annotate | Download (442.9 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

    
44
from ganeti import ssh
45
from ganeti import utils
46
from ganeti import errors
47
from ganeti import hypervisor
48
from ganeti import locking
49
from ganeti import constants
50
from ganeti import objects
51
from ganeti import serializer
52
from ganeti import ssconf
53
from ganeti import uidpool
54
from ganeti import compat
55
from ganeti import masterd
56
from ganeti import netutils
57
from ganeti import query
58
from ganeti import qlang
59
from ganeti import opcodes
60
from ganeti import ht
61

    
62
import ganeti.masterd.instance # pylint: disable-msg=W0611
63

    
64

    
65
def _SupportsOob(cfg, node):
66
  """Tells if node supports OOB.
67

68
  @type cfg: L{config.ConfigWriter}
69
  @param cfg: The cluster configuration
70
  @type node: L{objects.Node}
71
  @param node: The node
72
  @return: The OOB script if supported or an empty string otherwise
73

74
  """
75
  return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
76

    
77

    
78
class ResultWithJobs:
79
  """Data container for LU results with jobs.
80

81
  Instances of this class returned from L{LogicalUnit.Exec} will be recognized
82
  by L{mcpu.Processor._ProcessResult}. The latter will then submit the jobs
83
  contained in the C{jobs} attribute and include the job IDs in the opcode
84
  result.
85

86
  """
87
  def __init__(self, jobs, **kwargs):
88
    """Initializes this class.
89

90
    Additional return values can be specified as keyword arguments.
91

92
    @type jobs: list of lists of L{opcode.OpCode}
93
    @param jobs: A list of lists of opcode objects
94

95
    """
96
    self.jobs = jobs
97
    self.other = kwargs
98

    
99

    
100
class LogicalUnit(object):
101
  """Logical Unit base class.
102

103
  Subclasses must follow these rules:
104
    - implement ExpandNames
105
    - implement CheckPrereq (except when tasklets are used)
106
    - implement Exec (except when tasklets are used)
107
    - implement BuildHooksEnv
108
    - implement BuildHooksNodes
109
    - redefine HPATH and HTYPE
110
    - optionally redefine their run requirements:
111
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
112

113
  Note that all commands require root permissions.
114

115
  @ivar dry_run_result: the value (if any) that will be returned to the caller
116
      in dry-run mode (signalled by opcode dry_run parameter)
117

118
  """
119
  HPATH = None
120
  HTYPE = None
121
  REQ_BGL = True
122

    
123
  def __init__(self, processor, op, context, rpc):
124
    """Constructor for LogicalUnit.
125

126
    This needs to be overridden in derived classes in order to check op
127
    validity.
128

129
    """
130
    self.proc = processor
131
    self.op = op
132
    self.cfg = context.cfg
133
    self.glm = context.glm
134
    self.context = context
135
    self.rpc = rpc
136
    # Dicts used to declare locking needs to mcpu
137
    self.needed_locks = None
138
    self.share_locks = dict.fromkeys(locking.LEVELS, 0)
139
    self.add_locks = {}
140
    self.remove_locks = {}
141
    # Used to force good behavior when calling helper functions
142
    self.recalculate_locks = {}
143
    # logging
144
    self.Log = processor.Log # pylint: disable-msg=C0103
145
    self.LogWarning = processor.LogWarning # pylint: disable-msg=C0103
146
    self.LogInfo = processor.LogInfo # pylint: disable-msg=C0103
147
    self.LogStep = processor.LogStep # pylint: disable-msg=C0103
148
    # support for dry-run
149
    self.dry_run_result = None
150
    # support for generic debug attribute
151
    if (not hasattr(self.op, "debug_level") or
152
        not isinstance(self.op.debug_level, int)):
153
      self.op.debug_level = 0
154

    
155
    # Tasklets
156
    self.tasklets = None
157

    
158
    # Validate opcode parameters and set defaults
159
    self.op.Validate(True)
160

    
161
    self.CheckArguments()
162

    
163
  def CheckArguments(self):
164
    """Check syntactic validity for the opcode arguments.
165

166
    This method is for doing a simple syntactic check and ensure
167
    validity of opcode parameters, without any cluster-related
168
    checks. While the same can be accomplished in ExpandNames and/or
169
    CheckPrereq, doing these separate is better because:
170

171
      - ExpandNames is left as as purely a lock-related function
172
      - CheckPrereq is run after we have acquired locks (and possible
173
        waited for them)
174

175
    The function is allowed to change the self.op attribute so that
176
    later methods can no longer worry about missing parameters.
177

178
    """
179
    pass
180

    
181
  def ExpandNames(self):
182
    """Expand names for this LU.
183

184
    This method is called before starting to execute the opcode, and it should
185
    update all the parameters of the opcode to their canonical form (e.g. a
186
    short node name must be fully expanded after this method has successfully
187
    completed). This way locking, hooks, logging, etc. can work correctly.
188

189
    LUs which implement this method must also populate the self.needed_locks
190
    member, as a dict with lock levels as keys, and a list of needed lock names
191
    as values. Rules:
192

193
      - use an empty dict if you don't need any lock
194
      - if you don't need any lock at a particular level omit that level
195
      - don't put anything for the BGL level
196
      - if you want all locks at a level use locking.ALL_SET as a value
197

198
    If you need to share locks (rather than acquire them exclusively) at one
199
    level you can modify self.share_locks, setting a true value (usually 1) for
200
    that level. By default locks are not shared.
201

202
    This function can also define a list of tasklets, which then will be
203
    executed in order instead of the usual LU-level CheckPrereq and Exec
204
    functions, if those are not defined by the LU.
205

206
    Examples::
207

208
      # Acquire all nodes and one instance
209
      self.needed_locks = {
210
        locking.LEVEL_NODE: locking.ALL_SET,
211
        locking.LEVEL_INSTANCE: ['instance1.example.com'],
212
      }
213
      # Acquire just two nodes
214
      self.needed_locks = {
215
        locking.LEVEL_NODE: ['node1.example.com', 'node2.example.com'],
216
      }
217
      # Acquire no locks
218
      self.needed_locks = {} # No, you can't leave it to the default value None
219

220
    """
221
    # The implementation of this method is mandatory only if the new LU is
222
    # concurrent, so that old LUs don't need to be changed all at the same
223
    # time.
224
    if self.REQ_BGL:
225
      self.needed_locks = {} # Exclusive LUs don't need locks.
226
    else:
227
      raise NotImplementedError
228

    
229
  def DeclareLocks(self, level):
230
    """Declare LU locking needs for a level
231

232
    While most LUs can just declare their locking needs at ExpandNames time,
233
    sometimes there's the need to calculate some locks after having acquired
234
    the ones before. This function is called just before acquiring locks at a
235
    particular level, but after acquiring the ones at lower levels, and permits
236
    such calculations. It can be used to modify self.needed_locks, and by
237
    default it does nothing.
238

239
    This function is only called if you have something already set in
240
    self.needed_locks for the level.
241

242
    @param level: Locking level which is going to be locked
243
    @type level: member of ganeti.locking.LEVELS
244

245
    """
246

    
247
  def CheckPrereq(self):
248
    """Check prerequisites for this LU.
249

250
    This method should check that the prerequisites for the execution
251
    of this LU are fulfilled. It can do internode communication, but
252
    it should be idempotent - no cluster or system changes are
253
    allowed.
254

255
    The method should raise errors.OpPrereqError in case something is
256
    not fulfilled. Its return value is ignored.
257

258
    This method should also update all the parameters of the opcode to
259
    their canonical form if it hasn't been done by ExpandNames before.
260

261
    """
262
    if self.tasklets is not None:
263
      for (idx, tl) in enumerate(self.tasklets):
264
        logging.debug("Checking prerequisites for tasklet %s/%s",
265
                      idx + 1, len(self.tasklets))
266
        tl.CheckPrereq()
267
    else:
268
      pass
269

    
270
  def Exec(self, feedback_fn):
271
    """Execute the LU.
272

273
    This method should implement the actual work. It should raise
274
    errors.OpExecError for failures that are somewhat dealt with in
275
    code, or expected.
276

277
    """
278
    if self.tasklets is not None:
279
      for (idx, tl) in enumerate(self.tasklets):
280
        logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
281
        tl.Exec(feedback_fn)
282
    else:
283
      raise NotImplementedError
284

    
285
  def BuildHooksEnv(self):
286
    """Build hooks environment for this LU.
287

288
    @rtype: dict
289
    @return: Dictionary containing the environment that will be used for
290
      running the hooks for this LU. The keys of the dict must not be prefixed
291
      with "GANETI_"--that'll be added by the hooks runner. The hooks runner
292
      will extend the environment with additional variables. If no environment
293
      should be defined, an empty dictionary should be returned (not C{None}).
294
    @note: If the C{HPATH} attribute of the LU class is C{None}, this function
295
      will not be called.
296

297
    """
298
    raise NotImplementedError
299

    
300
  def BuildHooksNodes(self):
301
    """Build list of nodes to run LU's hooks.
302

303
    @rtype: tuple; (list, list)
304
    @return: Tuple containing a list of node names on which the hook
305
      should run before the execution and a list of node names on which the
306
      hook should run after the execution. No nodes should be returned as an
307
      empty list (and not None).
308
    @note: If the C{HPATH} attribute of the LU class is C{None}, this function
309
      will not be called.
310

311
    """
312
    raise NotImplementedError
313

    
314
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
315
    """Notify the LU about the results of its hooks.
316

317
    This method is called every time a hooks phase is executed, and notifies
318
    the Logical Unit about the hooks' result. The LU can then use it to alter
319
    its result based on the hooks.  By default the method does nothing and the
320
    previous result is passed back unchanged but any LU can define it if it
321
    wants to use the local cluster hook-scripts somehow.
322

323
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
324
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
325
    @param hook_results: the results of the multi-node hooks rpc call
326
    @param feedback_fn: function used send feedback back to the caller
327
    @param lu_result: the previous Exec result this LU had, or None
328
        in the PRE phase
329
    @return: the new Exec result, based on the previous result
330
        and hook results
331

332
    """
333
    # API must be kept, thus we ignore the unused argument and could
334
    # be a function warnings
335
    # pylint: disable-msg=W0613,R0201
336
    return lu_result
337

    
338
  def _ExpandAndLockInstance(self):
339
    """Helper function to expand and lock an instance.
340

341
    Many LUs that work on an instance take its name in self.op.instance_name
342
    and need to expand it and then declare the expanded name for locking. This
343
    function does it, and then updates self.op.instance_name to the expanded
344
    name. It also initializes needed_locks as a dict, if this hasn't been done
345
    before.
346

347
    """
348
    if self.needed_locks is None:
349
      self.needed_locks = {}
350
    else:
351
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
352
        "_ExpandAndLockInstance called with instance-level locks set"
353
    self.op.instance_name = _ExpandInstanceName(self.cfg,
354
                                                self.op.instance_name)
355
    self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
356

    
357
  def _LockInstancesNodes(self, primary_only=False):
358
    """Helper function to declare instances' nodes for locking.
359

360
    This function should be called after locking one or more instances to lock
361
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
362
    with all primary or secondary nodes for instances already locked and
363
    present in self.needed_locks[locking.LEVEL_INSTANCE].
364

365
    It should be called from DeclareLocks, and for safety only works if
366
    self.recalculate_locks[locking.LEVEL_NODE] is set.
367

368
    In the future it may grow parameters to just lock some instance's nodes, or
369
    to just lock primaries or secondary nodes, if needed.
370

371
    If should be called in DeclareLocks in a way similar to::
372

373
      if level == locking.LEVEL_NODE:
374
        self._LockInstancesNodes()
375

376
    @type primary_only: boolean
377
    @param primary_only: only lock primary nodes of locked instances
378

379
    """
380
    assert locking.LEVEL_NODE in self.recalculate_locks, \
381
      "_LockInstancesNodes helper function called with no nodes to recalculate"
382

    
383
    # TODO: check if we're really been called with the instance locks held
384

    
385
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
386
    # future we might want to have different behaviors depending on the value
387
    # of self.recalculate_locks[locking.LEVEL_NODE]
388
    wanted_nodes = []
389
    for instance_name in self.glm.list_owned(locking.LEVEL_INSTANCE):
390
      instance = self.context.cfg.GetInstanceInfo(instance_name)
391
      wanted_nodes.append(instance.primary_node)
392
      if not primary_only:
393
        wanted_nodes.extend(instance.secondary_nodes)
394

    
395
    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
396
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
397
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
398
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
399

    
400
    del self.recalculate_locks[locking.LEVEL_NODE]
401

    
402

    
403
class NoHooksLU(LogicalUnit): # pylint: disable-msg=W0223
404
  """Simple LU which runs no hooks.
405

406
  This LU is intended as a parent for other LogicalUnits which will
407
  run no hooks, in order to reduce duplicate code.
408

409
  """
410
  HPATH = None
411
  HTYPE = None
412

    
413
  def BuildHooksEnv(self):
414
    """Empty BuildHooksEnv for NoHooksLu.
415

416
    This just raises an error.
417

418
    """
419
    raise AssertionError("BuildHooksEnv called for NoHooksLUs")
420

    
421
  def BuildHooksNodes(self):
422
    """Empty BuildHooksNodes for NoHooksLU.
423

424
    """
425
    raise AssertionError("BuildHooksNodes called for NoHooksLU")
426

    
427

    
428
class Tasklet:
429
  """Tasklet base class.
430

431
  Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
432
  they can mix legacy code with tasklets. Locking needs to be done in the LU,
433
  tasklets know nothing about locks.
434

435
  Subclasses must follow these rules:
436
    - Implement CheckPrereq
437
    - Implement Exec
438

439
  """
440
  def __init__(self, lu):
441
    self.lu = lu
442

    
443
    # Shortcuts
444
    self.cfg = lu.cfg
445
    self.rpc = lu.rpc
446

    
447
  def CheckPrereq(self):
448
    """Check prerequisites for this tasklets.
449

450
    This method should check whether the prerequisites for the execution of
451
    this tasklet are fulfilled. It can do internode communication, but it
452
    should be idempotent - no cluster or system changes are allowed.
453

454
    The method should raise errors.OpPrereqError in case something is not
455
    fulfilled. Its return value is ignored.
456

457
    This method should also update all parameters to their canonical form if it
458
    hasn't been done before.
459

460
    """
461
    pass
462

    
463
  def Exec(self, feedback_fn):
464
    """Execute the tasklet.
465

466
    This method should implement the actual work. It should raise
467
    errors.OpExecError for failures that are somewhat dealt with in code, or
468
    expected.
469

470
    """
471
    raise NotImplementedError
472

    
473

    
474
class _QueryBase:
475
  """Base for query utility classes.
476

477
  """
478
  #: Attribute holding field definitions
479
  FIELDS = None
480

    
481
  def __init__(self, filter_, fields, use_locking):
482
    """Initializes this class.
483

484
    """
485
    self.use_locking = use_locking
486

    
487
    self.query = query.Query(self.FIELDS, fields, filter_=filter_,
488
                             namefield="name")
489
    self.requested_data = self.query.RequestedData()
490
    self.names = self.query.RequestedNames()
491

    
492
    # Sort only if no names were requested
493
    self.sort_by_name = not self.names
494

    
495
    self.do_locking = None
496
    self.wanted = None
497

    
498
  def _GetNames(self, lu, all_names, lock_level):
499
    """Helper function to determine names asked for in the query.
500

501
    """
502
    if self.do_locking:
503
      names = lu.glm.list_owned(lock_level)
504
    else:
505
      names = all_names
506

    
507
    if self.wanted == locking.ALL_SET:
508
      assert not self.names
509
      # caller didn't specify names, so ordering is not important
510
      return utils.NiceSort(names)
511

    
512
    # caller specified names and we must keep the same order
513
    assert self.names
514
    assert not self.do_locking or lu.glm.is_owned(lock_level)
515

    
516
    missing = set(self.wanted).difference(names)
517
    if missing:
518
      raise errors.OpExecError("Some items were removed before retrieving"
519
                               " their data: %s" % missing)
520

    
521
    # Return expanded names
522
    return self.wanted
523

    
524
  def ExpandNames(self, lu):
525
    """Expand names for this query.
526

527
    See L{LogicalUnit.ExpandNames}.
528

529
    """
530
    raise NotImplementedError()
531

    
532
  def DeclareLocks(self, lu, level):
533
    """Declare locks for this query.
534

535
    See L{LogicalUnit.DeclareLocks}.
536

537
    """
538
    raise NotImplementedError()
539

    
540
  def _GetQueryData(self, lu):
541
    """Collects all data for this query.
542

543
    @return: Query data object
544

545
    """
546
    raise NotImplementedError()
547

    
548
  def NewStyleQuery(self, lu):
549
    """Collect data and execute query.
550

551
    """
552
    return query.GetQueryResponse(self.query, self._GetQueryData(lu),
553
                                  sort_by_name=self.sort_by_name)
554

    
555
  def OldStyleQuery(self, lu):
556
    """Collect data and execute query.
557

558
    """
559
    return self.query.OldStyleQuery(self._GetQueryData(lu),
560
                                    sort_by_name=self.sort_by_name)
561

    
562

    
563
def _GetWantedNodes(lu, nodes):
564
  """Returns list of checked and expanded node names.
565

566
  @type lu: L{LogicalUnit}
567
  @param lu: the logical unit on whose behalf we execute
568
  @type nodes: list
569
  @param nodes: list of node names or None for all nodes
570
  @rtype: list
571
  @return: the list of nodes, sorted
572
  @raise errors.ProgrammerError: if the nodes parameter is wrong type
573

574
  """
575
  if nodes:
576
    return [_ExpandNodeName(lu.cfg, name) for name in nodes]
577

    
578
  return utils.NiceSort(lu.cfg.GetNodeList())
579

    
580

    
581
def _GetWantedInstances(lu, instances):
582
  """Returns list of checked and expanded instance names.
583

584
  @type lu: L{LogicalUnit}
585
  @param lu: the logical unit on whose behalf we execute
586
  @type instances: list
587
  @param instances: list of instance names or None for all instances
588
  @rtype: list
589
  @return: the list of instances, sorted
590
  @raise errors.OpPrereqError: if the instances parameter is wrong type
591
  @raise errors.OpPrereqError: if any of the passed instances is not found
592

593
  """
594
  if instances:
595
    wanted = [_ExpandInstanceName(lu.cfg, name) for name in instances]
596
  else:
597
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
598
  return wanted
599

    
600

    
601
def _GetUpdatedParams(old_params, update_dict,
602
                      use_default=True, use_none=False):
603
  """Return the new version of a parameter dictionary.
604

605
  @type old_params: dict
606
  @param old_params: old parameters
607
  @type update_dict: dict
608
  @param update_dict: dict containing new parameter values, or
609
      constants.VALUE_DEFAULT to reset the parameter to its default
610
      value
611
  @param use_default: boolean
612
  @type use_default: whether to recognise L{constants.VALUE_DEFAULT}
613
      values as 'to be deleted' values
614
  @param use_none: boolean
615
  @type use_none: whether to recognise C{None} values as 'to be
616
      deleted' values
617
  @rtype: dict
618
  @return: the new parameter dictionary
619

620
  """
621
  params_copy = copy.deepcopy(old_params)
622
  for key, val in update_dict.iteritems():
623
    if ((use_default and val == constants.VALUE_DEFAULT) or
624
        (use_none and val is None)):
625
      try:
626
        del params_copy[key]
627
      except KeyError:
628
        pass
629
    else:
630
      params_copy[key] = val
631
  return params_copy
632

    
633

    
634
def _ReleaseLocks(lu, level, names=None, keep=None):
635
  """Releases locks owned by an LU.
636

637
  @type lu: L{LogicalUnit}
638
  @param level: Lock level
639
  @type names: list or None
640
  @param names: Names of locks to release
641
  @type keep: list or None
642
  @param keep: Names of locks to retain
643

644
  """
645
  assert not (keep is not None and names is not None), \
646
         "Only one of the 'names' and the 'keep' parameters can be given"
647

    
648
  if names is not None:
649
    should_release = names.__contains__
650
  elif keep:
651
    should_release = lambda name: name not in keep
652
  else:
653
    should_release = None
654

    
655
  if should_release:
656
    retain = []
657
    release = []
658

    
659
    # Determine which locks to release
660
    for name in lu.glm.list_owned(level):
661
      if should_release(name):
662
        release.append(name)
663
      else:
664
        retain.append(name)
665

    
666
    assert len(lu.glm.list_owned(level)) == (len(retain) + len(release))
667

    
668
    # Release just some locks
669
    lu.glm.release(level, names=release)
670

    
671
    assert frozenset(lu.glm.list_owned(level)) == frozenset(retain)
672
  else:
673
    # Release everything
674
    lu.glm.release(level)
675

    
676
    assert not lu.glm.is_owned(level), "No locks should be owned"
677

    
678

    
679
def _RunPostHook(lu, node_name):
680
  """Runs the post-hook for an opcode on a single node.
681

682
  """
683
  hm = lu.proc.hmclass(lu.rpc.call_hooks_runner, lu)
684
  try:
685
    hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
686
  except:
687
    # pylint: disable-msg=W0702
688
    lu.LogWarning("Errors occurred running hooks on %s" % node_name)
689

    
690

    
691
def _CheckOutputFields(static, dynamic, selected):
692
  """Checks whether all selected fields are valid.
693

694
  @type static: L{utils.FieldSet}
695
  @param static: static fields set
696
  @type dynamic: L{utils.FieldSet}
697
  @param dynamic: dynamic fields set
698

699
  """
700
  f = utils.FieldSet()
701
  f.Extend(static)
702
  f.Extend(dynamic)
703

    
704
  delta = f.NonMatching(selected)
705
  if delta:
706
    raise errors.OpPrereqError("Unknown output fields selected: %s"
707
                               % ",".join(delta), errors.ECODE_INVAL)
708

    
709

    
710
def _CheckGlobalHvParams(params):
711
  """Validates that given hypervisor params are not global ones.
712

713
  This will ensure that instances don't get customised versions of
714
  global params.
715

716
  """
717
  used_globals = constants.HVC_GLOBALS.intersection(params)
718
  if used_globals:
719
    msg = ("The following hypervisor parameters are global and cannot"
720
           " be customized at instance level, please modify them at"
721
           " cluster level: %s" % utils.CommaJoin(used_globals))
722
    raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
723

    
724

    
725
def _CheckNodeOnline(lu, node, msg=None):
726
  """Ensure that a given node is online.
727

728
  @param lu: the LU on behalf of which we make the check
729
  @param node: the node to check
730
  @param msg: if passed, should be a message to replace the default one
731
  @raise errors.OpPrereqError: if the node is offline
732

733
  """
734
  if msg is None:
735
    msg = "Can't use offline node"
736
  if lu.cfg.GetNodeInfo(node).offline:
737
    raise errors.OpPrereqError("%s: %s" % (msg, node), errors.ECODE_STATE)
738

    
739

    
740
def _CheckNodeNotDrained(lu, node):
741
  """Ensure that a given node is not drained.
742

743
  @param lu: the LU on behalf of which we make the check
744
  @param node: the node to check
745
  @raise errors.OpPrereqError: if the node is drained
746

747
  """
748
  if lu.cfg.GetNodeInfo(node).drained:
749
    raise errors.OpPrereqError("Can't use drained node %s" % node,
750
                               errors.ECODE_STATE)
751

    
752

    
753
def _CheckNodeVmCapable(lu, node):
754
  """Ensure that a given node is vm capable.
755

756
  @param lu: the LU on behalf of which we make the check
757
  @param node: the node to check
758
  @raise errors.OpPrereqError: if the node is not vm capable
759

760
  """
761
  if not lu.cfg.GetNodeInfo(node).vm_capable:
762
    raise errors.OpPrereqError("Can't use non-vm_capable node %s" % node,
763
                               errors.ECODE_STATE)
764

    
765

    
766
def _CheckNodeHasOS(lu, node, os_name, force_variant):
767
  """Ensure that a node supports a given OS.
768

769
  @param lu: the LU on behalf of which we make the check
770
  @param node: the node to check
771
  @param os_name: the OS to query about
772
  @param force_variant: whether to ignore variant errors
773
  @raise errors.OpPrereqError: if the node is not supporting the OS
774

775
  """
776
  result = lu.rpc.call_os_get(node, os_name)
777
  result.Raise("OS '%s' not in supported OS list for node %s" %
778
               (os_name, node),
779
               prereq=True, ecode=errors.ECODE_INVAL)
780
  if not force_variant:
781
    _CheckOSVariant(result.payload, os_name)
782

    
783

    
784
def _CheckNodeHasSecondaryIP(lu, node, secondary_ip, prereq):
785
  """Ensure that a node has the given secondary ip.
786

787
  @type lu: L{LogicalUnit}
788
  @param lu: the LU on behalf of which we make the check
789
  @type node: string
790
  @param node: the node to check
791
  @type secondary_ip: string
792
  @param secondary_ip: the ip to check
793
  @type prereq: boolean
794
  @param prereq: whether to throw a prerequisite or an execute error
795
  @raise errors.OpPrereqError: if the node doesn't have the ip, and prereq=True
796
  @raise errors.OpExecError: if the node doesn't have the ip, and prereq=False
797

798
  """
799
  result = lu.rpc.call_node_has_ip_address(node, secondary_ip)
800
  result.Raise("Failure checking secondary ip on node %s" % node,
801
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
802
  if not result.payload:
803
    msg = ("Node claims it doesn't have the secondary ip you gave (%s),"
804
           " please fix and re-run this command" % secondary_ip)
805
    if prereq:
806
      raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
807
    else:
808
      raise errors.OpExecError(msg)
809

    
810

    
811
def _GetClusterDomainSecret():
812
  """Reads the cluster domain secret.
813

814
  """
815
  return utils.ReadOneLineFile(constants.CLUSTER_DOMAIN_SECRET_FILE,
816
                               strict=True)
817

    
818

    
819
def _CheckInstanceDown(lu, instance, reason):
820
  """Ensure that an instance is not running."""
821
  if instance.admin_up:
822
    raise errors.OpPrereqError("Instance %s is marked to be up, %s" %
823
                               (instance.name, reason), errors.ECODE_STATE)
824

    
825
  pnode = instance.primary_node
826
  ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
827
  ins_l.Raise("Can't contact node %s for instance information" % pnode,
828
              prereq=True, ecode=errors.ECODE_ENVIRON)
829

    
830
  if instance.name in ins_l.payload:
831
    raise errors.OpPrereqError("Instance %s is running, %s" %
832
                               (instance.name, reason), errors.ECODE_STATE)
833

    
834

    
835
def _ExpandItemName(fn, name, kind):
836
  """Expand an item name.
837

838
  @param fn: the function to use for expansion
839
  @param name: requested item name
840
  @param kind: text description ('Node' or 'Instance')
841
  @return: the resolved (full) name
842
  @raise errors.OpPrereqError: if the item is not found
843

844
  """
845
  full_name = fn(name)
846
  if full_name is None:
847
    raise errors.OpPrereqError("%s '%s' not known" % (kind, name),
848
                               errors.ECODE_NOENT)
849
  return full_name
850

    
851

    
852
def _ExpandNodeName(cfg, name):
853
  """Wrapper over L{_ExpandItemName} for nodes."""
854
  return _ExpandItemName(cfg.ExpandNodeName, name, "Node")
855

    
856

    
857
def _ExpandInstanceName(cfg, name):
858
  """Wrapper over L{_ExpandItemName} for instance."""
859
  return _ExpandItemName(cfg.ExpandInstanceName, name, "Instance")
860

    
861

    
862
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
863
                          memory, vcpus, nics, disk_template, disks,
864
                          bep, hvp, hypervisor_name):
865
  """Builds instance related env variables for hooks
866

867
  This builds the hook environment from individual variables.
868

869
  @type name: string
870
  @param name: the name of the instance
871
  @type primary_node: string
872
  @param primary_node: the name of the instance's primary node
873
  @type secondary_nodes: list
874
  @param secondary_nodes: list of secondary nodes as strings
875
  @type os_type: string
876
  @param os_type: the name of the instance's OS
877
  @type status: boolean
878
  @param status: the should_run status of the instance
879
  @type memory: string
880
  @param memory: the memory size of the instance
881
  @type vcpus: string
882
  @param vcpus: the count of VCPUs the instance has
883
  @type nics: list
884
  @param nics: list of tuples (ip, mac, mode, link) representing
885
      the NICs the instance has
886
  @type disk_template: string
887
  @param disk_template: the disk template of the instance
888
  @type disks: list
889
  @param disks: the list of (size, mode) pairs
890
  @type bep: dict
891
  @param bep: the backend parameters for the instance
892
  @type hvp: dict
893
  @param hvp: the hypervisor parameters for the instance
894
  @type hypervisor_name: string
895
  @param hypervisor_name: the hypervisor for the instance
896
  @rtype: dict
897
  @return: the hook environment for this instance
898

899
  """
900
  if status:
901
    str_status = "up"
902
  else:
903
    str_status = "down"
904
  env = {
905
    "OP_TARGET": name,
906
    "INSTANCE_NAME": name,
907
    "INSTANCE_PRIMARY": primary_node,
908
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
909
    "INSTANCE_OS_TYPE": os_type,
910
    "INSTANCE_STATUS": str_status,
911
    "INSTANCE_MEMORY": memory,
912
    "INSTANCE_VCPUS": vcpus,
913
    "INSTANCE_DISK_TEMPLATE": disk_template,
914
    "INSTANCE_HYPERVISOR": hypervisor_name,
915
  }
916

    
917
  if nics:
918
    nic_count = len(nics)
919
    for idx, (ip, mac, mode, link) in enumerate(nics):
920
      if ip is None:
921
        ip = ""
922
      env["INSTANCE_NIC%d_IP" % idx] = ip
923
      env["INSTANCE_NIC%d_MAC" % idx] = mac
924
      env["INSTANCE_NIC%d_MODE" % idx] = mode
925
      env["INSTANCE_NIC%d_LINK" % idx] = link
926
      if mode == constants.NIC_MODE_BRIDGED:
927
        env["INSTANCE_NIC%d_BRIDGE" % idx] = link
928
  else:
929
    nic_count = 0
930

    
931
  env["INSTANCE_NIC_COUNT"] = nic_count
932

    
933
  if disks:
934
    disk_count = len(disks)
935
    for idx, (size, mode) in enumerate(disks):
936
      env["INSTANCE_DISK%d_SIZE" % idx] = size
937
      env["INSTANCE_DISK%d_MODE" % idx] = mode
938
  else:
939
    disk_count = 0
940

    
941
  env["INSTANCE_DISK_COUNT"] = disk_count
942

    
943
  for source, kind in [(bep, "BE"), (hvp, "HV")]:
944
    for key, value in source.items():
945
      env["INSTANCE_%s_%s" % (kind, key)] = value
946

    
947
  return env
948

    
949

    
950
def _NICListToTuple(lu, nics):
951
  """Build a list of nic information tuples.
952

953
  This list is suitable to be passed to _BuildInstanceHookEnv or as a return
954
  value in LUInstanceQueryData.
955

956
  @type lu:  L{LogicalUnit}
957
  @param lu: the logical unit on whose behalf we execute
958
  @type nics: list of L{objects.NIC}
959
  @param nics: list of nics to convert to hooks tuples
960

961
  """
962
  hooks_nics = []
963
  cluster = lu.cfg.GetClusterInfo()
964
  for nic in nics:
965
    ip = nic.ip
966
    mac = nic.mac
967
    filled_params = cluster.SimpleFillNIC(nic.nicparams)
968
    mode = filled_params[constants.NIC_MODE]
969
    link = filled_params[constants.NIC_LINK]
970
    hooks_nics.append((ip, mac, mode, link))
971
  return hooks_nics
972

    
973

    
974
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
975
  """Builds instance related env variables for hooks from an object.
976

977
  @type lu: L{LogicalUnit}
978
  @param lu: the logical unit on whose behalf we execute
979
  @type instance: L{objects.Instance}
980
  @param instance: the instance for which we should build the
981
      environment
982
  @type override: dict
983
  @param override: dictionary with key/values that will override
984
      our values
985
  @rtype: dict
986
  @return: the hook environment dictionary
987

988
  """
989
  cluster = lu.cfg.GetClusterInfo()
990
  bep = cluster.FillBE(instance)
991
  hvp = cluster.FillHV(instance)
992
  args = {
993
    'name': instance.name,
994
    'primary_node': instance.primary_node,
995
    'secondary_nodes': instance.secondary_nodes,
996
    'os_type': instance.os,
997
    'status': instance.admin_up,
998
    'memory': bep[constants.BE_MEMORY],
999
    'vcpus': bep[constants.BE_VCPUS],
1000
    'nics': _NICListToTuple(lu, instance.nics),
1001
    'disk_template': instance.disk_template,
1002
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
1003
    'bep': bep,
1004
    'hvp': hvp,
1005
    'hypervisor_name': instance.hypervisor,
1006
  }
1007
  if override:
1008
    args.update(override)
1009
  return _BuildInstanceHookEnv(**args) # pylint: disable-msg=W0142
1010

    
1011

    
1012
def _AdjustCandidatePool(lu, exceptions):
1013
  """Adjust the candidate pool after node operations.
1014

1015
  """
1016
  mod_list = lu.cfg.MaintainCandidatePool(exceptions)
1017
  if mod_list:
1018
    lu.LogInfo("Promoted nodes to master candidate role: %s",
1019
               utils.CommaJoin(node.name for node in mod_list))
1020
    for name in mod_list:
1021
      lu.context.ReaddNode(name)
1022
  mc_now, mc_max, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1023
  if mc_now > mc_max:
1024
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
1025
               (mc_now, mc_max))
1026

    
1027

    
1028
def _DecideSelfPromotion(lu, exceptions=None):
1029
  """Decide whether I should promote myself as a master candidate.
1030

1031
  """
1032
  cp_size = lu.cfg.GetClusterInfo().candidate_pool_size
1033
  mc_now, mc_should, _ = lu.cfg.GetMasterCandidateStats(exceptions)
1034
  # the new node will increase mc_max with one, so:
1035
  mc_should = min(mc_should + 1, cp_size)
1036
  return mc_now < mc_should
1037

    
1038

    
1039
def _CheckNicsBridgesExist(lu, target_nics, target_node):
1040
  """Check that the brigdes needed by a list of nics exist.
1041

1042
  """
1043
  cluster = lu.cfg.GetClusterInfo()
1044
  paramslist = [cluster.SimpleFillNIC(nic.nicparams) for nic in target_nics]
1045
  brlist = [params[constants.NIC_LINK] for params in paramslist
1046
            if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
1047
  if brlist:
1048
    result = lu.rpc.call_bridges_exist(target_node, brlist)
1049
    result.Raise("Error checking bridges on destination node '%s'" %
1050
                 target_node, prereq=True, ecode=errors.ECODE_ENVIRON)
1051

    
1052

    
1053
def _CheckInstanceBridgesExist(lu, instance, node=None):
1054
  """Check that the brigdes needed by an instance exist.
1055

1056
  """
1057
  if node is None:
1058
    node = instance.primary_node
1059
  _CheckNicsBridgesExist(lu, instance.nics, node)
1060

    
1061

    
1062
def _CheckOSVariant(os_obj, name):
1063
  """Check whether an OS name conforms to the os variants specification.
1064

1065
  @type os_obj: L{objects.OS}
1066
  @param os_obj: OS object to check
1067
  @type name: string
1068
  @param name: OS name passed by the user, to check for validity
1069

1070
  """
1071
  if not os_obj.supported_variants:
1072
    return
1073
  variant = objects.OS.GetVariant(name)
1074
  if not variant:
1075
    raise errors.OpPrereqError("OS name must include a variant",
1076
                               errors.ECODE_INVAL)
1077

    
1078
  if variant not in os_obj.supported_variants:
1079
    raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)
1080

    
1081

    
1082
def _GetNodeInstancesInner(cfg, fn):
1083
  return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]
1084

    
1085

    
1086
def _GetNodeInstances(cfg, node_name):
1087
  """Returns a list of all primary and secondary instances on a node.
1088

1089
  """
1090

    
1091
  return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)
1092

    
1093

    
1094
def _GetNodePrimaryInstances(cfg, node_name):
1095
  """Returns primary instances on a node.
1096

1097
  """
1098
  return _GetNodeInstancesInner(cfg,
1099
                                lambda inst: node_name == inst.primary_node)
1100

    
1101

    
1102
def _GetNodeSecondaryInstances(cfg, node_name):
1103
  """Returns secondary instances on a node.
1104

1105
  """
1106
  return _GetNodeInstancesInner(cfg,
1107
                                lambda inst: node_name in inst.secondary_nodes)
1108

    
1109

    
1110
def _GetStorageTypeArgs(cfg, storage_type):
1111
  """Returns the arguments for a storage type.
1112

1113
  """
1114
  # Special case for file storage
1115
  if storage_type == constants.ST_FILE:
1116
    # storage.FileStorage wants a list of storage directories
1117
    return [[cfg.GetFileStorageDir(), cfg.GetSharedFileStorageDir()]]
1118

    
1119
  return []
1120

    
1121

    
1122
def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
1123
  faulty = []
1124

    
1125
  for dev in instance.disks:
1126
    cfg.SetDiskID(dev, node_name)
1127

    
1128
  result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
1129
  result.Raise("Failed to get disk status from node %s" % node_name,
1130
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
1131

    
1132
  for idx, bdev_status in enumerate(result.payload):
1133
    if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
1134
      faulty.append(idx)
1135

    
1136
  return faulty
1137

    
1138

    
1139
def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
1140
  """Check the sanity of iallocator and node arguments and use the
1141
  cluster-wide iallocator if appropriate.
1142

1143
  Check that at most one of (iallocator, node) is specified. If none is
1144
  specified, then the LU's opcode's iallocator slot is filled with the
1145
  cluster-wide default iallocator.
1146

1147
  @type iallocator_slot: string
1148
  @param iallocator_slot: the name of the opcode iallocator slot
1149
  @type node_slot: string
1150
  @param node_slot: the name of the opcode target node slot
1151

1152
  """
1153
  node = getattr(lu.op, node_slot, None)
1154
  iallocator = getattr(lu.op, iallocator_slot, None)
1155

    
1156
  if node is not None and iallocator is not None:
1157
    raise errors.OpPrereqError("Do not specify both, iallocator and node",
1158
                               errors.ECODE_INVAL)
1159
  elif node is None and iallocator is None:
1160
    default_iallocator = lu.cfg.GetDefaultIAllocator()
1161
    if default_iallocator:
1162
      setattr(lu.op, iallocator_slot, default_iallocator)
1163
    else:
1164
      raise errors.OpPrereqError("No iallocator or node given and no"
1165
                                 " cluster-wide default iallocator found;"
1166
                                 " please specify either an iallocator or a"
1167
                                 " node, or set a cluster-wide default"
1168
                                 " iallocator")
1169

    
1170

    
1171
class LUClusterPostInit(LogicalUnit):
1172
  """Logical unit for running hooks after cluster initialization.
1173

1174
  """
1175
  HPATH = "cluster-init"
1176
  HTYPE = constants.HTYPE_CLUSTER
1177

    
1178
  def BuildHooksEnv(self):
1179
    """Build hooks env.
1180

1181
    """
1182
    return {
1183
      "OP_TARGET": self.cfg.GetClusterName(),
1184
      }
1185

    
1186
  def BuildHooksNodes(self):
1187
    """Build hooks nodes.
1188

1189
    """
1190
    return ([], [self.cfg.GetMasterNode()])
1191

    
1192
  def Exec(self, feedback_fn):
1193
    """Nothing to do.
1194

1195
    """
1196
    return True
1197

    
1198

    
1199
class LUClusterDestroy(LogicalUnit):
1200
  """Logical unit for destroying the cluster.
1201

1202
  """
1203
  HPATH = "cluster-destroy"
1204
  HTYPE = constants.HTYPE_CLUSTER
1205

    
1206
  def BuildHooksEnv(self):
1207
    """Build hooks env.
1208

1209
    """
1210
    return {
1211
      "OP_TARGET": self.cfg.GetClusterName(),
1212
      }
1213

    
1214
  def BuildHooksNodes(self):
1215
    """Build hooks nodes.
1216

1217
    """
1218
    return ([], [])
1219

    
1220
  def CheckPrereq(self):
1221
    """Check prerequisites.
1222

1223
    This checks whether the cluster is empty.
1224

1225
    Any errors are signaled by raising errors.OpPrereqError.
1226

1227
    """
1228
    master = self.cfg.GetMasterNode()
1229

    
1230
    nodelist = self.cfg.GetNodeList()
1231
    if len(nodelist) != 1 or nodelist[0] != master:
1232
      raise errors.OpPrereqError("There are still %d node(s) in"
1233
                                 " this cluster." % (len(nodelist) - 1),
1234
                                 errors.ECODE_INVAL)
1235
    instancelist = self.cfg.GetInstanceList()
1236
    if instancelist:
1237
      raise errors.OpPrereqError("There are still %d instance(s) in"
1238
                                 " this cluster." % len(instancelist),
1239
                                 errors.ECODE_INVAL)
1240

    
1241
  def Exec(self, feedback_fn):
1242
    """Destroys the cluster.
1243

1244
    """
1245
    master = self.cfg.GetMasterNode()
1246

    
1247
    # Run post hooks on master node before it's removed
1248
    _RunPostHook(self, master)
1249

    
1250
    result = self.rpc.call_node_stop_master(master, False)
1251
    result.Raise("Could not disable the master role")
1252

    
1253
    return master
1254

    
1255

    
1256
def _VerifyCertificate(filename):
1257
  """Verifies a certificate for LUClusterVerifyConfig.
1258

1259
  @type filename: string
1260
  @param filename: Path to PEM file
1261

1262
  """
1263
  try:
1264
    cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1265
                                           utils.ReadFile(filename))
1266
  except Exception, err: # pylint: disable-msg=W0703
1267
    return (LUClusterVerifyConfig.ETYPE_ERROR,
1268
            "Failed to load X509 certificate %s: %s" % (filename, err))
1269

    
1270
  (errcode, msg) = \
1271
    utils.VerifyX509Certificate(cert, constants.SSL_CERT_EXPIRATION_WARN,
1272
                                constants.SSL_CERT_EXPIRATION_ERROR)
1273

    
1274
  if msg:
1275
    fnamemsg = "While verifying %s: %s" % (filename, msg)
1276
  else:
1277
    fnamemsg = None
1278

    
1279
  if errcode is None:
1280
    return (None, fnamemsg)
1281
  elif errcode == utils.CERT_WARNING:
1282
    return (LUClusterVerifyConfig.ETYPE_WARNING, fnamemsg)
1283
  elif errcode == utils.CERT_ERROR:
1284
    return (LUClusterVerifyConfig.ETYPE_ERROR, fnamemsg)
1285

    
1286
  raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
1287

    
1288

    
1289
def _GetAllHypervisorParameters(cluster, instances):
1290
  """Compute the set of all hypervisor parameters.
1291

1292
  @type cluster: L{objects.Cluster}
1293
  @param cluster: the cluster object
1294
  @param instances: list of L{objects.Instance}
1295
  @param instances: additional instances from which to obtain parameters
1296
  @rtype: list of (origin, hypervisor, parameters)
1297
  @return: a list with all parameters found, indicating the hypervisor they
1298
       apply to, and the origin (can be "cluster", "os X", or "instance Y")
1299

1300
  """
1301
  hvp_data = []
1302

    
1303
  for hv_name in cluster.enabled_hypervisors:
1304
    hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
1305

    
1306
  for os_name, os_hvp in cluster.os_hvp.items():
1307
    for hv_name, hv_params in os_hvp.items():
1308
      if hv_params:
1309
        full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
1310
        hvp_data.append(("os %s" % os_name, hv_name, full_params))
1311

    
1312
  # TODO: collapse identical parameter values in a single one
1313
  for instance in instances:
1314
    if instance.hvparams:
1315
      hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
1316
                       cluster.FillHV(instance)))
1317

    
1318
  return hvp_data
1319

    
1320

    
1321
class _VerifyErrors(object):
1322
  """Mix-in for cluster/group verify LUs.
1323

1324
  It provides _Error and _ErrorIf, and updates the self.bad boolean. (Expects
1325
  self.op and self._feedback_fn to be available.)
1326

1327
  """
1328
  TCLUSTER = "cluster"
1329
  TNODE = "node"
1330
  TINSTANCE = "instance"
1331

    
1332
  ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
1333
  ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
1334
  ECLUSTERFILECHECK = (TCLUSTER, "ECLUSTERFILECHECK")
1335
  ECLUSTERDANGLINGNODES = (TNODE, "ECLUSTERDANGLINGNODES")
1336
  ECLUSTERDANGLINGINST = (TNODE, "ECLUSTERDANGLINGINST")
1337
  EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
1338
  EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
1339
  EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
1340
  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
1341
  EINSTANCEFAULTYDISK = (TINSTANCE, "EINSTANCEFAULTYDISK")
1342
  EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
1343
  EINSTANCESPLITGROUPS = (TINSTANCE, "EINSTANCESPLITGROUPS")
1344
  ENODEDRBD = (TNODE, "ENODEDRBD")
1345
  ENODEDRBDHELPER = (TNODE, "ENODEDRBDHELPER")
1346
  ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
1347
  ENODEHOOKS = (TNODE, "ENODEHOOKS")
1348
  ENODEHV = (TNODE, "ENODEHV")
1349
  ENODELVM = (TNODE, "ENODELVM")
1350
  ENODEN1 = (TNODE, "ENODEN1")
1351
  ENODENET = (TNODE, "ENODENET")
1352
  ENODEOS = (TNODE, "ENODEOS")
1353
  ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
1354
  ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
1355
  ENODERPC = (TNODE, "ENODERPC")
1356
  ENODESSH = (TNODE, "ENODESSH")
1357
  ENODEVERSION = (TNODE, "ENODEVERSION")
1358
  ENODESETUP = (TNODE, "ENODESETUP")
1359
  ENODETIME = (TNODE, "ENODETIME")
1360
  ENODEOOBPATH = (TNODE, "ENODEOOBPATH")
1361

    
1362
  ETYPE_FIELD = "code"
1363
  ETYPE_ERROR = "ERROR"
1364
  ETYPE_WARNING = "WARNING"
1365

    
1366
  def _Error(self, ecode, item, msg, *args, **kwargs):
1367
    """Format an error message.
1368

1369
    Based on the opcode's error_codes parameter, either format a
1370
    parseable error code, or a simpler error string.
1371

1372
    This must be called only from Exec and functions called from Exec.
1373

1374
    """
1375
    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
1376
    itype, etxt = ecode
1377
    # first complete the msg
1378
    if args:
1379
      msg = msg % args
1380
    # then format the whole message
1381
    if self.op.error_codes: # This is a mix-in. pylint: disable-msg=E1101
1382
      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
1383
    else:
1384
      if item:
1385
        item = " " + item
1386
      else:
1387
        item = ""
1388
      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
1389
    # and finally report it via the feedback_fn
1390
    self._feedback_fn("  - %s" % msg) # Mix-in. pylint: disable-msg=E1101
1391

    
1392
  def _ErrorIf(self, cond, *args, **kwargs):
1393
    """Log an error message if the passed condition is True.
1394

1395
    """
1396
    cond = (bool(cond)
1397
            or self.op.debug_simulate_errors) # pylint: disable-msg=E1101
1398
    if cond:
1399
      self._Error(*args, **kwargs)
1400
    # do not mark the operation as failed for WARN cases only
1401
    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
1402
      self.bad = self.bad or cond
1403

    
1404

    
1405
class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
1406
  """Verifies the cluster config.
1407

1408
  """
1409
  REQ_BGL = False
1410

    
1411
  def _VerifyHVP(self, hvp_data):
1412
    """Verifies locally the syntax of the hypervisor parameters.
1413

1414
    """
1415
    for item, hv_name, hv_params in hvp_data:
1416
      msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
1417
             (item, hv_name))
1418
      try:
1419
        hv_class = hypervisor.GetHypervisor(hv_name)
1420
        utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1421
        hv_class.CheckParameterSyntax(hv_params)
1422
      except errors.GenericError, err:
1423
        self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
1424

    
1425
  def ExpandNames(self):
1426
    self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
1427
    self.all_node_info = self.cfg.GetAllNodesInfo()
1428
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1429
    self.needed_locks = {}
1430

    
1431
  def Exec(self, feedback_fn):
1432
    """Verify integrity of cluster, performing various test on nodes.
1433

1434
    """
1435
    self.bad = False
1436
    self._feedback_fn = feedback_fn
1437

    
1438
    feedback_fn("* Verifying cluster config")
1439

    
1440
    for msg in self.cfg.VerifyConfig():
1441
      self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
1442

    
1443
    feedback_fn("* Verifying cluster certificate files")
1444

    
1445
    for cert_filename in constants.ALL_CERT_FILES:
1446
      (errcode, msg) = _VerifyCertificate(cert_filename)
1447
      self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
1448

    
1449
    feedback_fn("* Verifying hypervisor parameters")
1450

    
1451
    self._VerifyHVP(_GetAllHypervisorParameters(self.cfg.GetClusterInfo(),
1452
                                                self.all_inst_info.values()))
1453

    
1454
    feedback_fn("* Verifying all nodes belong to an existing group")
1455

    
1456
    # We do this verification here because, should this bogus circumstance
1457
    # occur, it would never be catched by VerifyGroup, which only acts on
1458
    # nodes/instances reachable from existing node groups.
1459

    
1460
    dangling_nodes = set(node.name for node in self.all_node_info.values()
1461
                         if node.group not in self.all_group_info)
1462

    
1463
    dangling_instances = {}
1464
    no_node_instances = []
1465

    
1466
    for inst in self.all_inst_info.values():
1467
      if inst.primary_node in dangling_nodes:
1468
        dangling_instances.setdefault(inst.primary_node, []).append(inst.name)
1469
      elif inst.primary_node not in self.all_node_info:
1470
        no_node_instances.append(inst.name)
1471

    
1472
    pretty_dangling = [
1473
        "%s (%s)" %
1474
        (node.name,
1475
         utils.CommaJoin(dangling_instances.get(node.name,
1476
                                                ["no instances"])))
1477
        for node in dangling_nodes]
1478

    
1479
    self._ErrorIf(bool(dangling_nodes), self.ECLUSTERDANGLINGNODES, None,
1480
                  "the following nodes (and their instances) belong to a non"
1481
                  " existing group: %s", utils.CommaJoin(pretty_dangling))
1482

    
1483
    self._ErrorIf(bool(no_node_instances), self.ECLUSTERDANGLINGINST, None,
1484
                  "the following instances have a non-existing primary-node:"
1485
                  " %s", utils.CommaJoin(no_node_instances))
1486

    
1487
    return (not self.bad, [g.name for g in self.all_group_info.values()])
1488

    
1489

    
1490
class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
1491
  """Verifies the status of a node group.
1492

1493
  """
1494
  HPATH = "cluster-verify"
1495
  HTYPE = constants.HTYPE_CLUSTER
1496
  REQ_BGL = False
1497

    
1498
  _HOOKS_INDENT_RE = re.compile("^", re.M)
1499

    
1500
  class NodeImage(object):
1501
    """A class representing the logical and physical status of a node.
1502

1503
    @type name: string
1504
    @ivar name: the node name to which this object refers
1505
    @ivar volumes: a structure as returned from
1506
        L{ganeti.backend.GetVolumeList} (runtime)
1507
    @ivar instances: a list of running instances (runtime)
1508
    @ivar pinst: list of configured primary instances (config)
1509
    @ivar sinst: list of configured secondary instances (config)
1510
    @ivar sbp: dictionary of {primary-node: list of instances} for all
1511
        instances for which this node is secondary (config)
1512
    @ivar mfree: free memory, as reported by hypervisor (runtime)
1513
    @ivar dfree: free disk, as reported by the node (runtime)
1514
    @ivar offline: the offline status (config)
1515
    @type rpc_fail: boolean
1516
    @ivar rpc_fail: whether the RPC verify call was successfull (overall,
1517
        not whether the individual keys were correct) (runtime)
1518
    @type lvm_fail: boolean
1519
    @ivar lvm_fail: whether the RPC call didn't return valid LVM data
1520
    @type hyp_fail: boolean
1521
    @ivar hyp_fail: whether the RPC call didn't return the instance list
1522
    @type ghost: boolean
1523
    @ivar ghost: whether this is a known node or not (config)
1524
    @type os_fail: boolean
1525
    @ivar os_fail: whether the RPC call didn't return valid OS data
1526
    @type oslist: list
1527
    @ivar oslist: list of OSes as diagnosed by DiagnoseOS
1528
    @type vm_capable: boolean
1529
    @ivar vm_capable: whether the node can host instances
1530

1531
    """
1532
    def __init__(self, offline=False, name=None, vm_capable=True):
1533
      self.name = name
1534
      self.volumes = {}
1535
      self.instances = []
1536
      self.pinst = []
1537
      self.sinst = []
1538
      self.sbp = {}
1539
      self.mfree = 0
1540
      self.dfree = 0
1541
      self.offline = offline
1542
      self.vm_capable = vm_capable
1543
      self.rpc_fail = False
1544
      self.lvm_fail = False
1545
      self.hyp_fail = False
1546
      self.ghost = False
1547
      self.os_fail = False
1548
      self.oslist = {}
1549

    
1550
  def ExpandNames(self):
1551
    # This raises errors.OpPrereqError on its own:
1552
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
1553

    
1554
    all_node_info = self.cfg.GetAllNodesInfo()
1555
    all_inst_info = self.cfg.GetAllInstancesInfo()
1556

    
1557
    node_names = set(node.name
1558
                     for node in all_node_info.values()
1559
                     if node.group == self.group_uuid)
1560

    
1561
    inst_names = [inst.name
1562
                  for inst in all_inst_info.values()
1563
                  if inst.primary_node in node_names]
1564

    
1565
    # In Exec(), we warn about mirrored instances that have primary and
1566
    # secondary living in separate node groups. To fully verify that
1567
    # volumes for these instances are healthy, we will need to do an
1568
    # extra call to their secondaries. We ensure here those nodes will
1569
    # be locked.
1570
    for inst in inst_names:
1571
      if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
1572
        node_names.update(all_inst_info[inst].secondary_nodes)
1573

    
1574
    self.needed_locks = {
1575
      locking.LEVEL_NODEGROUP: [self.group_uuid],
1576
      locking.LEVEL_NODE: list(node_names),
1577
      locking.LEVEL_INSTANCE: inst_names,
1578
    }
1579

    
1580
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
1581

    
1582
  def CheckPrereq(self):
1583
    self.all_node_info = self.cfg.GetAllNodesInfo()
1584
    self.all_inst_info = self.cfg.GetAllInstancesInfo()
1585

    
1586
    group_nodes = set(node.name
1587
                      for node in self.all_node_info.values()
1588
                      if node.group == self.group_uuid)
1589

    
1590
    group_instances = set(inst.name
1591
                          for inst in self.all_inst_info.values()
1592
                          if inst.primary_node in group_nodes)
1593

    
1594
    unlocked_nodes = \
1595
        group_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
1596

    
1597
    unlocked_instances = \
1598
        group_instances.difference(self.glm.list_owned(locking.LEVEL_INSTANCE))
1599

    
1600
    if unlocked_nodes:
1601
      raise errors.OpPrereqError("missing lock for nodes: %s" %
1602
                                 utils.CommaJoin(unlocked_nodes))
1603

    
1604
    if unlocked_instances:
1605
      raise errors.OpPrereqError("missing lock for instances: %s" %
1606
                                 utils.CommaJoin(unlocked_instances))
1607

    
1608
    self.my_node_names = utils.NiceSort(group_nodes)
1609
    self.my_inst_names = utils.NiceSort(group_instances)
1610

    
1611
    self.my_node_info = dict((name, self.all_node_info[name])
1612
                             for name in self.my_node_names)
1613

    
1614
    self.my_inst_info = dict((name, self.all_inst_info[name])
1615
                             for name in self.my_inst_names)
1616

    
1617
    # We detect here the nodes that will need the extra RPC calls for verifying
1618
    # split LV volumes; they should be locked.
1619
    extra_lv_nodes = set()
1620

    
1621
    for inst in self.my_inst_info.values():
1622
      if inst.disk_template in constants.DTS_INT_MIRROR:
1623
        group = self.my_node_info[inst.primary_node].group
1624
        for nname in inst.secondary_nodes:
1625
          if self.all_node_info[nname].group != group:
1626
            extra_lv_nodes.add(nname)
1627

    
1628
    unlocked_lv_nodes = \
1629
        extra_lv_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
1630

    
1631
    if unlocked_lv_nodes:
1632
      raise errors.OpPrereqError("these nodes could be locked: %s" %
1633
                                 utils.CommaJoin(unlocked_lv_nodes))
1634
    self.extra_lv_nodes = list(extra_lv_nodes)
1635

    
1636
  def _VerifyNode(self, ninfo, nresult):
1637
    """Perform some basic validation on data returned from a node.
1638

1639
      - check the result data structure is well formed and has all the
1640
        mandatory fields
1641
      - check ganeti version
1642

1643
    @type ninfo: L{objects.Node}
1644
    @param ninfo: the node to check
1645
    @param nresult: the results from the node
1646
    @rtype: boolean
1647
    @return: whether overall this call was successful (and we can expect
1648
         reasonable values in the respose)
1649

1650
    """
1651
    node = ninfo.name
1652
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1653

    
1654
    # main result, nresult should be a non-empty dict
1655
    test = not nresult or not isinstance(nresult, dict)
1656
    _ErrorIf(test, self.ENODERPC, node,
1657
                  "unable to verify node: no data returned")
1658
    if test:
1659
      return False
1660

    
1661
    # compares ganeti version
1662
    local_version = constants.PROTOCOL_VERSION
1663
    remote_version = nresult.get("version", None)
1664
    test = not (remote_version and
1665
                isinstance(remote_version, (list, tuple)) and
1666
                len(remote_version) == 2)
1667
    _ErrorIf(test, self.ENODERPC, node,
1668
             "connection to node returned invalid data")
1669
    if test:
1670
      return False
1671

    
1672
    test = local_version != remote_version[0]
1673
    _ErrorIf(test, self.ENODEVERSION, node,
1674
             "incompatible protocol versions: master %s,"
1675
             " node %s", local_version, remote_version[0])
1676
    if test:
1677
      return False
1678

    
1679
    # node seems compatible, we can actually try to look into its results
1680

    
1681
    # full package version
1682
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1683
                  self.ENODEVERSION, node,
1684
                  "software version mismatch: master %s, node %s",
1685
                  constants.RELEASE_VERSION, remote_version[1],
1686
                  code=self.ETYPE_WARNING)
1687

    
1688
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1689
    if ninfo.vm_capable and isinstance(hyp_result, dict):
1690
      for hv_name, hv_result in hyp_result.iteritems():
1691
        test = hv_result is not None
1692
        _ErrorIf(test, self.ENODEHV, node,
1693
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1694

    
1695
    hvp_result = nresult.get(constants.NV_HVPARAMS, None)
1696
    if ninfo.vm_capable and isinstance(hvp_result, list):
1697
      for item, hv_name, hv_result in hvp_result:
1698
        _ErrorIf(True, self.ENODEHV, node,
1699
                 "hypervisor %s parameter verify failure (source %s): %s",
1700
                 hv_name, item, hv_result)
1701

    
1702
    test = nresult.get(constants.NV_NODESETUP,
1703
                       ["Missing NODESETUP results"])
1704
    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
1705
             "; ".join(test))
1706

    
1707
    return True
1708

    
1709
  def _VerifyNodeTime(self, ninfo, nresult,
1710
                      nvinfo_starttime, nvinfo_endtime):
1711
    """Check the node time.
1712

1713
    @type ninfo: L{objects.Node}
1714
    @param ninfo: the node to check
1715
    @param nresult: the remote results for the node
1716
    @param nvinfo_starttime: the start time of the RPC call
1717
    @param nvinfo_endtime: the end time of the RPC call
1718

1719
    """
1720
    node = ninfo.name
1721
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1722

    
1723
    ntime = nresult.get(constants.NV_TIME, None)
1724
    try:
1725
      ntime_merged = utils.MergeTime(ntime)
1726
    except (ValueError, TypeError):
1727
      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
1728
      return
1729

    
1730
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1731
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1732
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1733
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1734
    else:
1735
      ntime_diff = None
1736

    
1737
    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
1738
             "Node time diverges by at least %s from master node time",
1739
             ntime_diff)
1740

    
1741
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1742
    """Check the node LVM results.
1743

1744
    @type ninfo: L{objects.Node}
1745
    @param ninfo: the node to check
1746
    @param nresult: the remote results for the node
1747
    @param vg_name: the configured VG name
1748

1749
    """
1750
    if vg_name is None:
1751
      return
1752

    
1753
    node = ninfo.name
1754
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1755

    
1756
    # checks vg existence and size > 20G
1757
    vglist = nresult.get(constants.NV_VGLIST, None)
1758
    test = not vglist
1759
    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
1760
    if not test:
1761
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
1762
                                            constants.MIN_VG_SIZE)
1763
      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
1764

    
1765
    # check pv names
1766
    pvlist = nresult.get(constants.NV_PVLIST, None)
1767
    test = pvlist is None
1768
    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
1769
    if not test:
1770
      # check that ':' is not present in PV names, since it's a
1771
      # special character for lvcreate (denotes the range of PEs to
1772
      # use on the PV)
1773
      for _, pvname, owner_vg in pvlist:
1774
        test = ":" in pvname
1775
        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
1776
                 " '%s' of VG '%s'", pvname, owner_vg)
1777

    
1778
  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
1779
    """Check the node bridges.
1780

1781
    @type ninfo: L{objects.Node}
1782
    @param ninfo: the node to check
1783
    @param nresult: the remote results for the node
1784
    @param bridges: the expected list of bridges
1785

1786
    """
1787
    if not bridges:
1788
      return
1789

    
1790
    node = ninfo.name
1791
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1792

    
1793
    missing = nresult.get(constants.NV_BRIDGES, None)
1794
    test = not isinstance(missing, list)
1795
    _ErrorIf(test, self.ENODENET, node,
1796
             "did not return valid bridge information")
1797
    if not test:
1798
      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
1799
               utils.CommaJoin(sorted(missing)))
1800

    
1801
  def _VerifyNodeNetwork(self, ninfo, nresult):
1802
    """Check the node network connectivity results.
1803

1804
    @type ninfo: L{objects.Node}
1805
    @param ninfo: the node to check
1806
    @param nresult: the remote results for the node
1807

1808
    """
1809
    node = ninfo.name
1810
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1811

    
1812
    test = constants.NV_NODELIST not in nresult
1813
    _ErrorIf(test, self.ENODESSH, node,
1814
             "node hasn't returned node ssh connectivity data")
1815
    if not test:
1816
      if nresult[constants.NV_NODELIST]:
1817
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1818
          _ErrorIf(True, self.ENODESSH, node,
1819
                   "ssh communication with node '%s': %s", a_node, a_msg)
1820

    
1821
    test = constants.NV_NODENETTEST not in nresult
1822
    _ErrorIf(test, self.ENODENET, node,
1823
             "node hasn't returned node tcp connectivity data")
1824
    if not test:
1825
      if nresult[constants.NV_NODENETTEST]:
1826
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1827
        for anode in nlist:
1828
          _ErrorIf(True, self.ENODENET, node,
1829
                   "tcp communication with node '%s': %s",
1830
                   anode, nresult[constants.NV_NODENETTEST][anode])
1831

    
1832
    test = constants.NV_MASTERIP not in nresult
1833
    _ErrorIf(test, self.ENODENET, node,
1834
             "node hasn't returned node master IP reachability data")
1835
    if not test:
1836
      if not nresult[constants.NV_MASTERIP]:
1837
        if node == self.master_node:
1838
          msg = "the master node cannot reach the master IP (not configured?)"
1839
        else:
1840
          msg = "cannot reach the master IP"
1841
        _ErrorIf(True, self.ENODENET, node, msg)
1842

    
1843
  def _VerifyInstance(self, instance, instanceconfig, node_image,
1844
                      diskstatus):
1845
    """Verify an instance.
1846

1847
    This function checks to see if the required block devices are
1848
    available on the instance's node.
1849

1850
    """
1851
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1852
    node_current = instanceconfig.primary_node
1853

    
1854
    node_vol_should = {}
1855
    instanceconfig.MapLVsByNode(node_vol_should)
1856

    
1857
    for node in node_vol_should:
1858
      n_img = node_image[node]
1859
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1860
        # ignore missing volumes on offline or broken nodes
1861
        continue
1862
      for volume in node_vol_should[node]:
1863
        test = volume not in n_img.volumes
1864
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
1865
                 "volume %s missing on node %s", volume, node)
1866

    
1867
    if instanceconfig.admin_up:
1868
      pri_img = node_image[node_current]
1869
      test = instance not in pri_img.instances and not pri_img.offline
1870
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
1871
               "instance not running on its primary node %s",
1872
               node_current)
1873

    
1874
    diskdata = [(nname, success, status, idx)
1875
                for (nname, disks) in diskstatus.items()
1876
                for idx, (success, status) in enumerate(disks)]
1877

    
1878
    for nname, success, bdev_status, idx in diskdata:
1879
      # the 'ghost node' construction in Exec() ensures that we have a
1880
      # node here
1881
      snode = node_image[nname]
1882
      bad_snode = snode.ghost or snode.offline
1883
      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
1884
               self.EINSTANCEFAULTYDISK, instance,
1885
               "couldn't retrieve status for disk/%s on %s: %s",
1886
               idx, nname, bdev_status)
1887
      _ErrorIf((instanceconfig.admin_up and success and
1888
                bdev_status.ldisk_status == constants.LDS_FAULTY),
1889
               self.EINSTANCEFAULTYDISK, instance,
1890
               "disk/%s on %s is faulty", idx, nname)
1891

    
1892
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
1893
    """Verify if there are any unknown volumes in the cluster.
1894

1895
    The .os, .swap and backup volumes are ignored. All other volumes are
1896
    reported as unknown.
1897

1898
    @type reserved: L{ganeti.utils.FieldSet}
1899
    @param reserved: a FieldSet of reserved volume names
1900

1901
    """
1902
    for node, n_img in node_image.items():
1903
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1904
        # skip non-healthy nodes
1905
        continue
1906
      for volume in n_img.volumes:
1907
        test = ((node not in node_vol_should or
1908
                volume not in node_vol_should[node]) and
1909
                not reserved.Matches(volume))
1910
        self._ErrorIf(test, self.ENODEORPHANLV, node,
1911
                      "volume %s is unknown", volume)
1912

    
1913
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
1914
    """Verify N+1 Memory Resilience.
1915

1916
    Check that if one single node dies we can still start all the
1917
    instances it was primary for.
1918

1919
    """
1920
    cluster_info = self.cfg.GetClusterInfo()
1921
    for node, n_img in node_image.items():
1922
      # This code checks that every node which is now listed as
1923
      # secondary has enough memory to host all instances it is
1924
      # supposed to should a single other node in the cluster fail.
1925
      # FIXME: not ready for failover to an arbitrary node
1926
      # FIXME: does not support file-backed instances
1927
      # WARNING: we currently take into account down instances as well
1928
      # as up ones, considering that even if they're down someone
1929
      # might want to start them even in the event of a node failure.
1930
      if n_img.offline:
1931
        # we're skipping offline nodes from the N+1 warning, since
1932
        # most likely we don't have good memory infromation from them;
1933
        # we already list instances living on such nodes, and that's
1934
        # enough warning
1935
        continue
1936
      for prinode, instances in n_img.sbp.items():
1937
        needed_mem = 0
1938
        for instance in instances:
1939
          bep = cluster_info.FillBE(instance_cfg[instance])
1940
          if bep[constants.BE_AUTO_BALANCE]:
1941
            needed_mem += bep[constants.BE_MEMORY]
1942
        test = n_img.mfree < needed_mem
1943
        self._ErrorIf(test, self.ENODEN1, node,
1944
                      "not enough memory to accomodate instance failovers"
1945
                      " should node %s fail (%dMiB needed, %dMiB available)",
1946
                      prinode, needed_mem, n_img.mfree)
1947

    
1948
  @classmethod
1949
  def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
1950
                   (files_all, files_all_opt, files_mc, files_vm)):
1951
    """Verifies file checksums collected from all nodes.
1952

1953
    @param errorif: Callback for reporting errors
1954
    @param nodeinfo: List of L{objects.Node} objects
1955
    @param master_node: Name of master node
1956
    @param all_nvinfo: RPC results
1957

1958
    """
1959
    node_names = frozenset(node.name for node in nodeinfo)
1960

    
1961
    assert master_node in node_names
1962
    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
1963
            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
1964
           "Found file listed in more than one file list"
1965

    
1966
    # Define functions determining which nodes to consider for a file
1967
    file2nodefn = dict([(filename, fn)
1968
      for (files, fn) in [(files_all, None),
1969
                          (files_all_opt, None),
1970
                          (files_mc, lambda node: (node.master_candidate or
1971
                                                   node.name == master_node)),
1972
                          (files_vm, lambda node: node.vm_capable)]
1973
      for filename in files])
1974

    
1975
    fileinfo = dict((filename, {}) for filename in file2nodefn.keys())
1976

    
1977
    for node in nodeinfo:
1978
      nresult = all_nvinfo[node.name]
1979

    
1980
      if nresult.fail_msg or not nresult.payload:
1981
        node_files = None
1982
      else:
1983
        node_files = nresult.payload.get(constants.NV_FILELIST, None)
1984

    
1985
      test = not (node_files and isinstance(node_files, dict))
1986
      errorif(test, cls.ENODEFILECHECK, node.name,
1987
              "Node did not return file checksum data")
1988
      if test:
1989
        continue
1990

    
1991
      for (filename, checksum) in node_files.items():
1992
        # Check if the file should be considered for a node
1993
        fn = file2nodefn[filename]
1994
        if fn is None or fn(node):
1995
          fileinfo[filename].setdefault(checksum, set()).add(node.name)
1996

    
1997
    for (filename, checksums) in fileinfo.items():
1998
      assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
1999

    
2000
      # Nodes having the file
2001
      with_file = frozenset(node_name
2002
                            for nodes in fileinfo[filename].values()
2003
                            for node_name in nodes)
2004

    
2005
      # Nodes missing file
2006
      missing_file = node_names - with_file
2007

    
2008
      if filename in files_all_opt:
2009
        # All or no nodes
2010
        errorif(missing_file and missing_file != node_names,
2011
                cls.ECLUSTERFILECHECK, None,
2012
                "File %s is optional, but it must exist on all or no nodes (not"
2013
                " found on %s)",
2014
                filename, utils.CommaJoin(utils.NiceSort(missing_file)))
2015
      else:
2016
        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
2017
                "File %s is missing from node(s) %s", filename,
2018
                utils.CommaJoin(utils.NiceSort(missing_file)))
2019

    
2020
      # See if there are multiple versions of the file
2021
      test = len(checksums) > 1
2022
      if test:
2023
        variants = ["variant %s on %s" %
2024
                    (idx + 1, utils.CommaJoin(utils.NiceSort(nodes)))
2025
                    for (idx, (checksum, nodes)) in
2026
                      enumerate(sorted(checksums.items()))]
2027
      else:
2028
        variants = []
2029

    
2030
      errorif(test, cls.ECLUSTERFILECHECK, None,
2031
              "File %s found with %s different checksums (%s)",
2032
              filename, len(checksums), "; ".join(variants))
2033

    
2034
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
2035
                      drbd_map):
2036
    """Verifies and the node DRBD status.
2037

2038
    @type ninfo: L{objects.Node}
2039
    @param ninfo: the node to check
2040
    @param nresult: the remote results for the node
2041
    @param instanceinfo: the dict of instances
2042
    @param drbd_helper: the configured DRBD usermode helper
2043
    @param drbd_map: the DRBD map as returned by
2044
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
2045

2046
    """
2047
    node = ninfo.name
2048
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2049

    
2050
    if drbd_helper:
2051
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
2052
      test = (helper_result == None)
2053
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
2054
               "no drbd usermode helper returned")
2055
      if helper_result:
2056
        status, payload = helper_result
2057
        test = not status
2058
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2059
                 "drbd usermode helper check unsuccessful: %s", payload)
2060
        test = status and (payload != drbd_helper)
2061
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
2062
                 "wrong drbd usermode helper: %s", payload)
2063

    
2064
    # compute the DRBD minors
2065
    node_drbd = {}
2066
    for minor, instance in drbd_map[node].items():
2067
      test = instance not in instanceinfo
2068
      _ErrorIf(test, self.ECLUSTERCFG, None,
2069
               "ghost instance '%s' in temporary DRBD map", instance)
2070
        # ghost instance should not be running, but otherwise we
2071
        # don't give double warnings (both ghost instance and
2072
        # unallocated minor in use)
2073
      if test:
2074
        node_drbd[minor] = (instance, False)
2075
      else:
2076
        instance = instanceinfo[instance]
2077
        node_drbd[minor] = (instance.name, instance.admin_up)
2078

    
2079
    # and now check them
2080
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
2081
    test = not isinstance(used_minors, (tuple, list))
2082
    _ErrorIf(test, self.ENODEDRBD, node,
2083
             "cannot parse drbd status file: %s", str(used_minors))
2084
    if test:
2085
      # we cannot check drbd status
2086
      return
2087

    
2088
    for minor, (iname, must_exist) in node_drbd.items():
2089
      test = minor not in used_minors and must_exist
2090
      _ErrorIf(test, self.ENODEDRBD, node,
2091
               "drbd minor %d of instance %s is not active", minor, iname)
2092
    for minor in used_minors:
2093
      test = minor not in node_drbd
2094
      _ErrorIf(test, self.ENODEDRBD, node,
2095
               "unallocated drbd minor %d is in use", minor)
2096

    
2097
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
2098
    """Builds the node OS structures.
2099

2100
    @type ninfo: L{objects.Node}
2101
    @param ninfo: the node to check
2102
    @param nresult: the remote results for the node
2103
    @param nimg: the node image object
2104

2105
    """
2106
    node = ninfo.name
2107
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2108

    
2109
    remote_os = nresult.get(constants.NV_OSLIST, None)
2110
    test = (not isinstance(remote_os, list) or
2111
            not compat.all(isinstance(v, list) and len(v) == 7
2112
                           for v in remote_os))
2113

    
2114
    _ErrorIf(test, self.ENODEOS, node,
2115
             "node hasn't returned valid OS data")
2116

    
2117
    nimg.os_fail = test
2118

    
2119
    if test:
2120
      return
2121

    
2122
    os_dict = {}
2123

    
2124
    for (name, os_path, status, diagnose,
2125
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
2126

    
2127
      if name not in os_dict:
2128
        os_dict[name] = []
2129

    
2130
      # parameters is a list of lists instead of list of tuples due to
2131
      # JSON lacking a real tuple type, fix it:
2132
      parameters = [tuple(v) for v in parameters]
2133
      os_dict[name].append((os_path, status, diagnose,
2134
                            set(variants), set(parameters), set(api_ver)))
2135

    
2136
    nimg.oslist = os_dict
2137

    
2138
  def _VerifyNodeOS(self, ninfo, nimg, base):
2139
    """Verifies the node OS list.
2140

2141
    @type ninfo: L{objects.Node}
2142
    @param ninfo: the node to check
2143
    @param nimg: the node image object
2144
    @param base: the 'template' node we match against (e.g. from the master)
2145

2146
    """
2147
    node = ninfo.name
2148
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2149

    
2150
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
2151

    
2152
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
2153
    for os_name, os_data in nimg.oslist.items():
2154
      assert os_data, "Empty OS status for OS %s?!" % os_name
2155
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
2156
      _ErrorIf(not f_status, self.ENODEOS, node,
2157
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
2158
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
2159
               "OS '%s' has multiple entries (first one shadows the rest): %s",
2160
               os_name, utils.CommaJoin([v[0] for v in os_data]))
2161
      # this will catched in backend too
2162
      _ErrorIf(compat.any(v >= constants.OS_API_V15 for v in f_api)
2163
               and not f_var, self.ENODEOS, node,
2164
               "OS %s with API at least %d does not declare any variant",
2165
               os_name, constants.OS_API_V15)
2166
      # comparisons with the 'base' image
2167
      test = os_name not in base.oslist
2168
      _ErrorIf(test, self.ENODEOS, node,
2169
               "Extra OS %s not present on reference node (%s)",
2170
               os_name, base.name)
2171
      if test:
2172
        continue
2173
      assert base.oslist[os_name], "Base node has empty OS status?"
2174
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
2175
      if not b_status:
2176
        # base OS is invalid, skipping
2177
        continue
2178
      for kind, a, b in [("API version", f_api, b_api),
2179
                         ("variants list", f_var, b_var),
2180
                         ("parameters", beautify_params(f_param),
2181
                          beautify_params(b_param))]:
2182
        _ErrorIf(a != b, self.ENODEOS, node,
2183
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
2184
                 kind, os_name, base.name,
2185
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
2186

    
2187
    # check any missing OSes
2188
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
2189
    _ErrorIf(missing, self.ENODEOS, node,
2190
             "OSes present on reference node %s but missing on this node: %s",
2191
             base.name, utils.CommaJoin(missing))
2192

    
2193
  def _VerifyOob(self, ninfo, nresult):
2194
    """Verifies out of band functionality of a node.
2195

2196
    @type ninfo: L{objects.Node}
2197
    @param ninfo: the node to check
2198
    @param nresult: the remote results for the node
2199

2200
    """
2201
    node = ninfo.name
2202
    # We just have to verify the paths on master and/or master candidates
2203
    # as the oob helper is invoked on the master
2204
    if ((ninfo.master_candidate or ninfo.master_capable) and
2205
        constants.NV_OOB_PATHS in nresult):
2206
      for path_result in nresult[constants.NV_OOB_PATHS]:
2207
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
2208

    
2209
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
2210
    """Verifies and updates the node volume data.
2211

2212
    This function will update a L{NodeImage}'s internal structures
2213
    with data from the remote call.
2214

2215
    @type ninfo: L{objects.Node}
2216
    @param ninfo: the node to check
2217
    @param nresult: the remote results for the node
2218
    @param nimg: the node image object
2219
    @param vg_name: the configured VG name
2220

2221
    """
2222
    node = ninfo.name
2223
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2224

    
2225
    nimg.lvm_fail = True
2226
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
2227
    if vg_name is None:
2228
      pass
2229
    elif isinstance(lvdata, basestring):
2230
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
2231
               utils.SafeEncode(lvdata))
2232
    elif not isinstance(lvdata, dict):
2233
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
2234
    else:
2235
      nimg.volumes = lvdata
2236
      nimg.lvm_fail = False
2237

    
2238
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
2239
    """Verifies and updates the node instance list.
2240

2241
    If the listing was successful, then updates this node's instance
2242
    list. Otherwise, it marks the RPC call as failed for the instance
2243
    list key.
2244

2245
    @type ninfo: L{objects.Node}
2246
    @param ninfo: the node to check
2247
    @param nresult: the remote results for the node
2248
    @param nimg: the node image object
2249

2250
    """
2251
    idata = nresult.get(constants.NV_INSTANCELIST, None)
2252
    test = not isinstance(idata, list)
2253
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
2254
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
2255
    if test:
2256
      nimg.hyp_fail = True
2257
    else:
2258
      nimg.instances = idata
2259

    
2260
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
2261
    """Verifies and computes a node information map
2262

2263
    @type ninfo: L{objects.Node}
2264
    @param ninfo: the node to check
2265
    @param nresult: the remote results for the node
2266
    @param nimg: the node image object
2267
    @param vg_name: the configured VG name
2268

2269
    """
2270
    node = ninfo.name
2271
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2272

    
2273
    # try to read free memory (from the hypervisor)
2274
    hv_info = nresult.get(constants.NV_HVINFO, None)
2275
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
2276
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
2277
    if not test:
2278
      try:
2279
        nimg.mfree = int(hv_info["memory_free"])
2280
      except (ValueError, TypeError):
2281
        _ErrorIf(True, self.ENODERPC, node,
2282
                 "node returned invalid nodeinfo, check hypervisor")
2283

    
2284
    # FIXME: devise a free space model for file based instances as well
2285
    if vg_name is not None:
2286
      test = (constants.NV_VGLIST not in nresult or
2287
              vg_name not in nresult[constants.NV_VGLIST])
2288
      _ErrorIf(test, self.ENODELVM, node,
2289
               "node didn't return data for the volume group '%s'"
2290
               " - it is either missing or broken", vg_name)
2291
      if not test:
2292
        try:
2293
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
2294
        except (ValueError, TypeError):
2295
          _ErrorIf(True, self.ENODERPC, node,
2296
                   "node returned invalid LVM info, check LVM status")
2297

    
2298
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
2299
    """Gets per-disk status information for all instances.
2300

2301
    @type nodelist: list of strings
2302
    @param nodelist: Node names
2303
    @type node_image: dict of (name, L{objects.Node})
2304
    @param node_image: Node objects
2305
    @type instanceinfo: dict of (name, L{objects.Instance})
2306
    @param instanceinfo: Instance objects
2307
    @rtype: {instance: {node: [(succes, payload)]}}
2308
    @return: a dictionary of per-instance dictionaries with nodes as
2309
        keys and disk information as values; the disk information is a
2310
        list of tuples (success, payload)
2311

2312
    """
2313
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2314

    
2315
    node_disks = {}
2316
    node_disks_devonly = {}
2317
    diskless_instances = set()
2318
    diskless = constants.DT_DISKLESS
2319

    
2320
    for nname in nodelist:
2321
      node_instances = list(itertools.chain(node_image[nname].pinst,
2322
                                            node_image[nname].sinst))
2323
      diskless_instances.update(inst for inst in node_instances
2324
                                if instanceinfo[inst].disk_template == diskless)
2325
      disks = [(inst, disk)
2326
               for inst in node_instances
2327
               for disk in instanceinfo[inst].disks]
2328

    
2329
      if not disks:
2330
        # No need to collect data
2331
        continue
2332

    
2333
      node_disks[nname] = disks
2334

    
2335
      # Creating copies as SetDiskID below will modify the objects and that can
2336
      # lead to incorrect data returned from nodes
2337
      devonly = [dev.Copy() for (_, dev) in disks]
2338

    
2339
      for dev in devonly:
2340
        self.cfg.SetDiskID(dev, nname)
2341

    
2342
      node_disks_devonly[nname] = devonly
2343

    
2344
    assert len(node_disks) == len(node_disks_devonly)
2345

    
2346
    # Collect data from all nodes with disks
2347
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2348
                                                          node_disks_devonly)
2349

    
2350
    assert len(result) == len(node_disks)
2351

    
2352
    instdisk = {}
2353

    
2354
    for (nname, nres) in result.items():
2355
      disks = node_disks[nname]
2356

    
2357
      if nres.offline:
2358
        # No data from this node
2359
        data = len(disks) * [(False, "node offline")]
2360
      else:
2361
        msg = nres.fail_msg
2362
        _ErrorIf(msg, self.ENODERPC, nname,
2363
                 "while getting disk information: %s", msg)
2364
        if msg:
2365
          # No data from this node
2366
          data = len(disks) * [(False, msg)]
2367
        else:
2368
          data = []
2369
          for idx, i in enumerate(nres.payload):
2370
            if isinstance(i, (tuple, list)) and len(i) == 2:
2371
              data.append(i)
2372
            else:
2373
              logging.warning("Invalid result from node %s, entry %d: %s",
2374
                              nname, idx, i)
2375
              data.append((False, "Invalid result from the remote node"))
2376

    
2377
      for ((inst, _), status) in zip(disks, data):
2378
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2379

    
2380
    # Add empty entries for diskless instances.
2381
    for inst in diskless_instances:
2382
      assert inst not in instdisk
2383
      instdisk[inst] = {}
2384

    
2385
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2386
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2387
                      compat.all(isinstance(s, (tuple, list)) and
2388
                                 len(s) == 2 for s in statuses)
2389
                      for inst, nnames in instdisk.items()
2390
                      for nname, statuses in nnames.items())
2391
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2392

    
2393
    return instdisk
2394

    
2395
  def BuildHooksEnv(self):
2396
    """Build hooks env.
2397

2398
    Cluster-Verify hooks just ran in the post phase and their failure makes
2399
    the output be logged in the verify output and the verification to fail.
2400

2401
    """
2402
    env = {
2403
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2404
      }
2405

    
2406
    env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
2407
               for node in self.my_node_info.values())
2408

    
2409
    return env
2410

    
2411
  def BuildHooksNodes(self):
2412
    """Build hooks nodes.
2413

2414
    """
2415
    assert self.my_node_names, ("Node list not gathered,"
2416
      " has CheckPrereq been executed?")
2417
    return ([], self.my_node_names)
2418

    
2419
  def Exec(self, feedback_fn):
2420
    """Verify integrity of the node group, performing various test on nodes.
2421

2422
    """
2423
    # This method has too many local variables. pylint: disable-msg=R0914
2424
    self.bad = False
2425
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2426
    verbose = self.op.verbose
2427
    self._feedback_fn = feedback_fn
2428

    
2429
    vg_name = self.cfg.GetVGName()
2430
    drbd_helper = self.cfg.GetDRBDHelper()
2431
    cluster = self.cfg.GetClusterInfo()
2432
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2433
    hypervisors = cluster.enabled_hypervisors
2434
    node_data_list = [self.my_node_info[name] for name in self.my_node_names]
2435

    
2436
    i_non_redundant = [] # Non redundant instances
2437
    i_non_a_balanced = [] # Non auto-balanced instances
2438
    n_offline = 0 # Count of offline nodes
2439
    n_drained = 0 # Count of nodes being drained
2440
    node_vol_should = {}
2441

    
2442
    # FIXME: verify OS list
2443

    
2444
    # File verification
2445
    filemap = _ComputeAncillaryFiles(cluster, False)
2446

    
2447
    # do local checksums
2448
    master_node = self.master_node = self.cfg.GetMasterNode()
2449
    master_ip = self.cfg.GetMasterIP()
2450

    
2451
    feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
2452

    
2453
    # We will make nodes contact all nodes in their group, and one node from
2454
    # every other group.
2455
    # TODO: should it be a *random* node, different every time?
2456
    online_nodes = [node.name for node in node_data_list if not node.offline]
2457
    other_group_nodes = {}
2458

    
2459
    for name in sorted(self.all_node_info):
2460
      node = self.all_node_info[name]
2461
      if (node.group not in other_group_nodes
2462
          and node.group != self.group_uuid
2463
          and not node.offline):
2464
        other_group_nodes[node.group] = node.name
2465

    
2466
    node_verify_param = {
2467
      constants.NV_FILELIST:
2468
        utils.UniqueSequence(filename
2469
                             for files in filemap
2470
                             for filename in files),
2471
      constants.NV_NODELIST: online_nodes + other_group_nodes.values(),
2472
      constants.NV_HYPERVISOR: hypervisors,
2473
      constants.NV_HVPARAMS:
2474
        _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
2475
      constants.NV_NODENETTEST: [(node.name, node.primary_ip, node.secondary_ip)
2476
                                 for node in node_data_list
2477
                                 if not node.offline],
2478
      constants.NV_INSTANCELIST: hypervisors,
2479
      constants.NV_VERSION: None,
2480
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2481
      constants.NV_NODESETUP: None,
2482
      constants.NV_TIME: None,
2483
      constants.NV_MASTERIP: (master_node, master_ip),
2484
      constants.NV_OSLIST: None,
2485
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2486
      }
2487

    
2488
    if vg_name is not None:
2489
      node_verify_param[constants.NV_VGLIST] = None
2490
      node_verify_param[constants.NV_LVLIST] = vg_name
2491
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2492
      node_verify_param[constants.NV_DRBDLIST] = None
2493

    
2494
    if drbd_helper:
2495
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2496

    
2497
    # bridge checks
2498
    # FIXME: this needs to be changed per node-group, not cluster-wide
2499
    bridges = set()
2500
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2501
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2502
      bridges.add(default_nicpp[constants.NIC_LINK])
2503
    for instance in self.my_inst_info.values():
2504
      for nic in instance.nics:
2505
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2506
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2507
          bridges.add(full_nic[constants.NIC_LINK])
2508

    
2509
    if bridges:
2510
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2511

    
2512
    # Build our expected cluster state
2513
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2514
                                                 name=node.name,
2515
                                                 vm_capable=node.vm_capable))
2516
                      for node in node_data_list)
2517

    
2518
    # Gather OOB paths
2519
    oob_paths = []
2520
    for node in self.all_node_info.values():
2521
      path = _SupportsOob(self.cfg, node)
2522
      if path and path not in oob_paths:
2523
        oob_paths.append(path)
2524

    
2525
    if oob_paths:
2526
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2527

    
2528
    for instance in self.my_inst_names:
2529
      inst_config = self.my_inst_info[instance]
2530

    
2531
      for nname in inst_config.all_nodes:
2532
        if nname not in node_image:
2533
          gnode = self.NodeImage(name=nname)
2534
          gnode.ghost = (nname not in self.all_node_info)
2535
          node_image[nname] = gnode
2536

    
2537
      inst_config.MapLVsByNode(node_vol_should)
2538

    
2539
      pnode = inst_config.primary_node
2540
      node_image[pnode].pinst.append(instance)
2541

    
2542
      for snode in inst_config.secondary_nodes:
2543
        nimg = node_image[snode]
2544
        nimg.sinst.append(instance)
2545
        if pnode not in nimg.sbp:
2546
          nimg.sbp[pnode] = []
2547
        nimg.sbp[pnode].append(instance)
2548

    
2549
    # At this point, we have the in-memory data structures complete,
2550
    # except for the runtime information, which we'll gather next
2551

    
2552
    # Due to the way our RPC system works, exact response times cannot be
2553
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2554
    # time before and after executing the request, we can at least have a time
2555
    # window.
2556
    nvinfo_starttime = time.time()
2557
    all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
2558
                                           node_verify_param,
2559
                                           self.cfg.GetClusterName())
2560
    if self.extra_lv_nodes and vg_name is not None:
2561
      extra_lv_nvinfo = \
2562
          self.rpc.call_node_verify(self.extra_lv_nodes,
2563
                                    {constants.NV_LVLIST: vg_name},
2564
                                    self.cfg.GetClusterName())
2565
    else:
2566
      extra_lv_nvinfo = {}
2567
    nvinfo_endtime = time.time()
2568

    
2569
    all_drbd_map = self.cfg.ComputeDRBDMap()
2570

    
2571
    feedback_fn("* Gathering disk information (%s nodes)" %
2572
                len(self.my_node_names))
2573
    instdisk = self._CollectDiskInfo(self.my_node_names, node_image,
2574
                                     self.my_inst_info)
2575

    
2576
    feedback_fn("* Verifying configuration file consistency")
2577

    
2578
    # If not all nodes are being checked, we need to make sure the master node
2579
    # and a non-checked vm_capable node are in the list.
2580
    absent_nodes = set(self.all_node_info).difference(self.my_node_info)
2581
    if absent_nodes:
2582
      vf_nvinfo = all_nvinfo.copy()
2583
      vf_node_info = list(self.my_node_info.values())
2584
      additional_nodes = []
2585
      if master_node not in self.my_node_info:
2586
        additional_nodes.append(master_node)
2587
        vf_node_info.append(self.all_node_info[master_node])
2588
      # Add the first vm_capable node we find which is not included
2589
      for node in absent_nodes:
2590
        nodeinfo = self.all_node_info[node]
2591
        if nodeinfo.vm_capable and not nodeinfo.offline:
2592
          additional_nodes.append(node)
2593
          vf_node_info.append(self.all_node_info[node])
2594
          break
2595
      key = constants.NV_FILELIST
2596
      vf_nvinfo.update(self.rpc.call_node_verify(additional_nodes,
2597
                                                 {key: node_verify_param[key]},
2598
                                                 self.cfg.GetClusterName()))
2599
    else:
2600
      vf_nvinfo = all_nvinfo
2601
      vf_node_info = self.my_node_info.values()
2602

    
2603
    self._VerifyFiles(_ErrorIf, vf_node_info, master_node, vf_nvinfo, filemap)
2604

    
2605
    feedback_fn("* Verifying node status")
2606

    
2607
    refos_img = None
2608

    
2609
    for node_i in node_data_list:
2610
      node = node_i.name
2611
      nimg = node_image[node]
2612

    
2613
      if node_i.offline:
2614
        if verbose:
2615
          feedback_fn("* Skipping offline node %s" % (node,))
2616
        n_offline += 1
2617
        continue
2618

    
2619
      if node == master_node:
2620
        ntype = "master"
2621
      elif node_i.master_candidate:
2622
        ntype = "master candidate"
2623
      elif node_i.drained:
2624
        ntype = "drained"
2625
        n_drained += 1
2626
      else:
2627
        ntype = "regular"
2628
      if verbose:
2629
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2630

    
2631
      msg = all_nvinfo[node].fail_msg
2632
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2633
      if msg:
2634
        nimg.rpc_fail = True
2635
        continue
2636

    
2637
      nresult = all_nvinfo[node].payload
2638

    
2639
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2640
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2641
      self._VerifyNodeNetwork(node_i, nresult)
2642
      self._VerifyOob(node_i, nresult)
2643

    
2644
      if nimg.vm_capable:
2645
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2646
        self._VerifyNodeDrbd(node_i, nresult, self.all_inst_info, drbd_helper,
2647
                             all_drbd_map)
2648

    
2649
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2650
        self._UpdateNodeInstances(node_i, nresult, nimg)
2651
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2652
        self._UpdateNodeOS(node_i, nresult, nimg)
2653

    
2654
        if not nimg.os_fail:
2655
          if refos_img is None:
2656
            refos_img = nimg
2657
          self._VerifyNodeOS(node_i, nimg, refos_img)
2658
        self._VerifyNodeBridges(node_i, nresult, bridges)
2659

    
2660
        # Check whether all running instancies are primary for the node. (This
2661
        # can no longer be done from _VerifyInstance below, since some of the
2662
        # wrong instances could be from other node groups.)
2663
        non_primary_inst = set(nimg.instances).difference(nimg.pinst)
2664

    
2665
        for inst in non_primary_inst:
2666
          test = inst in self.all_inst_info
2667
          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
2668
                   "instance should not run on node %s", node_i.name)
2669
          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
2670
                   "node is running unknown instance %s", inst)
2671

    
2672
    for node, result in extra_lv_nvinfo.items():
2673
      self._UpdateNodeVolumes(self.all_node_info[node], result.payload,
2674
                              node_image[node], vg_name)
2675

    
2676
    feedback_fn("* Verifying instance status")
2677
    for instance in self.my_inst_names:
2678
      if verbose:
2679
        feedback_fn("* Verifying instance %s" % instance)
2680
      inst_config = self.my_inst_info[instance]
2681
      self._VerifyInstance(instance, inst_config, node_image,
2682
                           instdisk[instance])
2683
      inst_nodes_offline = []
2684

    
2685
      pnode = inst_config.primary_node
2686
      pnode_img = node_image[pnode]
2687
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2688
               self.ENODERPC, pnode, "instance %s, connection to"
2689
               " primary node failed", instance)
2690

    
2691
      _ErrorIf(inst_config.admin_up and pnode_img.offline,
2692
               self.EINSTANCEBADNODE, instance,
2693
               "instance is marked as running and lives on offline node %s",
2694
               inst_config.primary_node)
2695

    
2696
      # If the instance is non-redundant we cannot survive losing its primary
2697
      # node, so we are not N+1 compliant. On the other hand we have no disk
2698
      # templates with more than one secondary so that situation is not well
2699
      # supported either.
2700
      # FIXME: does not support file-backed instances
2701
      if not inst_config.secondary_nodes:
2702
        i_non_redundant.append(instance)
2703

    
2704
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2705
               instance, "instance has multiple secondary nodes: %s",
2706
               utils.CommaJoin(inst_config.secondary_nodes),
2707
               code=self.ETYPE_WARNING)
2708

    
2709
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2710
        pnode = inst_config.primary_node
2711
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2712
        instance_groups = {}
2713

    
2714
        for node in instance_nodes:
2715
          instance_groups.setdefault(self.all_node_info[node].group,
2716
                                     []).append(node)
2717

    
2718
        pretty_list = [
2719
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2720
          # Sort so that we always list the primary node first.
2721
          for group, nodes in sorted(instance_groups.items(),
2722
                                     key=lambda (_, nodes): pnode in nodes,
2723
                                     reverse=True)]
2724

    
2725
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2726
                      instance, "instance has primary and secondary nodes in"
2727
                      " different groups: %s", utils.CommaJoin(pretty_list),
2728
                      code=self.ETYPE_WARNING)
2729

    
2730
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2731
        i_non_a_balanced.append(instance)
2732

    
2733
      for snode in inst_config.secondary_nodes:
2734
        s_img = node_image[snode]
2735
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2736
                 "instance %s, connection to secondary node failed", instance)
2737

    
2738
        if s_img.offline:
2739
          inst_nodes_offline.append(snode)
2740

    
2741
      # warn that the instance lives on offline nodes
2742
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2743
               "instance has offline secondary node(s) %s",
2744
               utils.CommaJoin(inst_nodes_offline))
2745
      # ... or ghost/non-vm_capable nodes
2746
      for node in inst_config.all_nodes:
2747
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2748
                 "instance lives on ghost node %s", node)
2749
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2750
                 instance, "instance lives on non-vm_capable node %s", node)
2751

    
2752
    feedback_fn("* Verifying orphan volumes")
2753
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2754

    
2755
    # We will get spurious "unknown volume" warnings if any node of this group
2756
    # is secondary for an instance whose primary is in another group. To avoid
2757
    # them, we find these instances and add their volumes to node_vol_should.
2758
    for inst in self.all_inst_info.values():
2759
      for secondary in inst.secondary_nodes:
2760
        if (secondary in self.my_node_info
2761
            and inst.name not in self.my_inst_info):
2762
          inst.MapLVsByNode(node_vol_should)
2763
          break
2764

    
2765
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2766

    
2767
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2768
      feedback_fn("* Verifying N+1 Memory redundancy")
2769
      self._VerifyNPlusOneMemory(node_image, self.my_inst_info)
2770

    
2771
    feedback_fn("* Other Notes")
2772
    if i_non_redundant:
2773
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2774
                  % len(i_non_redundant))
2775

    
2776
    if i_non_a_balanced:
2777
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2778
                  % len(i_non_a_balanced))
2779

    
2780
    if n_offline:
2781
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2782

    
2783
    if n_drained:
2784
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2785

    
2786
    return not self.bad
2787

    
2788
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2789
    """Analyze the post-hooks' result
2790

2791
    This method analyses the hook result, handles it, and sends some
2792
    nicely-formatted feedback back to the user.
2793

2794
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
2795
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
2796
    @param hooks_results: the results of the multi-node hooks rpc call
2797
    @param feedback_fn: function used send feedback back to the caller
2798
    @param lu_result: previous Exec result
2799
    @return: the new Exec result, based on the previous result
2800
        and hook results
2801

2802
    """
2803
    # We only really run POST phase hooks, and are only interested in
2804
    # their results
2805
    if phase == constants.HOOKS_PHASE_POST:
2806
      # Used to change hooks' output to proper indentation
2807
      feedback_fn("* Hooks Results")
2808
      assert hooks_results, "invalid result from hooks"
2809

    
2810
      for node_name in hooks_results:
2811
        res = hooks_results[node_name]
2812
        msg = res.fail_msg
2813
        test = msg and not res.offline
2814
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
2815
                      "Communication failure in hooks execution: %s", msg)
2816
        if res.offline or msg:
2817
          # No need to investigate payload if node is offline or gave an error.
2818
          # override manually lu_result here as _ErrorIf only
2819
          # overrides self.bad
2820
          lu_result = 1
2821
          continue
2822
        for script, hkr, output in res.payload:
2823
          test = hkr == constants.HKR_FAIL
2824
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
2825
                        "Script %s failed, output:", script)
2826
          if test:
2827
            output = self._HOOKS_INDENT_RE.sub('      ', output)
2828
            feedback_fn("%s" % output)
2829
            lu_result = 0
2830

    
2831
      return lu_result
2832

    
2833

    
2834
class LUClusterVerifyDisks(NoHooksLU):
2835
  """Verifies the cluster disks status.
2836

2837
  """
2838
  REQ_BGL = False
2839

    
2840
  def ExpandNames(self):
2841
    self.needed_locks = {
2842
      locking.LEVEL_NODE: locking.ALL_SET,
2843
      locking.LEVEL_INSTANCE: locking.ALL_SET,
2844
    }
2845
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
2846

    
2847
  def Exec(self, feedback_fn):
2848
    """Verify integrity of cluster disks.
2849

2850
    @rtype: tuple of three items
2851
    @return: a tuple of (dict of node-to-node_error, list of instances
2852
        which need activate-disks, dict of instance: (node, volume) for
2853
        missing volumes
2854

2855
    """
2856
    result = res_nodes, res_instances, res_missing = {}, [], {}
2857

    
2858
    nodes = utils.NiceSort(self.cfg.GetVmCapableNodeList())
2859
    instances = self.cfg.GetAllInstancesInfo().values()
2860

    
2861
    nv_dict = {}
2862
    for inst in instances:
2863
      inst_lvs = {}
2864
      if not inst.admin_up:
2865
        continue
2866
      inst.MapLVsByNode(inst_lvs)
2867
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
2868
      for node, vol_list in inst_lvs.iteritems():
2869
        for vol in vol_list:
2870
          nv_dict[(node, vol)] = inst
2871

    
2872
    if not nv_dict:
2873
      return result
2874

    
2875
    node_lvs = self.rpc.call_lv_list(nodes, [])
2876
    for node, node_res in node_lvs.items():
2877
      if node_res.offline:
2878
        continue
2879
      msg = node_res.fail_msg
2880
      if msg:
2881
        logging.warning("Error enumerating LVs on node %s: %s", node, msg)
2882
        res_nodes[node] = msg
2883
        continue
2884

    
2885
      lvs = node_res.payload
2886
      for lv_name, (_, _, lv_online) in lvs.items():
2887
        inst = nv_dict.pop((node, lv_name), None)
2888
        if (not lv_online and inst is not None
2889
            and inst.name not in res_instances):
2890
          res_instances.append(inst.name)
2891

    
2892
    # any leftover items in nv_dict are missing LVs, let's arrange the
2893
    # data better
2894
    for key, inst in nv_dict.iteritems():
2895
      if inst.name not in res_missing:
2896
        res_missing[inst.name] = []
2897
      res_missing[inst.name].append(key)
2898

    
2899
    return result
2900

    
2901

    
2902
class LUClusterRepairDiskSizes(NoHooksLU):
2903
  """Verifies the cluster disks sizes.
2904

2905
  """
2906
  REQ_BGL = False
2907

    
2908
  def ExpandNames(self):
2909
    if self.op.instances:
2910
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
2911
      self.needed_locks = {
2912
        locking.LEVEL_NODE: [],
2913
        locking.LEVEL_INSTANCE: self.wanted_names,
2914
        }
2915
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2916
    else:
2917
      self.wanted_names = None
2918
      self.needed_locks = {
2919
        locking.LEVEL_NODE: locking.ALL_SET,
2920
        locking.LEVEL_INSTANCE: locking.ALL_SET,
2921
        }
2922
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
2923

    
2924
  def DeclareLocks(self, level):
2925
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
2926
      self._LockInstancesNodes(primary_only=True)
2927

    
2928
  def CheckPrereq(self):
2929
    """Check prerequisites.
2930

2931
    This only checks the optional instance list against the existing names.
2932

2933
    """
2934
    if self.wanted_names is None:
2935
      self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
2936

    
2937
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
2938
                             in self.wanted_names]
2939

    
2940
  def _EnsureChildSizes(self, disk):
2941
    """Ensure children of the disk have the needed disk size.
2942

2943
    This is valid mainly for DRBD8 and fixes an issue where the
2944
    children have smaller disk size.
2945

2946
    @param disk: an L{ganeti.objects.Disk} object
2947

2948
    """
2949
    if disk.dev_type == constants.LD_DRBD8:
2950
      assert disk.children, "Empty children for DRBD8?"
2951
      fchild = disk.children[0]
2952
      mismatch = fchild.size < disk.size
2953
      if mismatch:
2954
        self.LogInfo("Child disk has size %d, parent %d, fixing",
2955
                     fchild.size, disk.size)
2956
        fchild.size = disk.size
2957

    
2958
      # and we recurse on this child only, not on the metadev
2959
      return self._EnsureChildSizes(fchild) or mismatch
2960
    else:
2961
      return False
2962

    
2963
  def Exec(self, feedback_fn):
2964
    """Verify the size of cluster disks.
2965

2966
    """
2967
    # TODO: check child disks too
2968
    # TODO: check differences in size between primary/secondary nodes
2969
    per_node_disks = {}
2970
    for instance in self.wanted_instances:
2971
      pnode = instance.primary_node
2972
      if pnode not in per_node_disks:
2973
        per_node_disks[pnode] = []
2974
      for idx, disk in enumerate(instance.disks):
2975
        per_node_disks[pnode].append((instance, idx, disk))
2976

    
2977
    changed = []
2978
    for node, dskl in per_node_disks.items():
2979
      newl = [v[2].Copy() for v in dskl]
2980
      for dsk in newl:
2981
        self.cfg.SetDiskID(dsk, node)
2982
      result = self.rpc.call_blockdev_getsize(node, newl)
2983
      if result.fail_msg:
2984
        self.LogWarning("Failure in blockdev_getsize call to node"
2985
                        " %s, ignoring", node)
2986
        continue
2987
      if len(result.payload) != len(dskl):
2988
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
2989
                        " result.payload=%s", node, len(dskl), result.payload)
2990
        self.LogWarning("Invalid result from node %s, ignoring node results",
2991
                        node)
2992
        continue
2993
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
2994
        if size is None:
2995
          self.LogWarning("Disk %d of instance %s did not return size"
2996
                          " information, ignoring", idx, instance.name)
2997
          continue
2998
        if not isinstance(size, (int, long)):
2999
          self.LogWarning("Disk %d of instance %s did not return valid"
3000
                          " size information, ignoring", idx, instance.name)
3001
          continue
3002
        size = size >> 20
3003
        if size != disk.size:
3004
          self.LogInfo("Disk %d of instance %s has mismatched size,"
3005
                       " correcting: recorded %d, actual %d", idx,
3006
                       instance.name, disk.size, size)
3007
          disk.size = size
3008
          self.cfg.Update(instance, feedback_fn)
3009
          changed.append((instance.name, idx, size))
3010
        if self._EnsureChildSizes(disk):
3011
          self.cfg.Update(instance, feedback_fn)
3012
          changed.append((instance.name, idx, disk.size))
3013
    return changed
3014

    
3015

    
3016
class LUClusterRename(LogicalUnit):
3017
  """Rename the cluster.
3018

3019
  """
3020
  HPATH = "cluster-rename"
3021
  HTYPE = constants.HTYPE_CLUSTER
3022

    
3023
  def BuildHooksEnv(self):
3024
    """Build hooks env.
3025

3026
    """
3027
    return {
3028
      "OP_TARGET": self.cfg.GetClusterName(),
3029
      "NEW_NAME": self.op.name,
3030
      }
3031

    
3032
  def BuildHooksNodes(self):
3033
    """Build hooks nodes.
3034

3035
    """
3036
    return ([self.cfg.GetMasterNode()], self.cfg.GetNodeList())
3037

    
3038
  def CheckPrereq(self):
3039
    """Verify that the passed name is a valid one.
3040

3041
    """
3042
    hostname = netutils.GetHostname(name=self.op.name,
3043
                                    family=self.cfg.GetPrimaryIPFamily())
3044

    
3045
    new_name = hostname.name
3046
    self.ip = new_ip = hostname.ip
3047
    old_name = self.cfg.GetClusterName()
3048
    old_ip = self.cfg.GetMasterIP()
3049
    if new_name == old_name and new_ip == old_ip:
3050
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
3051
                                 " cluster has changed",
3052
                                 errors.ECODE_INVAL)
3053
    if new_ip != old_ip:
3054
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
3055
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
3056
                                   " reachable on the network" %
3057
                                   new_ip, errors.ECODE_NOTUNIQUE)
3058

    
3059
    self.op.name = new_name
3060

    
3061
  def Exec(self, feedback_fn):
3062
    """Rename the cluster.
3063

3064
    """
3065
    clustername = self.op.name
3066
    ip = self.ip
3067

    
3068
    # shutdown the master IP
3069
    master = self.cfg.GetMasterNode()
3070
    result = self.rpc.call_node_stop_master(master, False)
3071
    result.Raise("Could not disable the master role")
3072

    
3073
    try:
3074
      cluster = self.cfg.GetClusterInfo()
3075
      cluster.cluster_name = clustername
3076
      cluster.master_ip = ip
3077
      self.cfg.Update(cluster, feedback_fn)
3078

    
3079
      # update the known hosts file
3080
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
3081
      node_list = self.cfg.GetOnlineNodeList()
3082
      try:
3083
        node_list.remove(master)
3084
      except ValueError:
3085
        pass
3086
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
3087
    finally:
3088
      result = self.rpc.call_node_start_master(master, False, False)
3089
      msg = result.fail_msg
3090
      if msg:
3091
        self.LogWarning("Could not re-enable the master role on"
3092
                        " the master, please restart manually: %s", msg)
3093

    
3094
    return clustername
3095

    
3096

    
3097
class LUClusterSetParams(LogicalUnit):
3098
  """Change the parameters of the cluster.
3099

3100
  """
3101
  HPATH = "cluster-modify"
3102
  HTYPE = constants.HTYPE_CLUSTER
3103
  REQ_BGL = False
3104

    
3105
  def CheckArguments(self):
3106
    """Check parameters
3107

3108
    """
3109
    if self.op.uid_pool:
3110
      uidpool.CheckUidPool(self.op.uid_pool)
3111

    
3112
    if self.op.add_uids:
3113
      uidpool.CheckUidPool(self.op.add_uids)
3114

    
3115
    if self.op.remove_uids:
3116
      uidpool.CheckUidPool(self.op.remove_uids)
3117

    
3118
  def ExpandNames(self):
3119
    # FIXME: in the future maybe other cluster params won't require checking on
3120
    # all nodes to be modified.
3121
    self.needed_locks = {
3122
      locking.LEVEL_NODE: locking.ALL_SET,
3123
    }
3124
    self.share_locks[locking.LEVEL_NODE] = 1
3125

    
3126
  def BuildHooksEnv(self):
3127
    """Build hooks env.
3128

3129
    """
3130
    return {
3131
      "OP_TARGET": self.cfg.GetClusterName(),
3132
      "NEW_VG_NAME": self.op.vg_name,
3133
      }
3134

    
3135
  def BuildHooksNodes(self):
3136
    """Build hooks nodes.
3137

3138
    """
3139
    mn = self.cfg.GetMasterNode()
3140
    return ([mn], [mn])
3141

    
3142
  def CheckPrereq(self):
3143
    """Check prerequisites.
3144

3145
    This checks whether the given params don't conflict and
3146
    if the given volume group is valid.
3147

3148
    """
3149
    if self.op.vg_name is not None and not self.op.vg_name:
3150
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
3151
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
3152
                                   " instances exist", errors.ECODE_INVAL)
3153

    
3154
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
3155
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
3156
        raise errors.OpPrereqError("Cannot disable drbd helper while"
3157
                                   " drbd-based instances exist",
3158
                                   errors.ECODE_INVAL)
3159

    
3160
    node_list = self.glm.list_owned(locking.LEVEL_NODE)
3161

    
3162
    # if vg_name not None, checks given volume group on all nodes
3163
    if self.op.vg_name:
3164
      vglist = self.rpc.call_vg_list(node_list)
3165
      for node in node_list:
3166
        msg = vglist[node].fail_msg
3167
        if msg:
3168
          # ignoring down node
3169
          self.LogWarning("Error while gathering data on node %s"
3170
                          " (ignoring node): %s", node, msg)
3171
          continue
3172
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
3173
                                              self.op.vg_name,
3174
                                              constants.MIN_VG_SIZE)
3175
        if vgstatus:
3176
          raise errors.OpPrereqError("Error on node '%s': %s" %
3177
                                     (node, vgstatus), errors.ECODE_ENVIRON)
3178

    
3179
    if self.op.drbd_helper:
3180
      # checks given drbd helper on all nodes
3181
      helpers = self.rpc.call_drbd_helper(node_list)
3182
      for node in node_list:
3183
        ninfo = self.cfg.GetNodeInfo(node)
3184
        if ninfo.offline:
3185
          self.LogInfo("Not checking drbd helper on offline node %s", node)
3186
          continue
3187
        msg = helpers[node].fail_msg
3188
        if msg:
3189
          raise errors.OpPrereqError("Error checking drbd helper on node"
3190
                                     " '%s': %s" % (node, msg),
3191
                                     errors.ECODE_ENVIRON)
3192
        node_helper = helpers[node].payload
3193
        if node_helper != self.op.drbd_helper:
3194
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
3195
                                     (node, node_helper), errors.ECODE_ENVIRON)
3196

    
3197
    self.cluster = cluster = self.cfg.GetClusterInfo()
3198
    # validate params changes
3199
    if self.op.beparams:
3200
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
3201
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
3202

    
3203
    if self.op.ndparams:
3204
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
3205
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
3206

    
3207
      # TODO: we need a more general way to handle resetting
3208
      # cluster-level parameters to default values
3209
      if self.new_ndparams["oob_program"] == "":
3210
        self.new_ndparams["oob_program"] = \
3211
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
3212

    
3213
    if self.op.nicparams:
3214
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
3215
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
3216
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
3217
      nic_errors = []
3218

    
3219
      # check all instances for consistency
3220
      for instance in self.cfg.GetAllInstancesInfo().values():
3221
        for nic_idx, nic in enumerate(instance.nics):
3222
          params_copy = copy.deepcopy(nic.nicparams)
3223
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
3224

    
3225
          # check parameter syntax
3226
          try:
3227
            objects.NIC.CheckParameterSyntax(params_filled)
3228
          except errors.ConfigurationError, err:
3229
            nic_errors.append("Instance %s, nic/%d: %s" %
3230
                              (instance.name, nic_idx, err))
3231

    
3232
          # if we're moving instances to routed, check that they have an ip
3233
          target_mode = params_filled[constants.NIC_MODE]
3234
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
3235
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
3236
                              " address" % (instance.name, nic_idx))
3237
      if nic_errors:
3238
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
3239
                                   "\n".join(nic_errors))
3240

    
3241
    # hypervisor list/parameters
3242
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
3243
    if self.op.hvparams:
3244
      for hv_name, hv_dict in self.op.hvparams.items():
3245
        if hv_name not in self.new_hvparams:
3246
          self.new_hvparams[hv_name] = hv_dict
3247
        else:
3248
          self.new_hvparams[hv_name].update(hv_dict)
3249

    
3250
    # os hypervisor parameters
3251
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
3252
    if self.op.os_hvp:
3253
      for os_name, hvs in self.op.os_hvp.items():
3254
        if os_name not in self.new_os_hvp:
3255
          self.new_os_hvp[os_name] = hvs
3256
        else:
3257
          for hv_name, hv_dict in hvs.items():
3258
            if hv_name not in self.new_os_hvp[os_name]:
3259
              self.new_os_hvp[os_name][hv_name] = hv_dict
3260
            else:
3261
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
3262

    
3263
    # os parameters
3264
    self.new_osp = objects.FillDict(cluster.osparams, {})
3265
    if self.op.osparams:
3266
      for os_name, osp in self.op.osparams.items():
3267
        if os_name not in self.new_osp:
3268
          self.new_osp[os_name] = {}
3269

    
3270
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
3271
                                                  use_none=True)
3272

    
3273
        if not self.new_osp[os_name]:
3274
          # we removed all parameters
3275
          del self.new_osp[os_name]
3276
        else:
3277
          # check the parameter validity (remote check)
3278
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
3279
                         os_name, self.new_osp[os_name])
3280

    
3281
    # changes to the hypervisor list
3282
    if self.op.enabled_hypervisors is not None:
3283
      self.hv_list = self.op.enabled_hypervisors
3284
      for hv in self.hv_list:
3285
        # if the hypervisor doesn't already exist in the cluster
3286
        # hvparams, we initialize it to empty, and then (in both
3287
        # cases) we make sure to fill the defaults, as we might not
3288
        # have a complete defaults list if the hypervisor wasn't
3289
        # enabled before
3290
        if hv not in new_hvp:
3291
          new_hvp[hv] = {}
3292
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
3293
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
3294
    else:
3295
      self.hv_list = cluster.enabled_hypervisors
3296

    
3297
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
3298
      # either the enabled list has changed, or the parameters have, validate
3299
      for hv_name, hv_params in self.new_hvparams.items():
3300
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
3301
            (self.op.enabled_hypervisors and
3302
             hv_name in self.op.enabled_hypervisors)):
3303
          # either this is a new hypervisor, or its parameters have changed
3304
          hv_class = hypervisor.GetHypervisor(hv_name)
3305
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3306
          hv_class.CheckParameterSyntax(hv_params)
3307
          _CheckHVParams(self, node_list, hv_name, hv_params)
3308

    
3309
    if self.op.os_hvp:
3310
      # no need to check any newly-enabled hypervisors, since the
3311
      # defaults have already been checked in the above code-block
3312
      for os_name, os_hvp in self.new_os_hvp.items():
3313
        for hv_name, hv_params in os_hvp.items():
3314
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
3315
          # we need to fill in the new os_hvp on top of the actual hv_p
3316
          cluster_defaults = self.new_hvparams.get(hv_name, {})
3317
          new_osp = objects.FillDict(cluster_defaults, hv_params)
3318
          hv_class = hypervisor.GetHypervisor(hv_name)
3319
          hv_class.CheckParameterSyntax(new_osp)
3320
          _CheckHVParams(self, node_list, hv_name, new_osp)
3321

    
3322
    if self.op.default_iallocator:
3323
      alloc_script = utils.FindFile(self.op.default_iallocator,
3324
                                    constants.IALLOCATOR_SEARCH_PATH,
3325
                                    os.path.isfile)
3326
      if alloc_script is None:
3327
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
3328
                                   " specified" % self.op.default_iallocator,
3329
                                   errors.ECODE_INVAL)
3330

    
3331
  def Exec(self, feedback_fn):
3332
    """Change the parameters of the cluster.
3333

3334
    """
3335
    if self.op.vg_name is not None:
3336
      new_volume = self.op.vg_name
3337
      if not new_volume:
3338
        new_volume = None
3339
      if new_volume != self.cfg.GetVGName():
3340
        self.cfg.SetVGName(new_volume)
3341
      else:
3342
        feedback_fn("Cluster LVM configuration already in desired"
3343
                    " state, not changing")
3344
    if self.op.drbd_helper is not None:
3345
      new_helper = self.op.drbd_helper
3346
      if not new_helper:
3347
        new_helper = None
3348
      if new_helper != self.cfg.GetDRBDHelper():
3349
        self.cfg.SetDRBDHelper(new_helper)
3350
      else:
3351
        feedback_fn("Cluster DRBD helper already in desired state,"
3352
                    " not changing")
3353
    if self.op.hvparams:
3354
      self.cluster.hvparams = self.new_hvparams
3355
    if self.op.os_hvp:
3356
      self.cluster.os_hvp = self.new_os_hvp
3357
    if self.op.enabled_hypervisors is not None:
3358
      self.cluster.hvparams = self.new_hvparams
3359
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
3360
    if self.op.beparams:
3361
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3362
    if self.op.nicparams:
3363
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3364
    if self.op.osparams:
3365
      self.cluster.osparams = self.new_osp
3366
    if self.op.ndparams:
3367
      self.cluster.ndparams = self.new_ndparams
3368

    
3369
    if self.op.candidate_pool_size is not None:
3370
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3371
      # we need to update the pool size here, otherwise the save will fail
3372
      _AdjustCandidatePool(self, [])
3373

    
3374
    if self.op.maintain_node_health is not None:
3375
      self.cluster.maintain_node_health = self.op.maintain_node_health
3376

    
3377
    if self.op.prealloc_wipe_disks is not None:
3378
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3379

    
3380
    if self.op.add_uids is not None:
3381
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3382

    
3383
    if self.op.remove_uids is not None:
3384
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3385

    
3386
    if self.op.uid_pool is not None:
3387
      self.cluster.uid_pool = self.op.uid_pool
3388

    
3389
    if self.op.default_iallocator is not None:
3390
      self.cluster.default_iallocator = self.op.default_iallocator
3391

    
3392
    if self.op.reserved_lvs is not None:
3393
      self.cluster.reserved_lvs = self.op.reserved_lvs
3394

    
3395
    def helper_os(aname, mods, desc):
3396
      desc += " OS list"
3397
      lst = getattr(self.cluster, aname)
3398
      for key, val in mods:
3399
        if key == constants.DDM_ADD:
3400
          if val in lst:
3401
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3402
          else:
3403
            lst.append(val)
3404
        elif key == constants.DDM_REMOVE:
3405
          if val in lst:
3406
            lst.remove(val)
3407
          else:
3408
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3409
        else:
3410
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3411

    
3412
    if self.op.hidden_os:
3413
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3414

    
3415
    if self.op.blacklisted_os:
3416
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3417

    
3418
    if self.op.master_netdev:
3419
      master = self.cfg.GetMasterNode()
3420
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3421
                  self.cluster.master_netdev)
3422
      result = self.rpc.call_node_stop_master(master, False)
3423
      result.Raise("Could not disable the master ip")
3424
      feedback_fn("Changing master_netdev from %s to %s" %
3425
                  (self.cluster.master_netdev, self.op.master_netdev))
3426
      self.cluster.master_netdev = self.op.master_netdev
3427

    
3428
    self.cfg.Update(self.cluster, feedback_fn)
3429

    
3430
    if self.op.master_netdev:
3431
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3432
                  self.op.master_netdev)
3433
      result = self.rpc.call_node_start_master(master, False, False)
3434
      if result.fail_msg:
3435
        self.LogWarning("Could not re-enable the master ip on"
3436
                        " the master, please restart manually: %s",
3437
                        result.fail_msg)
3438

    
3439

    
3440
def _UploadHelper(lu, nodes, fname):
3441
  """Helper for uploading a file and showing warnings.
3442

3443
  """
3444
  if os.path.exists(fname):
3445
    result = lu.rpc.call_upload_file(nodes, fname)
3446
    for to_node, to_result in result.items():
3447
      msg = to_result.fail_msg
3448
      if msg:
3449
        msg = ("Copy of file %s to node %s failed: %s" %
3450
               (fname, to_node, msg))
3451
        lu.proc.LogWarning(msg)
3452

    
3453

    
3454
def _ComputeAncillaryFiles(cluster, redist):
3455
  """Compute files external to Ganeti which need to be consistent.
3456

3457
  @type redist: boolean
3458
  @param redist: Whether to include files which need to be redistributed
3459

3460
  """
3461
  # Compute files for all nodes
3462
  files_all = set([
3463
    constants.SSH_KNOWN_HOSTS_FILE,
3464
    constants.CONFD_HMAC_KEY,
3465
    constants.CLUSTER_DOMAIN_SECRET_FILE,
3466
    ])
3467

    
3468
  if not redist:
3469
    files_all.update(constants.ALL_CERT_FILES)
3470
    files_all.update(ssconf.SimpleStore().GetFileList())
3471

    
3472
  if cluster.modify_etc_hosts:
3473
    files_all.add(constants.ETC_HOSTS)
3474

    
3475
  # Files which must either exist on all nodes or on none
3476
  files_all_opt = set([
3477
    constants.RAPI_USERS_FILE,
3478
    ])
3479

    
3480
  # Files which should only be on master candidates
3481
  files_mc = set()
3482
  if not redist:
3483
    files_mc.add(constants.CLUSTER_CONF_FILE)
3484

    
3485
  # Files which should only be on VM-capable nodes
3486
  files_vm = set(filename
3487
    for hv_name in cluster.enabled_hypervisors
3488
    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
3489

    
3490
  # Filenames must be unique
3491
  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
3492
          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
3493
         "Found file listed in more than one file list"
3494

    
3495
  return (files_all, files_all_opt, files_mc, files_vm)
3496

    
3497

    
3498
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3499
  """Distribute additional files which are part of the cluster configuration.
3500

3501
  ConfigWriter takes care of distributing the config and ssconf files, but
3502
  there are more files which should be distributed to all nodes. This function
3503
  makes sure those are copied.
3504

3505
  @param lu: calling logical unit
3506
  @param additional_nodes: list of nodes not in the config to distribute to
3507
  @type additional_vm: boolean
3508
  @param additional_vm: whether the additional nodes are vm-capable or not
3509

3510
  """
3511
  # Gather target nodes
3512
  cluster = lu.cfg.GetClusterInfo()
3513
  master_info = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3514

    
3515
  online_nodes = lu.cfg.GetOnlineNodeList()
3516
  vm_nodes = lu.cfg.GetVmCapableNodeList()
3517

    
3518
  if additional_nodes is not None:
3519
    online_nodes.extend(additional_nodes)
3520
    if additional_vm:
3521
      vm_nodes.extend(additional_nodes)
3522

    
3523
  # Never distribute to master node
3524
  for nodelist in [online_nodes, vm_nodes]:
3525
    if master_info.name in nodelist:
3526
      nodelist.remove(master_info.name)
3527

    
3528
  # Gather file lists
3529
  (files_all, files_all_opt, files_mc, files_vm) = \
3530
    _ComputeAncillaryFiles(cluster, True)
3531

    
3532
  # Never re-distribute configuration file from here
3533
  assert not (constants.CLUSTER_CONF_FILE in files_all or
3534
              constants.CLUSTER_CONF_FILE in files_vm)
3535
  assert not files_mc, "Master candidates not handled in this function"
3536

    
3537
  filemap = [
3538
    (online_nodes, files_all),
3539
    (online_nodes, files_all_opt),
3540
    (vm_nodes, files_vm),
3541
    ]
3542

    
3543
  # Upload the files
3544
  for (node_list, files) in filemap:
3545
    for fname in files:
3546
      _UploadHelper(lu, node_list, fname)
3547

    
3548

    
3549
class LUClusterRedistConf(NoHooksLU):
3550
  """Force the redistribution of cluster configuration.
3551

3552
  This is a very simple LU.
3553

3554
  """
3555
  REQ_BGL = False
3556

    
3557
  def ExpandNames(self):
3558
    self.needed_locks = {
3559
      locking.LEVEL_NODE: locking.ALL_SET,
3560
    }
3561
    self.share_locks[locking.LEVEL_NODE] = 1
3562

    
3563
  def Exec(self, feedback_fn):
3564
    """Redistribute the configuration.
3565

3566
    """
3567
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3568
    _RedistributeAncillaryFiles(self)
3569

    
3570

    
3571
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3572
  """Sleep and poll for an instance's disk to sync.
3573

3574
  """
3575
  if not instance.disks or disks is not None and not disks:
3576
    return True
3577

    
3578
  disks = _ExpandCheckDisks(instance, disks)
3579

    
3580
  if not oneshot:
3581
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3582

    
3583
  node = instance.primary_node
3584

    
3585
  for dev in disks:
3586
    lu.cfg.SetDiskID(dev, node)
3587

    
3588
  # TODO: Convert to utils.Retry
3589

    
3590
  retries = 0
3591
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3592
  while True:
3593
    max_time = 0
3594
    done = True
3595
    cumul_degraded = False
3596
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3597
    msg = rstats.fail_msg
3598
    if msg:
3599
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3600
      retries += 1
3601
      if retries >= 10:
3602
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3603
                                 " aborting." % node)
3604
      time.sleep(6)
3605
      continue
3606
    rstats = rstats.payload
3607
    retries = 0
3608
    for i, mstat in enumerate(rstats):
3609
      if mstat is None:
3610
        lu.LogWarning("Can't compute data for node %s/%s",
3611
                           node, disks[i].iv_name)
3612
        continue
3613

    
3614
      cumul_degraded = (cumul_degraded or
3615
                        (mstat.is_degraded and mstat.sync_percent is None))
3616
      if mstat.sync_percent is not None:
3617
        done = False
3618
        if mstat.estimated_time is not None:
3619
          rem_time = ("%s remaining (estimated)" %
3620
                      utils.FormatSeconds(mstat.estimated_time))
3621
          max_time = mstat.estimated_time
3622
        else:
3623
          rem_time = "no time estimate"
3624
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3625
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3626

    
3627
    # if we're done but degraded, let's do a few small retries, to
3628
    # make sure we see a stable and not transient situation; therefore
3629
    # we force restart of the loop
3630
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3631
      logging.info("Degraded disks found, %d retries left", degr_retries)
3632
      degr_retries -= 1
3633
      time.sleep(1)
3634
      continue
3635

    
3636
    if done or oneshot:
3637
      break
3638

    
3639
    time.sleep(min(60, max_time))
3640

    
3641
  if done:
3642
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3643
  return not cumul_degraded
3644

    
3645

    
3646
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3647
  """Check that mirrors are not degraded.
3648

3649
  The ldisk parameter, if True, will change the test from the
3650
  is_degraded attribute (which represents overall non-ok status for
3651
  the device(s)) to the ldisk (representing the local storage status).
3652

3653
  """
3654
  lu.cfg.SetDiskID(dev, node)
3655

    
3656
  result = True
3657

    
3658
  if on_primary or dev.AssembleOnSecondary():
3659
    rstats = lu.rpc.call_blockdev_find(node, dev)
3660
    msg = rstats.fail_msg
3661
    if msg:
3662
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3663
      result = False
3664
    elif not rstats.payload:
3665
      lu.LogWarning("Can't find disk on node %s", node)
3666
      result = False
3667
    else:
3668
      if ldisk:
3669
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3670
      else:
3671
        result = result and not rstats.payload.is_degraded
3672

    
3673
  if dev.children:
3674
    for child in dev.children:
3675
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3676

    
3677
  return result
3678

    
3679

    
3680
class LUOobCommand(NoHooksLU):
3681
  """Logical unit for OOB handling.
3682

3683
  """
3684
  REG_BGL = False
3685
  _SKIP_MASTER = (constants.OOB_POWER_OFF, constants.OOB_POWER_CYCLE)
3686

    
3687
  def ExpandNames(self):
3688
    """Gather locks we need.
3689

3690
    """
3691
    if self.op.node_names:
3692
      self.op.node_names = _GetWantedNodes(self, self.op.node_names)
3693
      lock_names = self.op.node_names
3694
    else:
3695
      lock_names = locking.ALL_SET
3696

    
3697
    self.needed_locks = {
3698
      locking.LEVEL_NODE: lock_names,
3699
      }
3700

    
3701
  def CheckPrereq(self):
3702
    """Check prerequisites.
3703

3704
    This checks:
3705
     - the node exists in the configuration
3706
     - OOB is supported
3707

3708
    Any errors are signaled by raising errors.OpPrereqError.
3709

3710
    """
3711
    self.nodes = []
3712
    self.master_node = self.cfg.GetMasterNode()
3713

    
3714
    assert self.op.power_delay >= 0.0
3715

    
3716
    if self.op.node_names:
3717
      if (self.op.command in self._SKIP_MASTER and
3718
          self.master_node in self.op.node_names):
3719
        master_node_obj = self.cfg.GetNodeInfo(self.master_node)
3720
        master_oob_handler = _SupportsOob(self.cfg, master_node_obj)
3721

    
3722
        if master_oob_handler:
3723
          additional_text = ("run '%s %s %s' if you want to operate on the"
3724
                             " master regardless") % (master_oob_handler,
3725
                                                      self.op.command,
3726
                                                      self.master_node)
3727
        else:
3728
          additional_text = "it does not support out-of-band operations"
3729

    
3730
        raise errors.OpPrereqError(("Operating on the master node %s is not"
3731
                                    " allowed for %s; %s") %
3732
                                   (self.master_node, self.op.command,
3733
                                    additional_text), errors.ECODE_INVAL)
3734
    else:
3735
      self.op.node_names = self.cfg.GetNodeList()
3736
      if self.op.command in self._SKIP_MASTER:
3737
        self.op.node_names.remove(self.master_node)
3738

    
3739
    if self.op.command in self._SKIP_MASTER:
3740
      assert self.master_node not in self.op.node_names
3741

    
3742
    for node_name in self.op.node_names:
3743
      node = self.cfg.GetNodeInfo(node_name)
3744

    
3745
      if node is None:
3746
        raise errors.OpPrereqError("Node %s not found" % node_name,
3747
                                   errors.ECODE_NOENT)
3748
      else:
3749
        self.nodes.append(node)
3750

    
3751
      if (not self.op.ignore_status and
3752
          (self.op.command == constants.OOB_POWER_OFF and not node.offline)):
3753
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
3754
                                    " not marked offline") % node_name,
3755
                                   errors.ECODE_STATE)
3756

    
3757
  def Exec(self, feedback_fn):
3758
    """Execute OOB and return result if we expect any.
3759

3760
    """
3761
    master_node = self.master_node
3762
    ret = []
3763

    
3764
    for idx, node in enumerate(utils.NiceSort(self.nodes,
3765
                                              key=lambda node: node.name)):
3766
      node_entry = [(constants.RS_NORMAL, node.name)]
3767
      ret.append(node_entry)
3768

    
3769
      oob_program = _SupportsOob(self.cfg, node)
3770

    
3771
      if not oob_program:
3772
        node_entry.append((constants.RS_UNAVAIL, None))
3773
        continue
3774

    
3775
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
3776
                   self.op.command, oob_program, node.name)
3777
      result = self.rpc.call_run_oob(master_node, oob_program,
3778
                                     self.op.command, node.name,
3779
                                     self.op.timeout)
3780

    
3781
      if result.fail_msg:
3782
        self.LogWarning("Out-of-band RPC failed on node '%s': %s",
3783
                        node.name, result.fail_msg)
3784
        node_entry.append((constants.RS_NODATA, None))
3785
      else:
3786
        try:
3787
          self._CheckPayload(result)
3788
        except errors.OpExecError, err:
3789
          self.LogWarning("Payload returned by node '%s' is not valid: %s",
3790
                          node.name, err)
3791
          node_entry.append((constants.RS_NODATA, None))
3792
        else:
3793
          if self.op.command == constants.OOB_HEALTH:
3794
            # For health we should log important events
3795
            for item, status in result.payload:
3796
              if status in [constants.OOB_STATUS_WARNING,
3797
                            constants.OOB_STATUS_CRITICAL]:
3798
                self.LogWarning("Item '%s' on node '%s' has status '%s'",
3799
                                item, node.name, status)
3800

    
3801
          if self.op.command == constants.OOB_POWER_ON:
3802
            node.powered = True
3803
          elif self.op.command == constants.OOB_POWER_OFF:
3804
            node.powered = False
3805
          elif self.op.command == constants.OOB_POWER_STATUS:
3806
            powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
3807
            if powered != node.powered:
3808
              logging.warning(("Recorded power state (%s) of node '%s' does not"
3809
                               " match actual power state (%s)"), node.powered,
3810
                              node.name, powered)
3811

    
3812
          # For configuration changing commands we should update the node
3813
          if self.op.command in (constants.OOB_POWER_ON,
3814
                                 constants.OOB_POWER_OFF):
3815
            self.cfg.Update(node, feedback_fn)
3816

    
3817
          node_entry.append((constants.RS_NORMAL, result.payload))
3818

    
3819
          if (self.op.command == constants.OOB_POWER_ON and
3820
              idx < len(self.nodes) - 1):
3821
            time.sleep(self.op.power_delay)
3822

    
3823
    return ret
3824

    
3825
  def _CheckPayload(self, result):
3826
    """Checks if the payload is valid.
3827

3828
    @param result: RPC result
3829
    @raises errors.OpExecError: If payload is not valid
3830

3831
    """
3832
    errs = []
3833
    if self.op.command == constants.OOB_HEALTH:
3834
      if not isinstance(result.payload, list):
3835
        errs.append("command 'health' is expected to return a list but got %s" %
3836
                    type(result.payload))
3837
      else:
3838
        for item, status in result.payload:
3839
          if status not in constants.OOB_STATUSES:
3840
            errs.append("health item '%s' has invalid status '%s'" %
3841
                        (item, status))
3842

    
3843
    if self.op.command == constants.OOB_POWER_STATUS:
3844
      if not isinstance(result.payload, dict):
3845
        errs.append("power-status is expected to return a dict but got %s" %
3846
                    type(result.payload))
3847

    
3848
    if self.op.command in [
3849
        constants.OOB_POWER_ON,
3850
        constants.OOB_POWER_OFF,
3851
        constants.OOB_POWER_CYCLE,
3852
        ]:
3853
      if result.payload is not None:
3854
        errs.append("%s is expected to not return payload but got '%s'" %
3855
                    (self.op.command, result.payload))
3856

    
3857
    if errs:
3858
      raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
3859
                               utils.CommaJoin(errs))
3860

    
3861
class _OsQuery(_QueryBase):
3862
  FIELDS = query.OS_FIELDS
3863

    
3864
  def ExpandNames(self, lu):
3865
    # Lock all nodes in shared mode
3866
    # Temporary removal of locks, should be reverted later
3867
    # TODO: reintroduce locks when they are lighter-weight
3868
    lu.needed_locks = {}
3869
    #self.share_locks[locking.LEVEL_NODE] = 1
3870
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3871

    
3872
    # The following variables interact with _QueryBase._GetNames
3873
    if self.names:
3874
      self.wanted = self.names
3875
    else:
3876
      self.wanted = locking.ALL_SET
3877

    
3878
    self.do_locking = self.use_locking
3879

    
3880
  def DeclareLocks(self, lu, level):
3881
    pass
3882

    
3883
  @staticmethod
3884
  def _DiagnoseByOS(rlist):
3885
    """Remaps a per-node return list into an a per-os per-node dictionary
3886

3887
    @param rlist: a map with node names as keys and OS objects as values
3888

3889
    @rtype: dict
3890
    @return: a dictionary with osnames as keys and as value another
3891
        map, with nodes as keys and tuples of (path, status, diagnose,
3892
        variants, parameters, api_versions) as values, eg::
3893

3894
          {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
3895
                                     (/srv/..., False, "invalid api")],
3896
                           "node2": [(/srv/..., True, "", [], [])]}
3897
          }
3898

3899
    """
3900
    all_os = {}
3901
    # we build here the list of nodes that didn't fail the RPC (at RPC
3902
    # level), so that nodes with a non-responding node daemon don't
3903
    # make all OSes invalid
3904
    good_nodes = [node_name for node_name in rlist
3905
                  if not rlist[node_name].fail_msg]
3906
    for node_name, nr in rlist.items():
3907
      if nr.fail_msg or not nr.payload:
3908
        continue
3909
      for (name, path, status, diagnose, variants,
3910
           params, api_versions) in nr.payload:
3911
        if name not in all_os:
3912
          # build a list of nodes for this os containing empty lists
3913
          # for each node in node_list
3914
          all_os[name] = {}
3915
          for nname in good_nodes:
3916
            all_os[name][nname] = []
3917
        # convert params from [name, help] to (name, help)
3918
        params = [tuple(v) for v in params]
3919
        all_os[name][node_name].append((path, status, diagnose,
3920
                                        variants, params, api_versions))
3921
    return all_os
3922

    
3923
  def _GetQueryData(self, lu):
3924
    """Computes the list of nodes and their attributes.
3925

3926
    """
3927
    # Locking is not used
3928
    assert not (compat.any(lu.glm.is_owned(level)
3929
                           for level in locking.LEVELS
3930
                           if level != locking.LEVEL_CLUSTER) or
3931
                self.do_locking or self.use_locking)
3932

    
3933
    valid_nodes = [node.name
3934
                   for node in lu.cfg.GetAllNodesInfo().values()
3935
                   if not node.offline and node.vm_capable]
3936
    pol = self._DiagnoseByOS(lu.rpc.call_os_diagnose(valid_nodes))
3937
    cluster = lu.cfg.GetClusterInfo()
3938

    
3939
    data = {}
3940

    
3941
    for (os_name, os_data) in pol.items():
3942
      info = query.OsInfo(name=os_name, valid=True, node_status=os_data,
3943
                          hidden=(os_name in cluster.hidden_os),
3944
                          blacklisted=(os_name in cluster.blacklisted_os))
3945

    
3946
      variants = set()
3947
      parameters = set()
3948
      api_versions = set()
3949

    
3950
      for idx, osl in enumerate(os_data.values()):
3951
        info.valid = bool(info.valid and osl and osl[0][1])
3952
        if not info.valid:
3953
          break
3954

    
3955
        (node_variants, node_params, node_api) = osl[0][3:6]
3956
        if idx == 0:
3957
          # First entry
3958
          variants.update(node_variants)
3959
          parameters.update(node_params)
3960
          api_versions.update(node_api)
3961
        else:
3962
          # Filter out inconsistent values
3963
          variants.intersection_update(node_variants)
3964
          parameters.intersection_update(node_params)
3965
          api_versions.intersection_update(node_api)
3966

    
3967
      info.variants = list(variants)
3968
      info.parameters = list(parameters)
3969
      info.api_versions = list(api_versions)
3970

    
3971
      data[os_name] = info
3972

    
3973
    # Prepare data in requested order
3974
    return [data[name] for name in self._GetNames(lu, pol.keys(), None)
3975
            if name in data]
3976

    
3977

    
3978
class LUOsDiagnose(NoHooksLU):
3979
  """Logical unit for OS diagnose/query.
3980

3981
  """
3982
  REQ_BGL = False
3983

    
3984
  @staticmethod
3985
  def _BuildFilter(fields, names):
3986
    """Builds a filter for querying OSes.
3987

3988
    """
3989
    name_filter = qlang.MakeSimpleFilter("name", names)
3990

    
3991
    # Legacy behaviour: Hide hidden, blacklisted or invalid OSes if the
3992
    # respective field is not requested
3993
    status_filter = [[qlang.OP_NOT, [qlang.OP_TRUE, fname]]
3994
                     for fname in ["hidden", "blacklisted"]
3995
                     if fname not in fields]
3996
    if "valid" not in fields:
3997
      status_filter.append([qlang.OP_TRUE, "valid"])
3998

    
3999
    if status_filter:
4000
      status_filter.insert(0, qlang.OP_AND)
4001
    else:
4002
      status_filter = None
4003

    
4004
    if name_filter and status_filter:
4005
      return [qlang.OP_AND, name_filter, status_filter]
4006
    elif name_filter:
4007
      return name_filter
4008
    else:
4009
      return status_filter
4010

    
4011
  def CheckArguments(self):
4012
    self.oq = _OsQuery(self._BuildFilter(self.op.output_fields, self.op.names),
4013
                       self.op.output_fields, False)
4014

    
4015
  def ExpandNames(self):
4016
    self.oq.ExpandNames(self)
4017

    
4018
  def Exec(self, feedback_fn):
4019
    return self.oq.OldStyleQuery(self)
4020

    
4021

    
4022
class LUNodeRemove(LogicalUnit):
4023
  """Logical unit for removing a node.
4024

4025
  """
4026
  HPATH = "node-remove"
4027
  HTYPE = constants.HTYPE_NODE
4028

    
4029
  def BuildHooksEnv(self):
4030
    """Build hooks env.
4031

4032
    This doesn't run on the target node in the pre phase as a failed
4033
    node would then be impossible to remove.
4034

4035
    """
4036
    return {
4037
      "OP_TARGET": self.op.node_name,
4038
      "NODE_NAME": self.op.node_name,
4039
      }
4040

    
4041
  def BuildHooksNodes(self):
4042
    """Build hooks nodes.
4043

4044
    """
4045
    all_nodes = self.cfg.GetNodeList()
4046
    try:
4047
      all_nodes.remove(self.op.node_name)
4048
    except ValueError:
4049
      logging.warning("Node '%s', which is about to be removed, was not found"
4050
                      " in the list of all nodes", self.op.node_name)
4051
    return (all_nodes, all_nodes)
4052

    
4053
  def CheckPrereq(self):
4054
    """Check prerequisites.
4055

4056
    This checks:
4057
     - the node exists in the configuration
4058
     - it does not have primary or secondary instances
4059
     - it's not the master
4060

4061
    Any errors are signaled by raising errors.OpPrereqError.
4062

4063
    """
4064
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4065
    node = self.cfg.GetNodeInfo(self.op.node_name)
4066
    assert node is not None
4067

    
4068
    instance_list = self.cfg.GetInstanceList()
4069

    
4070
    masternode = self.cfg.GetMasterNode()
4071
    if node.name == masternode:
4072
      raise errors.OpPrereqError("Node is the master node, failover to another"
4073
                                 " node is required", errors.ECODE_INVAL)
4074

    
4075
    for instance_name in instance_list:
4076
      instance = self.cfg.GetInstanceInfo(instance_name)
4077
      if node.name in instance.all_nodes:
4078
        raise errors.OpPrereqError("Instance %s is still running on the node,"
4079
                                   " please remove first" % instance_name,
4080
                                   errors.ECODE_INVAL)
4081
    self.op.node_name = node.name
4082
    self.node = node
4083

    
4084
  def Exec(self, feedback_fn):
4085
    """Removes the node from the cluster.
4086

4087
    """
4088
    node = self.node
4089
    logging.info("Stopping the node daemon and removing configs from node %s",
4090
                 node.name)
4091

    
4092
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
4093

    
4094
    # Promote nodes to master candidate as needed
4095
    _AdjustCandidatePool(self, exceptions=[node.name])
4096
    self.context.RemoveNode(node.name)
4097

    
4098
    # Run post hooks on the node before it's removed
4099
    _RunPostHook(self, node.name)
4100

    
4101
    result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
4102
    msg = result.fail_msg
4103
    if msg:
4104
      self.LogWarning("Errors encountered on the remote node while leaving"
4105
                      " the cluster: %s", msg)
4106

    
4107
    # Remove node from our /etc/hosts
4108
    if self.cfg.GetClusterInfo().modify_etc_hosts:
4109
      master_node = self.cfg.GetMasterNode()
4110
      result = self.rpc.call_etc_hosts_modify(master_node,
4111
                                              constants.ETC_HOSTS_REMOVE,
4112
                                              node.name, None)
4113
      result.Raise("Can't update hosts file with new host data")
4114
      _RedistributeAncillaryFiles(self)
4115

    
4116

    
4117
class _NodeQuery(_QueryBase):
4118
  FIELDS = query.NODE_FIELDS
4119

    
4120
  def ExpandNames(self, lu):
4121
    lu.needed_locks = {}
4122
    lu.share_locks[locking.LEVEL_NODE] = 1
4123

    
4124
    if self.names:
4125
      self.wanted = _GetWantedNodes(lu, self.names)
4126
    else:
4127
      self.wanted = locking.ALL_SET
4128

    
4129
    self.do_locking = (self.use_locking and
4130
                       query.NQ_LIVE in self.requested_data)
4131

    
4132
    if self.do_locking:
4133
      # if we don't request only static fields, we need to lock the nodes
4134
      lu.needed_locks[locking.LEVEL_NODE] = self.wanted
4135

    
4136
  def DeclareLocks(self, lu, level):
4137
    pass
4138

    
4139
  def _GetQueryData(self, lu):
4140
    """Computes the list of nodes and their attributes.
4141

4142
    """
4143
    all_info = lu.cfg.GetAllNodesInfo()
4144

    
4145
    nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
4146

    
4147
    # Gather data as requested
4148
    if query.NQ_LIVE in self.requested_data:
4149
      # filter out non-vm_capable nodes
4150
      toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
4151

    
4152
      node_data = lu.rpc.call_node_info(toquery_nodes, lu.cfg.GetVGName(),
4153
                                        lu.cfg.GetHypervisorType())
4154
      live_data = dict((name, nresult.payload)
4155
                       for (name, nresult) in node_data.items()
4156
                       if not nresult.fail_msg and nresult.payload)
4157
    else:
4158
      live_data = None
4159

    
4160
    if query.NQ_INST in self.requested_data:
4161
      node_to_primary = dict([(name, set()) for name in nodenames])
4162
      node_to_secondary = dict([(name, set()) for name in nodenames])
4163

    
4164
      inst_data = lu.cfg.GetAllInstancesInfo()
4165

    
4166
      for inst in inst_data.values():
4167
        if inst.primary_node in node_to_primary:
4168
          node_to_primary[inst.primary_node].add(inst.name)
4169
        for secnode in inst.secondary_nodes:
4170
          if secnode in node_to_secondary:
4171
            node_to_secondary[secnode].add(inst.name)
4172
    else:
4173
      node_to_primary = None
4174
      node_to_secondary = None
4175

    
4176
    if query.NQ_OOB in self.requested_data:
4177
      oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
4178
                         for name, node in all_info.iteritems())
4179
    else:
4180
      oob_support = None
4181

    
4182
    if query.NQ_GROUP in self.requested_data:
4183
      groups = lu.cfg.GetAllNodeGroupsInfo()
4184
    else:
4185
      groups = {}
4186

    
4187
    return query.NodeQueryData([all_info[name] for name in nodenames],
4188
                               live_data, lu.cfg.GetMasterNode(),
4189
                               node_to_primary, node_to_secondary, groups,
4190
                               oob_support, lu.cfg.GetClusterInfo())
4191

    
4192

    
4193
class LUNodeQuery(NoHooksLU):
4194
  """Logical unit for querying nodes.
4195

4196
  """
4197
  # pylint: disable-msg=W0142
4198
  REQ_BGL = False
4199

    
4200
  def CheckArguments(self):
4201
    self.nq = _NodeQuery(qlang.MakeSimpleFilter("name", self.op.names),
4202
                         self.op.output_fields, self.op.use_locking)
4203

    
4204
  def ExpandNames(self):
4205
    self.nq.ExpandNames(self)
4206

    
4207
  def Exec(self, feedback_fn):
4208
    return self.nq.OldStyleQuery(self)
4209

    
4210

    
4211
class LUNodeQueryvols(NoHooksLU):
4212
  """Logical unit for getting volumes on node(s).
4213

4214
  """
4215
  REQ_BGL = False
4216
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
4217
  _FIELDS_STATIC = utils.FieldSet("node")
4218

    
4219
  def CheckArguments(self):
4220
    _CheckOutputFields(static=self._FIELDS_STATIC,
4221
                       dynamic=self._FIELDS_DYNAMIC,
4222
                       selected=self.op.output_fields)
4223

    
4224
  def ExpandNames(self):
4225
    self.needed_locks = {}
4226
    self.share_locks[locking.LEVEL_NODE] = 1
4227
    if not self.op.nodes:
4228
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4229
    else:
4230
      self.needed_locks[locking.LEVEL_NODE] = \
4231
        _GetWantedNodes(self, self.op.nodes)
4232

    
4233
  def Exec(self, feedback_fn):
4234
    """Computes the list of nodes and their attributes.
4235

4236
    """
4237
    nodenames = self.glm.list_owned(locking.LEVEL_NODE)
4238
    volumes = self.rpc.call_node_volumes(nodenames)
4239

    
4240
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
4241
             in self.cfg.GetInstanceList()]
4242

    
4243
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
4244

    
4245
    output = []
4246
    for node in nodenames:
4247
      nresult = volumes[node]
4248
      if nresult.offline:
4249
        continue
4250
      msg = nresult.fail_msg
4251
      if msg:
4252
        self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
4253
        continue
4254

    
4255
      node_vols = nresult.payload[:]
4256
      node_vols.sort(key=lambda vol: vol['dev'])
4257

    
4258
      for vol in node_vols:
4259
        node_output = []
4260
        for field in self.op.output_fields:
4261
          if field == "node":
4262
            val = node
4263
          elif field == "phys":
4264
            val = vol['dev']
4265
          elif field == "vg":
4266
            val = vol['vg']
4267
          elif field == "name":
4268
            val = vol['name']
4269
          elif field == "size":
4270
            val = int(float(vol['size']))
4271
          elif field == "instance":
4272
            for inst in ilist:
4273
              if node not in lv_by_node[inst]:
4274
                continue
4275
              if vol['name'] in lv_by_node[inst][node]:
4276
                val = inst.name
4277
                break
4278
            else:
4279
              val = '-'
4280
          else:
4281
            raise errors.ParameterError(field)
4282
          node_output.append(str(val))
4283

    
4284
        output.append(node_output)
4285

    
4286
    return output
4287

    
4288

    
4289
class LUNodeQueryStorage(NoHooksLU):
4290
  """Logical unit for getting information on storage units on node(s).
4291

4292
  """
4293
  _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
4294
  REQ_BGL = False
4295

    
4296
  def CheckArguments(self):
4297
    _CheckOutputFields(static=self._FIELDS_STATIC,
4298
                       dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
4299
                       selected=self.op.output_fields)
4300

    
4301
  def ExpandNames(self):
4302
    self.needed_locks = {}
4303
    self.share_locks[locking.LEVEL_NODE] = 1
4304

    
4305
    if self.op.nodes:
4306
      self.needed_locks[locking.LEVEL_NODE] = \
4307
        _GetWantedNodes(self, self.op.nodes)
4308
    else:
4309
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4310

    
4311
  def Exec(self, feedback_fn):
4312
    """Computes the list of nodes and their attributes.
4313

4314
    """
4315
    self.nodes = self.glm.list_owned(locking.LEVEL_NODE)
4316

    
4317
    # Always get name to sort by
4318
    if constants.SF_NAME in self.op.output_fields:
4319
      fields = self.op.output_fields[:]
4320
    else:
4321
      fields = [constants.SF_NAME] + self.op.output_fields
4322

    
4323
    # Never ask for node or type as it's only known to the LU
4324
    for extra in [constants.SF_NODE, constants.SF_TYPE]:
4325
      while extra in fields:
4326
        fields.remove(extra)
4327

    
4328
    field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
4329
    name_idx = field_idx[constants.SF_NAME]
4330

    
4331
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4332
    data = self.rpc.call_storage_list(self.nodes,
4333
                                      self.op.storage_type, st_args,
4334
                                      self.op.name, fields)
4335

    
4336
    result = []
4337

    
4338
    for node in utils.NiceSort(self.nodes):
4339
      nresult = data[node]
4340
      if nresult.offline:
4341
        continue
4342

    
4343
      msg = nresult.fail_msg
4344
      if msg:
4345
        self.LogWarning("Can't get storage data from node %s: %s", node, msg)
4346
        continue
4347

    
4348
      rows = dict([(row[name_idx], row) for row in nresult.payload])
4349

    
4350
      for name in utils.NiceSort(rows.keys()):
4351
        row = rows[name]
4352

    
4353
        out = []
4354

    
4355
        for field in self.op.output_fields:
4356
          if field == constants.SF_NODE:
4357
            val = node
4358
          elif field == constants.SF_TYPE:
4359
            val = self.op.storage_type
4360
          elif field in field_idx:
4361
            val = row[field_idx[field]]
4362
          else:
4363
            raise errors.ParameterError(field)
4364

    
4365
          out.append(val)
4366

    
4367
        result.append(out)
4368

    
4369
    return result
4370

    
4371

    
4372
class _InstanceQuery(_QueryBase):
4373
  FIELDS = query.INSTANCE_FIELDS
4374

    
4375
  def ExpandNames(self, lu):
4376
    lu.needed_locks = {}
4377
    lu.share_locks[locking.LEVEL_INSTANCE] = 1
4378
    lu.share_locks[locking.LEVEL_NODE] = 1
4379

    
4380
    if self.names:
4381
      self.wanted = _GetWantedInstances(lu, self.names)
4382
    else:
4383
      self.wanted = locking.ALL_SET
4384

    
4385
    self.do_locking = (self.use_locking and
4386
                       query.IQ_LIVE in self.requested_data)
4387
    if self.do_locking:
4388
      lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
4389
      lu.needed_locks[locking.LEVEL_NODE] = []
4390
      lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4391

    
4392
  def DeclareLocks(self, lu, level):
4393
    if level == locking.LEVEL_NODE and self.do_locking:
4394
      lu._LockInstancesNodes() # pylint: disable-msg=W0212
4395

    
4396
  def _GetQueryData(self, lu):
4397
    """Computes the list of instances and their attributes.
4398

4399
    """
4400
    cluster = lu.cfg.GetClusterInfo()
4401
    all_info = lu.cfg.GetAllInstancesInfo()
4402

    
4403
    instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
4404

    
4405
    instance_list = [all_info[name] for name in instance_names]
4406
    nodes = frozenset(itertools.chain(*(inst.all_nodes
4407
                                        for inst in instance_list)))
4408
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
4409
    bad_nodes = []
4410
    offline_nodes = []
4411
    wrongnode_inst = set()
4412

    
4413
    # Gather data as requested
4414
    if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
4415
      live_data = {}
4416
      node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
4417
      for name in nodes:
4418
        result = node_data[name]
4419
        if result.offline:
4420
          # offline nodes will be in both lists
4421
          assert result.fail_msg
4422
          offline_nodes.append(name)
4423
        if result.fail_msg:
4424
          bad_nodes.append(name)
4425
        elif result.payload:
4426
          for inst in result.payload:
4427
            if inst in all_info:
4428
              if all_info[inst].primary_node == name:
4429
                live_data.update(result.payload)
4430
              else:
4431
                wrongnode_inst.add(inst)
4432
            else:
4433
              # orphan instance; we don't list it here as we don't
4434
              # handle this case yet in the output of instance listing
4435
              logging.warning("Orphan instance '%s' found on node %s",
4436
                              inst, name)
4437
        # else no instance is alive
4438
    else:
4439
      live_data = {}
4440

    
4441
    if query.IQ_DISKUSAGE in self.requested_data:
4442
      disk_usage = dict((inst.name,
4443
                         _ComputeDiskSize(inst.disk_template,
4444
                                          [{constants.IDISK_SIZE: disk.size}
4445
                                           for disk in inst.disks]))
4446
                        for inst in instance_list)
4447
    else:
4448
      disk_usage = None
4449

    
4450
    if query.IQ_CONSOLE in self.requested_data:
4451
      consinfo = {}
4452
      for inst in instance_list:
4453
        if inst.name in live_data:
4454
          # Instance is running
4455
          consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
4456
        else:
4457
          consinfo[inst.name] = None
4458
      assert set(consinfo.keys()) == set(instance_names)
4459
    else:
4460
      consinfo = None
4461

    
4462
    return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
4463
                                   disk_usage, offline_nodes, bad_nodes,
4464
                                   live_data, wrongnode_inst, consinfo)
4465

    
4466

    
4467
class LUQuery(NoHooksLU):
4468
  """Query for resources/items of a certain kind.
4469

4470
  """
4471
  # pylint: disable-msg=W0142
4472
  REQ_BGL = False
4473

    
4474
  def CheckArguments(self):
4475
    qcls = _GetQueryImplementation(self.op.what)
4476

    
4477
    self.impl = qcls(self.op.filter, self.op.fields, False)
4478

    
4479
  def ExpandNames(self):
4480
    self.impl.ExpandNames(self)
4481

    
4482
  def DeclareLocks(self, level):
4483
    self.impl.DeclareLocks(self, level)
4484

    
4485
  def Exec(self, feedback_fn):
4486
    return self.impl.NewStyleQuery(self)
4487

    
4488

    
4489
class LUQueryFields(NoHooksLU):
4490
  """Query for resources/items of a certain kind.
4491

4492
  """
4493
  # pylint: disable-msg=W0142
4494
  REQ_BGL = False
4495

    
4496
  def CheckArguments(self):
4497
    self.qcls = _GetQueryImplementation(self.op.what)
4498

    
4499
  def ExpandNames(self):
4500
    self.needed_locks = {}
4501

    
4502
  def Exec(self, feedback_fn):
4503
    return query.QueryFields(self.qcls.FIELDS, self.op.fields)
4504

    
4505

    
4506
class LUNodeModifyStorage(NoHooksLU):
4507
  """Logical unit for modifying a storage volume on a node.
4508

4509
  """
4510
  REQ_BGL = False
4511

    
4512
  def CheckArguments(self):
4513
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4514

    
4515
    storage_type = self.op.storage_type
4516

    
4517
    try:
4518
      modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
4519
    except KeyError:
4520
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
4521
                                 " modified" % storage_type,
4522
                                 errors.ECODE_INVAL)
4523

    
4524
    diff = set(self.op.changes.keys()) - modifiable
4525
    if diff:
4526
      raise errors.OpPrereqError("The following fields can not be modified for"
4527
                                 " storage units of type '%s': %r" %
4528
                                 (storage_type, list(diff)),
4529
                                 errors.ECODE_INVAL)
4530

    
4531
  def ExpandNames(self):
4532
    self.needed_locks = {
4533
      locking.LEVEL_NODE: self.op.node_name,
4534
      }
4535

    
4536
  def Exec(self, feedback_fn):
4537
    """Computes the list of nodes and their attributes.
4538

4539
    """
4540
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4541
    result = self.rpc.call_storage_modify(self.op.node_name,
4542
                                          self.op.storage_type, st_args,
4543
                                          self.op.name, self.op.changes)
4544
    result.Raise("Failed to modify storage unit '%s' on %s" %
4545
                 (self.op.name, self.op.node_name))
4546

    
4547

    
4548
class LUNodeAdd(LogicalUnit):
4549
  """Logical unit for adding node to the cluster.
4550

4551
  """
4552
  HPATH = "node-add"
4553
  HTYPE = constants.HTYPE_NODE
4554
  _NFLAGS = ["master_capable", "vm_capable"]
4555

    
4556
  def CheckArguments(self):
4557
    self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
4558
    # validate/normalize the node name
4559
    self.hostname = netutils.GetHostname(name=self.op.node_name,
4560
                                         family=self.primary_ip_family)
4561
    self.op.node_name = self.hostname.name
4562

    
4563
    if self.op.readd and self.op.node_name == self.cfg.GetMasterNode():
4564
      raise errors.OpPrereqError("Cannot readd the master node",
4565
                                 errors.ECODE_STATE)
4566

    
4567
    if self.op.readd and self.op.group:
4568
      raise errors.OpPrereqError("Cannot pass a node group when a node is"
4569
                                 " being readded", errors.ECODE_INVAL)
4570

    
4571
  def BuildHooksEnv(self):
4572
    """Build hooks env.
4573

4574
    This will run on all nodes before, and on all nodes + the new node after.
4575

4576
    """
4577
    return {
4578
      "OP_TARGET": self.op.node_name,
4579
      "NODE_NAME": self.op.node_name,
4580
      "NODE_PIP": self.op.primary_ip,
4581
      "NODE_SIP": self.op.secondary_ip,
4582
      "MASTER_CAPABLE": str(self.op.master_capable),
4583
      "VM_CAPABLE": str(self.op.vm_capable),
4584
      }
4585

    
4586
  def BuildHooksNodes(self):
4587
    """Build hooks nodes.
4588

4589
    """
4590
    # Exclude added node
4591
    pre_nodes = list(set(self.cfg.GetNodeList()) - set([self.op.node_name]))
4592
    post_nodes = pre_nodes + [self.op.node_name, ]
4593

    
4594
    return (pre_nodes, post_nodes)
4595

    
4596
  def CheckPrereq(self):
4597
    """Check prerequisites.
4598

4599
    This checks:
4600
     - the new node is not already in the config
4601
     - it is resolvable
4602
     - its parameters (single/dual homed) matches the cluster
4603

4604
    Any errors are signaled by raising errors.OpPrereqError.
4605

4606
    """
4607
    cfg = self.cfg
4608
    hostname = self.hostname
4609
    node = hostname.name
4610
    primary_ip = self.op.primary_ip = hostname.ip
4611
    if self.op.secondary_ip is None:
4612
      if self.primary_ip_family == netutils.IP6Address.family:
4613
        raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
4614
                                   " IPv4 address must be given as secondary",
4615
                                   errors.ECODE_INVAL)
4616
      self.op.secondary_ip = primary_ip
4617

    
4618
    secondary_ip = self.op.secondary_ip
4619
    if not netutils.IP4Address.IsValid(secondary_ip):
4620
      raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
4621
                                 " address" % secondary_ip, errors.ECODE_INVAL)
4622

    
4623
    node_list = cfg.GetNodeList()
4624
    if not self.op.readd and node in node_list:
4625
      raise errors.OpPrereqError("Node %s is already in the configuration" %
4626
                                 node, errors.ECODE_EXISTS)
4627
    elif self.op.readd and node not in node_list:
4628
      raise errors.OpPrereqError("Node %s is not in the configuration" % node,
4629
                                 errors.ECODE_NOENT)
4630

    
4631
    self.changed_primary_ip = False
4632

    
4633
    for existing_node_name in node_list:
4634
      existing_node = cfg.GetNodeInfo(existing_node_name)
4635

    
4636
      if self.op.readd and node == existing_node_name:
4637
        if existing_node.secondary_ip != secondary_ip:
4638
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
4639
                                     " address configuration as before",
4640
                                     errors.ECODE_INVAL)
4641
        if existing_node.primary_ip != primary_ip:
4642
          self.changed_primary_ip = True
4643

    
4644
        continue
4645

    
4646
      if (existing_node.primary_ip == primary_ip or
4647
          existing_node.secondary_ip == primary_ip or
4648
          existing_node.primary_ip == secondary_ip or
4649
          existing_node.secondary_ip == secondary_ip):
4650
        raise errors.OpPrereqError("New node ip address(es) conflict with"
4651
                                   " existing node %s" % existing_node.name,
4652
                                   errors.ECODE_NOTUNIQUE)
4653

    
4654
    # After this 'if' block, None is no longer a valid value for the
4655
    # _capable op attributes
4656
    if self.op.readd:
4657
      old_node = self.cfg.GetNodeInfo(node)
4658
      assert old_node is not None, "Can't retrieve locked node %s" % node
4659
      for attr in self._NFLAGS:
4660
        if getattr(self.op, attr) is None:
4661
          setattr(self.op, attr, getattr(old_node, attr))
4662
    else:
4663
      for attr in self._NFLAGS:
4664
        if getattr(self.op, attr) is None:
4665
          setattr(self.op, attr, True)
4666

    
4667
    if self.op.readd and not self.op.vm_capable:
4668
      pri, sec = cfg.GetNodeInstances(node)
4669
      if pri or sec:
4670
        raise errors.OpPrereqError("Node %s being re-added with vm_capable"
4671
                                   " flag set to false, but it already holds"
4672
                                   " instances" % node,
4673
                                   errors.ECODE_STATE)
4674

    
4675
    # check that the type of the node (single versus dual homed) is the
4676
    # same as for the master
4677
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
4678
    master_singlehomed = myself.secondary_ip == myself.primary_ip
4679
    newbie_singlehomed = secondary_ip == primary_ip
4680
    if master_singlehomed != newbie_singlehomed:
4681
      if master_singlehomed:
4682
        raise errors.OpPrereqError("The master has no secondary ip but the"
4683
                                   " new node has one",
4684
                                   errors.ECODE_INVAL)
4685
      else:
4686
        raise errors.OpPrereqError("The master has a secondary ip but the"
4687
                                   " new node doesn't have one",
4688
                                   errors.ECODE_INVAL)
4689

    
4690
    # checks reachability
4691
    if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
4692
      raise errors.OpPrereqError("Node not reachable by ping",
4693
                                 errors.ECODE_ENVIRON)
4694

    
4695
    if not newbie_singlehomed:
4696
      # check reachability from my secondary ip to newbie's secondary ip
4697
      if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
4698
                           source=myself.secondary_ip):
4699
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
4700
                                   " based ping to node daemon port",
4701
                                   errors.ECODE_ENVIRON)
4702

    
4703
    if self.op.readd:
4704
      exceptions = [node]
4705
    else:
4706
      exceptions = []
4707

    
4708
    if self.op.master_capable:
4709
      self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
4710
    else:
4711
      self.master_candidate = False
4712

    
4713
    if self.op.readd:
4714
      self.new_node = old_node
4715
    else:
4716
      node_group = cfg.LookupNodeGroup(self.op.group)
4717
      self.new_node = objects.Node(name=node,
4718
                                   primary_ip=primary_ip,
4719
                                   secondary_ip=secondary_ip,
4720
                                   master_candidate=self.master_candidate,
4721
                                   offline=False, drained=False,
4722
                                   group=node_group)
4723

    
4724
    if self.op.ndparams:
4725
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
4726

    
4727
  def Exec(self, feedback_fn):
4728
    """Adds the new node to the cluster.
4729

4730
    """
4731
    new_node = self.new_node
4732
    node = new_node.name
4733

    
4734
    # We adding a new node so we assume it's powered
4735
    new_node.powered = True
4736

    
4737
    # for re-adds, reset the offline/drained/master-candidate flags;
4738
    # we need to reset here, otherwise offline would prevent RPC calls
4739
    # later in the procedure; this also means that if the re-add
4740
    # fails, we are left with a non-offlined, broken node
4741
    if self.op.readd:
4742
      new_node.drained = new_node.offline = False # pylint: disable-msg=W0201
4743
      self.LogInfo("Readding a node, the offline/drained flags were reset")
4744
      # if we demote the node, we do cleanup later in the procedure
4745
      new_node.master_candidate = self.master_candidate
4746
      if self.changed_primary_ip:
4747
        new_node.primary_ip = self.op.primary_ip
4748

    
4749
    # copy the master/vm_capable flags
4750
    for attr in self._NFLAGS:
4751
      setattr(new_node, attr, getattr(self.op, attr))
4752

    
4753
    # notify the user about any possible mc promotion
4754
    if new_node.master_candidate:
4755
      self.LogInfo("Node will be a master candidate")
4756

    
4757
    if self.op.ndparams:
4758
      new_node.ndparams = self.op.ndparams
4759
    else:
4760
      new_node.ndparams = {}
4761

    
4762
    # check connectivity
4763
    result = self.rpc.call_version([node])[node]
4764
    result.Raise("Can't get version information from node %s" % node)
4765
    if constants.PROTOCOL_VERSION == result.payload:
4766
      logging.info("Communication to node %s fine, sw version %s match",
4767
                   node, result.payload)
4768
    else:
4769
      raise errors.OpExecError("Version mismatch master version %s,"
4770
                               " node version %s" %
4771
                               (constants.PROTOCOL_VERSION, result.payload))
4772

    
4773
    # Add node to our /etc/hosts, and add key to known_hosts
4774
    if self.cfg.GetClusterInfo().modify_etc_hosts:
4775
      master_node = self.cfg.GetMasterNode()
4776
      result = self.rpc.call_etc_hosts_modify(master_node,
4777
                                              constants.ETC_HOSTS_ADD,
4778
                                              self.hostname.name,
4779
                                              self.hostname.ip)
4780
      result.Raise("Can't update hosts file with new host data")
4781

    
4782
    if new_node.secondary_ip != new_node.primary_ip:
4783
      _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
4784
                               False)
4785

    
4786
    node_verify_list = [self.cfg.GetMasterNode()]
4787
    node_verify_param = {
4788
      constants.NV_NODELIST: [node],
4789
      # TODO: do a node-net-test as well?
4790
    }
4791

    
4792
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
4793
                                       self.cfg.GetClusterName())
4794
    for verifier in node_verify_list:
4795
      result[verifier].Raise("Cannot communicate with node %s" % verifier)
4796
      nl_payload = result[verifier].payload[constants.NV_NODELIST]
4797
      if nl_payload:
4798
        for failed in nl_payload:
4799
          feedback_fn("ssh/hostname verification failed"
4800
                      " (checking from %s): %s" %
4801
                      (verifier, nl_payload[failed]))
4802
        raise errors.OpExecError("ssh/hostname verification failed")
4803

    
4804
    if self.op.readd:
4805
      _RedistributeAncillaryFiles(self)
4806
      self.context.ReaddNode(new_node)
4807
      # make sure we redistribute the config
4808
      self.cfg.Update(new_node, feedback_fn)
4809
      # and make sure the new node will not have old files around
4810
      if not new_node.master_candidate:
4811
        result = self.rpc.call_node_demote_from_mc(new_node.name)
4812
        msg = result.fail_msg
4813
        if msg:
4814
          self.LogWarning("Node failed to demote itself from master"
4815
                          " candidate status: %s" % msg)
4816
    else:
4817
      _RedistributeAncillaryFiles(self, additional_nodes=[node],
4818
                                  additional_vm=self.op.vm_capable)
4819
      self.context.AddNode(new_node, self.proc.GetECId())
4820

    
4821

    
4822
class LUNodeSetParams(LogicalUnit):
4823
  """Modifies the parameters of a node.
4824

4825
  @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
4826
      to the node role (as _ROLE_*)
4827
  @cvar _R2F: a dictionary from node role to tuples of flags
4828
  @cvar _FLAGS: a list of attribute names corresponding to the flags
4829

4830
  """
4831
  HPATH = "node-modify"
4832
  HTYPE = constants.HTYPE_NODE
4833
  REQ_BGL = False
4834
  (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
4835
  _F2R = {
4836
    (True, False, False): _ROLE_CANDIDATE,
4837
    (False, True, False): _ROLE_DRAINED,
4838
    (False, False, True): _ROLE_OFFLINE,
4839
    (False, False, False): _ROLE_REGULAR,
4840
    }
4841
  _R2F = dict((v, k) for k, v in _F2R.items())
4842
  _FLAGS = ["master_candidate", "drained", "offline"]
4843

    
4844
  def CheckArguments(self):
4845
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4846
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
4847
                self.op.master_capable, self.op.vm_capable,
4848
                self.op.secondary_ip, self.op.ndparams]
4849
    if all_mods.count(None) == len(all_mods):
4850
      raise errors.OpPrereqError("Please pass at least one modification",
4851
                                 errors.ECODE_INVAL)
4852
    if all_mods.count(True) > 1:
4853
      raise errors.OpPrereqError("Can't set the node into more than one"
4854
                                 " state at the same time",
4855
                                 errors.ECODE_INVAL)
4856

    
4857
    # Boolean value that tells us whether we might be demoting from MC
4858
    self.might_demote = (self.op.master_candidate == False or
4859
                         self.op.offline == True or
4860
                         self.op.drained == True or
4861
                         self.op.master_capable == False)
4862

    
4863
    if self.op.secondary_ip:
4864
      if not netutils.IP4Address.IsValid(self.op.secondary_ip):
4865
        raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
4866
                                   " address" % self.op.secondary_ip,
4867
                                   errors.ECODE_INVAL)
4868

    
4869
    self.lock_all = self.op.auto_promote and self.might_demote
4870
    self.lock_instances = self.op.secondary_ip is not None
4871

    
4872
  def ExpandNames(self):
4873
    if self.lock_all:
4874
      self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
4875
    else:
4876
      self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
4877

    
4878
    if self.lock_instances:
4879
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
4880

    
4881
  def DeclareLocks(self, level):
4882
    # If we have locked all instances, before waiting to lock nodes, release
4883
    # all the ones living on nodes unrelated to the current operation.
4884
    if level == locking.LEVEL_NODE and self.lock_instances:
4885
      self.affected_instances = []
4886
      if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4887
        instances_keep = []
4888

    
4889
        # Build list of instances to release
4890
        for instance_name in self.glm.list_owned(locking.LEVEL_INSTANCE):
4891
          instance = self.context.cfg.GetInstanceInfo(instance_name)
4892
          if (instance.disk_template in constants.DTS_INT_MIRROR and
4893
              self.op.node_name in instance.all_nodes):
4894
            instances_keep.append(instance_name)
4895
            self.affected_instances.append(instance)
4896

    
4897
        _ReleaseLocks(self, locking.LEVEL_INSTANCE, keep=instances_keep)
4898

    
4899
        assert (set(self.glm.list_owned(locking.LEVEL_INSTANCE)) ==
4900
                set(instances_keep))
4901

    
4902
  def BuildHooksEnv(self):
4903
    """Build hooks env.
4904

4905
    This runs on the master node.
4906

4907
    """
4908
    return {
4909
      "OP_TARGET": self.op.node_name,
4910
      "MASTER_CANDIDATE": str(self.op.master_candidate),
4911
      "OFFLINE": str(self.op.offline),
4912
      "DRAINED": str(self.op.drained),
4913
      "MASTER_CAPABLE": str(self.op.master_capable),
4914
      "VM_CAPABLE": str(self.op.vm_capable),
4915
      }
4916

    
4917
  def BuildHooksNodes(self):
4918
    """Build hooks nodes.
4919

4920
    """
4921
    nl = [self.cfg.GetMasterNode(), self.op.node_name]
4922
    return (nl, nl)
4923

    
4924
  def CheckPrereq(self):
4925
    """Check prerequisites.
4926

4927
    This only checks the instance list against the existing names.
4928

4929
    """
4930
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
4931

    
4932
    if (self.op.master_candidate is not None or
4933
        self.op.drained is not None or
4934
        self.op.offline is not None):
4935
      # we can't change the master's node flags
4936
      if self.op.node_name == self.cfg.GetMasterNode():
4937
        raise errors.OpPrereqError("The master role can be changed"
4938
                                   " only via master-failover",
4939
                                   errors.ECODE_INVAL)
4940

    
4941
    if self.op.master_candidate and not node.master_capable:
4942
      raise errors.OpPrereqError("Node %s is not master capable, cannot make"
4943
                                 " it a master candidate" % node.name,
4944
                                 errors.ECODE_STATE)
4945

    
4946
    if self.op.vm_capable == False:
4947
      (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
4948
      if ipri or isec:
4949
        raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
4950
                                   " the vm_capable flag" % node.name,
4951
                                   errors.ECODE_STATE)
4952

    
4953
    if node.master_candidate and self.might_demote and not self.lock_all:
4954
      assert not self.op.auto_promote, "auto_promote set but lock_all not"
4955
      # check if after removing the current node, we're missing master
4956
      # candidates
4957
      (mc_remaining, mc_should, _) = \
4958
          self.cfg.GetMasterCandidateStats(exceptions=[node.name])
4959
      if mc_remaining < mc_should:
4960
        raise errors.OpPrereqError("Not enough master candidates, please"
4961
                                   " pass auto promote option to allow"
4962
                                   " promotion", errors.ECODE_STATE)
4963

    
4964
    self.old_flags = old_flags = (node.master_candidate,
4965
                                  node.drained, node.offline)
4966
    assert old_flags in self._F2R, "Un-handled old flags %s" % str(old_flags)
4967
    self.old_role = old_role = self._F2R[old_flags]
4968

    
4969
    # Check for ineffective changes
4970
    for attr in self._FLAGS:
4971
      if (getattr(self.op, attr) == False and getattr(node, attr) == False):
4972
        self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
4973
        setattr(self.op, attr, None)
4974

    
4975
    # Past this point, any flag change to False means a transition
4976
    # away from the respective state, as only real changes are kept
4977

    
4978
    # TODO: We might query the real power state if it supports OOB
4979
    if _SupportsOob(self.cfg, node):
4980
      if self.op.offline is False and not (node.powered or
4981
                                           self.op.powered == True):
4982
        raise errors.OpPrereqError(("Node %s needs to be turned on before its"
4983
                                    " offline status can be reset") %
4984
                                   self.op.node_name)
4985
    elif self.op.powered is not None:
4986
      raise errors.OpPrereqError(("Unable to change powered state for node %s"
4987
                                  " as it does not support out-of-band"
4988
                                  " handling") % self.op.node_name)
4989

    
4990
    # If we're being deofflined/drained, we'll MC ourself if needed
4991
    if (self.op.drained == False or self.op.offline == False or
4992
        (self.op.master_capable and not node.master_capable)):
4993
      if _DecideSelfPromotion(self):
4994
        self.op.master_candidate = True
4995
        self.LogInfo("Auto-promoting node to master candidate")
4996

    
4997
    # If we're no longer master capable, we'll demote ourselves from MC
4998
    if self.op.master_capable == False and node.master_candidate:
4999
      self.LogInfo("Demoting from master candidate")
5000
      self.op.master_candidate = False
5001

    
5002
    # Compute new role
5003
    assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
5004
    if self.op.master_candidate:
5005
      new_role = self._ROLE_CANDIDATE
5006
    elif self.op.drained:
5007
      new_role = self._ROLE_DRAINED
5008
    elif self.op.offline:
5009
      new_role = self._ROLE_OFFLINE
5010
    elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
5011
      # False is still in new flags, which means we're un-setting (the
5012
      # only) True flag
5013
      new_role = self._ROLE_REGULAR
5014
    else: # no new flags, nothing, keep old role
5015
      new_role = old_role
5016

    
5017
    self.new_role = new_role
5018

    
5019
    if old_role == self._ROLE_OFFLINE and new_role != old_role:
5020
      # Trying to transition out of offline status
5021
      result = self.rpc.call_version([node.name])[node.name]
5022
      if result.fail_msg:
5023
        raise errors.OpPrereqError("Node %s is being de-offlined but fails"
5024
                                   " to report its version: %s" %
5025
                                   (node.name, result.fail_msg),
5026
                                   errors.ECODE_STATE)
5027
      else:
5028
        self.LogWarning("Transitioning node from offline to online state"
5029
                        " without using re-add. Please make sure the node"
5030
                        " is healthy!")
5031

    
5032
    if self.op.secondary_ip:
5033
      # Ok even without locking, because this can't be changed by any LU
5034
      master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
5035
      master_singlehomed = master.secondary_ip == master.primary_ip
5036
      if master_singlehomed and self.op.secondary_ip:
5037
        raise errors.OpPrereqError("Cannot change the secondary ip on a single"
5038
                                   " homed cluster", errors.ECODE_INVAL)
5039

    
5040
      if node.offline:
5041
        if self.affected_instances:
5042
          raise errors.OpPrereqError("Cannot change secondary ip: offline"
5043
                                     " node has instances (%s) configured"
5044
                                     " to use it" % self.affected_instances)
5045
      else:
5046
        # On online nodes, check that no instances are running, and that
5047
        # the node has the new ip and we can reach it.
5048
        for instance in self.affected_instances:
5049
          _CheckInstanceDown(self, instance, "cannot change secondary ip")
5050

    
5051
        _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
5052
        if master.name != node.name:
5053
          # check reachability from master secondary ip to new secondary ip
5054
          if not netutils.TcpPing(self.op.secondary_ip,
5055
                                  constants.DEFAULT_NODED_PORT,
5056
                                  source=master.secondary_ip):
5057
            raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
5058
                                       " based ping to node daemon port",
5059
                                       errors.ECODE_ENVIRON)
5060

    
5061
    if self.op.ndparams:
5062
      new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
5063
      utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
5064
      self.new_ndparams = new_ndparams
5065

    
5066
  def Exec(self, feedback_fn):
5067
    """Modifies a node.
5068

5069
    """
5070
    node = self.node
5071
    old_role = self.old_role
5072
    new_role = self.new_role
5073

    
5074
    result = []
5075

    
5076
    if self.op.ndparams:
5077
      node.ndparams = self.new_ndparams
5078

    
5079
    if self.op.powered is not None:
5080
      node.powered = self.op.powered
5081

    
5082
    for attr in ["master_capable", "vm_capable"]:
5083
      val = getattr(self.op, attr)
5084
      if val is not None:
5085
        setattr(node, attr, val)
5086
        result.append((attr, str(val)))
5087

    
5088
    if new_role != old_role:
5089
      # Tell the node to demote itself, if no longer MC and not offline
5090
      if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
5091
        msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
5092
        if msg:
5093
          self.LogWarning("Node failed to demote itself: %s", msg)
5094

    
5095
      new_flags = self._R2F[new_role]
5096
      for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
5097
        if of != nf:
5098
          result.append((desc, str(nf)))
5099
      (node.master_candidate, node.drained, node.offline) = new_flags
5100

    
5101
      # we locked all nodes, we adjust the CP before updating this node
5102
      if self.lock_all:
5103
        _AdjustCandidatePool(self, [node.name])
5104

    
5105
    if self.op.secondary_ip:
5106
      node.secondary_ip = self.op.secondary_ip
5107
      result.append(("secondary_ip", self.op.secondary_ip))
5108

    
5109
    # this will trigger configuration file update, if needed
5110
    self.cfg.Update(node, feedback_fn)
5111

    
5112
    # this will trigger job queue propagation or cleanup if the mc
5113
    # flag changed
5114
    if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
5115
      self.context.ReaddNode(node)
5116

    
5117
    return result
5118

    
5119

    
5120
class LUNodePowercycle(NoHooksLU):
5121
  """Powercycles a node.
5122

5123
  """
5124
  REQ_BGL = False
5125

    
5126
  def CheckArguments(self):
5127
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5128
    if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
5129
      raise errors.OpPrereqError("The node is the master and the force"
5130
                                 " parameter was not set",
5131
                                 errors.ECODE_INVAL)
5132

    
5133
  def ExpandNames(self):
5134
    """Locking for PowercycleNode.
5135

5136
    This is a last-resort option and shouldn't block on other
5137
    jobs. Therefore, we grab no locks.
5138

5139
    """
5140
    self.needed_locks = {}
5141

    
5142
  def Exec(self, feedback_fn):
5143
    """Reboots a node.
5144

5145
    """
5146
    result = self.rpc.call_node_powercycle(self.op.node_name,
5147
                                           self.cfg.GetHypervisorType())
5148
    result.Raise("Failed to schedule the reboot")
5149
    return result.payload
5150

    
5151

    
5152
class LUClusterQuery(NoHooksLU):
5153
  """Query cluster configuration.
5154

5155
  """
5156
  REQ_BGL = False
5157

    
5158
  def ExpandNames(self):
5159
    self.needed_locks = {}
5160

    
5161
  def Exec(self, feedback_fn):
5162
    """Return cluster config.
5163

5164
    """
5165
    cluster = self.cfg.GetClusterInfo()
5166
    os_hvp = {}
5167

    
5168
    # Filter just for enabled hypervisors
5169
    for os_name, hv_dict in cluster.os_hvp.items():
5170
      os_hvp[os_name] = {}
5171
      for hv_name, hv_params in hv_dict.items():
5172
        if hv_name in cluster.enabled_hypervisors:
5173
          os_hvp[os_name][hv_name] = hv_params
5174

    
5175
    # Convert ip_family to ip_version
5176
    primary_ip_version = constants.IP4_VERSION
5177
    if cluster.primary_ip_family == netutils.IP6Address.family:
5178
      primary_ip_version = constants.IP6_VERSION
5179

    
5180
    result = {
5181
      "software_version": constants.RELEASE_VERSION,
5182
      "protocol_version": constants.PROTOCOL_VERSION,
5183
      "config_version": constants.CONFIG_VERSION,
5184
      "os_api_version": max(constants.OS_API_VERSIONS),
5185
      "export_version": constants.EXPORT_VERSION,
5186
      "architecture": (platform.architecture()[0], platform.machine()),
5187
      "name": cluster.cluster_name,
5188
      "master": cluster.master_node,
5189
      "default_hypervisor": cluster.enabled_hypervisors[0],
5190
      "enabled_hypervisors": cluster.enabled_hypervisors,
5191
      "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
5192
                        for hypervisor_name in cluster.enabled_hypervisors]),
5193
      "os_hvp": os_hvp,
5194
      "beparams": cluster.beparams,
5195
      "osparams": cluster.osparams,
5196
      "nicparams": cluster.nicparams,
5197
      "ndparams": cluster.ndparams,
5198
      "candidate_pool_size": cluster.candidate_pool_size,
5199
      "master_netdev": cluster.master_netdev,
5200
      "volume_group_name": cluster.volume_group_name,
5201
      "drbd_usermode_helper": cluster.drbd_usermode_helper,
5202
      "file_storage_dir": cluster.file_storage_dir,
5203
      "shared_file_storage_dir": cluster.shared_file_storage_dir,
5204
      "maintain_node_health": cluster.maintain_node_health,
5205
      "ctime": cluster.ctime,
5206
      "mtime": cluster.mtime,
5207
      "uuid": cluster.uuid,
5208
      "tags": list(cluster.GetTags()),
5209
      "uid_pool": cluster.uid_pool,
5210
      "default_iallocator": cluster.default_iallocator,
5211
      "reserved_lvs": cluster.reserved_lvs,
5212
      "primary_ip_version": primary_ip_version,
5213
      "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
5214
      "hidden_os": cluster.hidden_os,
5215
      "blacklisted_os": cluster.blacklisted_os,
5216
      }
5217

    
5218
    return result
5219

    
5220

    
5221
class LUClusterConfigQuery(NoHooksLU):
5222
  """Return configuration values.
5223

5224
  """
5225
  REQ_BGL = False
5226
  _FIELDS_DYNAMIC = utils.FieldSet()
5227
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag",
5228
                                  "watcher_pause", "volume_group_name")
5229

    
5230
  def CheckArguments(self):
5231
    _CheckOutputFields(static=self._FIELDS_STATIC,
5232
                       dynamic=self._FIELDS_DYNAMIC,
5233
                       selected=self.op.output_fields)
5234

    
5235
  def ExpandNames(self):
5236
    self.needed_locks = {}
5237

    
5238
  def Exec(self, feedback_fn):
5239
    """Dump a representation of the cluster config to the standard output.
5240

5241
    """
5242
    values = []
5243
    for field in self.op.output_fields:
5244
      if field == "cluster_name":
5245
        entry = self.cfg.GetClusterName()
5246
      elif field == "master_node":
5247
        entry = self.cfg.GetMasterNode()
5248
      elif field == "drain_flag":
5249
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
5250
      elif field == "watcher_pause":
5251
        entry = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
5252
      elif field == "volume_group_name":
5253
        entry = self.cfg.GetVGName()
5254
      else:
5255
        raise errors.ParameterError(field)
5256
      values.append(entry)
5257
    return values
5258

    
5259

    
5260
class LUInstanceActivateDisks(NoHooksLU):
5261
  """Bring up an instance's disks.
5262

5263
  """
5264
  REQ_BGL = False
5265

    
5266
  def ExpandNames(self):
5267
    self._ExpandAndLockInstance()
5268
    self.needed_locks[locking.LEVEL_NODE] = []
5269
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5270

    
5271
  def DeclareLocks(self, level):
5272
    if level == locking.LEVEL_NODE:
5273
      self._LockInstancesNodes()
5274

    
5275
  def CheckPrereq(self):
5276
    """Check prerequisites.
5277

5278
    This checks that the instance is in the cluster.
5279

5280
    """
5281
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5282
    assert self.instance is not None, \
5283
      "Cannot retrieve locked instance %s" % self.op.instance_name
5284
    _CheckNodeOnline(self, self.instance.primary_node)
5285

    
5286
  def Exec(self, feedback_fn):
5287
    """Activate the disks.
5288

5289
    """
5290
    disks_ok, disks_info = \
5291
              _AssembleInstanceDisks(self, self.instance,
5292
                                     ignore_size=self.op.ignore_size)
5293
    if not disks_ok:
5294
      raise errors.OpExecError("Cannot activate block devices")
5295

    
5296
    return disks_info
5297

    
5298

    
5299
def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
5300
                           ignore_size=False):
5301
  """Prepare the block devices for an instance.
5302

5303
  This sets up the block devices on all nodes.
5304

5305
  @type lu: L{LogicalUnit}
5306
  @param lu: the logical unit on whose behalf we execute
5307
  @type instance: L{objects.Instance}
5308
  @param instance: the instance for whose disks we assemble
5309
  @type disks: list of L{objects.Disk} or None
5310
  @param disks: which disks to assemble (or all, if None)
5311
  @type ignore_secondaries: boolean
5312
  @param ignore_secondaries: if true, errors on secondary nodes
5313
      won't result in an error return from the function
5314
  @type ignore_size: boolean
5315
  @param ignore_size: if true, the current known size of the disk
5316
      will not be used during the disk activation, useful for cases
5317
      when the size is wrong
5318
  @return: False if the operation failed, otherwise a list of
5319
      (host, instance_visible_name, node_visible_name)
5320
      with the mapping from node devices to instance devices
5321

5322
  """
5323
  device_info = []
5324
  disks_ok = True
5325
  iname = instance.name
5326
  disks = _ExpandCheckDisks(instance, disks)
5327

    
5328
  # With the two passes mechanism we try to reduce the window of
5329
  # opportunity for the race condition of switching DRBD to primary
5330
  # before handshaking occured, but we do not eliminate it
5331

    
5332
  # The proper fix would be to wait (with some limits) until the
5333
  # connection has been made and drbd transitions from WFConnection
5334
  # into any other network-connected state (Connected, SyncTarget,
5335
  # SyncSource, etc.)
5336

    
5337
  # 1st pass, assemble on all nodes in secondary mode
5338
  for idx, inst_disk in enumerate(disks):
5339
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
5340
      if ignore_size:
5341
        node_disk = node_disk.Copy()
5342
        node_disk.UnsetSize()
5343
      lu.cfg.SetDiskID(node_disk, node)
5344
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False, idx)
5345
      msg = result.fail_msg
5346
      if msg:
5347
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
5348
                           " (is_primary=False, pass=1): %s",
5349
                           inst_disk.iv_name, node, msg)
5350
        if not ignore_secondaries:
5351
          disks_ok = False
5352

    
5353
  # FIXME: race condition on drbd migration to primary
5354

    
5355
  # 2nd pass, do only the primary node
5356
  for idx, inst_disk in enumerate(disks):
5357
    dev_path = None
5358

    
5359
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
5360
      if node != instance.primary_node:
5361
        continue
5362
      if ignore_size:
5363
        node_disk = node_disk.Copy()
5364
        node_disk.UnsetSize()
5365
      lu.cfg.SetDiskID(node_disk, node)
5366
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True, idx)
5367
      msg = result.fail_msg
5368
      if msg:
5369
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
5370
                           " (is_primary=True, pass=2): %s",
5371
                           inst_disk.iv_name, node, msg)
5372
        disks_ok = False
5373
      else:
5374
        dev_path = result.payload
5375

    
5376
    device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
5377

    
5378
  # leave the disks configured for the primary node
5379
  # this is a workaround that would be fixed better by
5380
  # improving the logical/physical id handling
5381
  for disk in disks:
5382
    lu.cfg.SetDiskID(disk, instance.primary_node)
5383

    
5384
  return disks_ok, device_info
5385

    
5386

    
5387
def _StartInstanceDisks(lu, instance, force):
5388
  """Start the disks of an instance.
5389

5390
  """
5391
  disks_ok, _ = _AssembleInstanceDisks(lu, instance,
5392
                                           ignore_secondaries=force)
5393
  if not disks_ok:
5394
    _ShutdownInstanceDisks(lu, instance)
5395
    if force is not None and not force:
5396
      lu.proc.LogWarning("", hint="If the message above refers to a"
5397
                         " secondary node,"
5398
                         " you can retry the operation using '--force'.")
5399
    raise errors.OpExecError("Disk consistency error")
5400

    
5401

    
5402
class LUInstanceDeactivateDisks(NoHooksLU):
5403
  """Shutdown an instance's disks.
5404

5405
  """
5406
  REQ_BGL = False
5407

    
5408
  def ExpandNames(self):
5409
    self._ExpandAndLockInstance()
5410
    self.needed_locks[locking.LEVEL_NODE] = []
5411
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5412

    
5413
  def DeclareLocks(self, level):
5414
    if level == locking.LEVEL_NODE:
5415
      self._LockInstancesNodes()
5416

    
5417
  def CheckPrereq(self):
5418
    """Check prerequisites.
5419

5420
    This checks that the instance is in the cluster.
5421

5422
    """
5423
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5424
    assert self.instance is not None, \
5425
      "Cannot retrieve locked instance %s" % self.op.instance_name
5426

    
5427
  def Exec(self, feedback_fn):
5428
    """Deactivate the disks
5429

5430
    """
5431
    instance = self.instance
5432
    if self.op.force:
5433
      _ShutdownInstanceDisks(self, instance)
5434
    else:
5435
      _SafeShutdownInstanceDisks(self, instance)
5436

    
5437

    
5438
def _SafeShutdownInstanceDisks(lu, instance, disks=None):
5439
  """Shutdown block devices of an instance.
5440

5441
  This function checks if an instance is running, before calling
5442
  _ShutdownInstanceDisks.
5443

5444
  """
5445
  _CheckInstanceDown(lu, instance, "cannot shutdown disks")
5446
  _ShutdownInstanceDisks(lu, instance, disks=disks)
5447

    
5448

    
5449
def _ExpandCheckDisks(instance, disks):
5450
  """Return the instance disks selected by the disks list
5451

5452
  @type disks: list of L{objects.Disk} or None
5453
  @param disks: selected disks
5454
  @rtype: list of L{objects.Disk}
5455
  @return: selected instance disks to act on
5456

5457
  """
5458
  if disks is None:
5459
    return instance.disks
5460
  else:
5461
    if not set(disks).issubset(instance.disks):
5462
      raise errors.ProgrammerError("Can only act on disks belonging to the"
5463
                                   " target instance")
5464
    return disks
5465

    
5466

    
5467
def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
5468
  """Shutdown block devices of an instance.
5469

5470
  This does the shutdown on all nodes of the instance.
5471

5472
  If the ignore_primary is false, errors on the primary node are
5473
  ignored.
5474

5475
  """
5476
  all_result = True
5477
  disks = _ExpandCheckDisks(instance, disks)
5478

    
5479
  for disk in disks:
5480
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
5481
      lu.cfg.SetDiskID(top_disk, node)
5482
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
5483
      msg = result.fail_msg
5484
      if msg:
5485
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
5486
                      disk.iv_name, node, msg)
5487
        if ((node == instance.primary_node and not ignore_primary) or
5488
            (node != instance.primary_node and not result.offline)):
5489
          all_result = False
5490
  return all_result
5491

    
5492

    
5493
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
5494
  """Checks if a node has enough free memory.
5495

5496
  This function check if a given node has the needed amount of free
5497
  memory. In case the node has less memory or we cannot get the
5498
  information from the node, this function raise an OpPrereqError
5499
  exception.
5500

5501
  @type lu: C{LogicalUnit}
5502
  @param lu: a logical unit from which we get configuration data
5503
  @type node: C{str}
5504
  @param node: the node to check
5505
  @type reason: C{str}
5506
  @param reason: string to use in the error message
5507
  @type requested: C{int}
5508
  @param requested: the amount of memory in MiB to check for
5509
  @type hypervisor_name: C{str}
5510
  @param hypervisor_name: the hypervisor to ask for memory stats
5511
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
5512
      we cannot check the node
5513

5514
  """
5515
  nodeinfo = lu.rpc.call_node_info([node], None, hypervisor_name)
5516
  nodeinfo[node].Raise("Can't get data from node %s" % node,
5517
                       prereq=True, ecode=errors.ECODE_ENVIRON)
5518
  free_mem = nodeinfo[node].payload.get('memory_free', None)
5519
  if not isinstance(free_mem, int):
5520
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
5521
                               " was '%s'" % (node, free_mem),
5522
                               errors.ECODE_ENVIRON)
5523
  if requested > free_mem:
5524
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
5525
                               " needed %s MiB, available %s MiB" %
5526
                               (node, reason, requested, free_mem),
5527
                               errors.ECODE_NORES)
5528

    
5529

    
5530
def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
5531
  """Checks if nodes have enough free disk space in the all VGs.
5532

5533
  This function check if all given nodes have the needed amount of
5534
  free disk. In case any node has less disk or we cannot get the
5535
  information from the node, this function raise an OpPrereqError
5536
  exception.
5537

5538
  @type lu: C{LogicalUnit}
5539
  @param lu: a logical unit from which we get configuration data
5540
  @type nodenames: C{list}
5541
  @param nodenames: the list of node names to check
5542
  @type req_sizes: C{dict}
5543
  @param req_sizes: the hash of vg and corresponding amount of disk in
5544
      MiB to check for
5545
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5546
      or we cannot check the node
5547

5548
  """
5549
  for vg, req_size in req_sizes.items():
5550
    _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
5551

    
5552

    
5553
def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
5554
  """Checks if nodes have enough free disk space in the specified VG.
5555

5556
  This function check if all given nodes have the needed amount of
5557
  free disk. In case any node has less disk or we cannot get the
5558
  information from the node, this function raise an OpPrereqError
5559
  exception.
5560

5561
  @type lu: C{LogicalUnit}
5562
  @param lu: a logical unit from which we get configuration data
5563
  @type nodenames: C{list}
5564
  @param nodenames: the list of node names to check
5565
  @type vg: C{str}
5566
  @param vg: the volume group to check
5567
  @type requested: C{int}
5568
  @param requested: the amount of disk in MiB to check for
5569
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5570
      or we cannot check the node
5571

5572
  """
5573
  nodeinfo = lu.rpc.call_node_info(nodenames, vg, None)
5574
  for node in nodenames:
5575
    info = nodeinfo[node]
5576
    info.Raise("Cannot get current information from node %s" % node,
5577
               prereq=True, ecode=errors.ECODE_ENVIRON)
5578
    vg_free = info.payload.get("vg_free", None)
5579
    if not isinstance(vg_free, int):
5580
      raise errors.OpPrereqError("Can't compute free disk space on node"
5581
                                 " %s for vg %s, result was '%s'" %
5582
                                 (node, vg, vg_free), errors.ECODE_ENVIRON)
5583
    if requested > vg_free:
5584
      raise errors.OpPrereqError("Not enough disk space on target node %s"
5585
                                 " vg %s: required %d MiB, available %d MiB" %
5586
                                 (node, vg, requested, vg_free),
5587
                                 errors.ECODE_NORES)
5588

    
5589

    
5590
class LUInstanceStartup(LogicalUnit):
5591
  """Starts an instance.
5592

5593
  """
5594
  HPATH = "instance-start"
5595
  HTYPE = constants.HTYPE_INSTANCE
5596
  REQ_BGL = False
5597

    
5598
  def CheckArguments(self):
5599
    # extra beparams
5600
    if self.op.beparams:
5601
      # fill the beparams dict
5602
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
5603

    
5604
  def ExpandNames(self):
5605
    self._ExpandAndLockInstance()
5606

    
5607
  def BuildHooksEnv(self):
5608
    """Build hooks env.
5609

5610
    This runs on master, primary and secondary nodes of the instance.
5611

5612
    """
5613
    env = {
5614
      "FORCE": self.op.force,
5615
      }
5616

    
5617
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5618

    
5619
    return env
5620

    
5621
  def BuildHooksNodes(self):
5622
    """Build hooks nodes.
5623

5624
    """
5625
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5626
    return (nl, nl)
5627

    
5628
  def CheckPrereq(self):
5629
    """Check prerequisites.
5630

5631
    This checks that the instance is in the cluster.
5632

5633
    """
5634
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5635
    assert self.instance is not None, \
5636
      "Cannot retrieve locked instance %s" % self.op.instance_name
5637

    
5638
    # extra hvparams
5639
    if self.op.hvparams:
5640
      # check hypervisor parameter syntax (locally)
5641
      cluster = self.cfg.GetClusterInfo()
5642
      utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
5643
      filled_hvp = cluster.FillHV(instance)
5644
      filled_hvp.update(self.op.hvparams)
5645
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
5646
      hv_type.CheckParameterSyntax(filled_hvp)
5647
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
5648

    
5649
    self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
5650

    
5651
    if self.primary_offline and self.op.ignore_offline_nodes:
5652
      self.proc.LogWarning("Ignoring offline primary node")
5653

    
5654
      if self.op.hvparams or self.op.beparams:
5655
        self.proc.LogWarning("Overridden parameters are ignored")
5656
    else:
5657
      _CheckNodeOnline(self, instance.primary_node)
5658

    
5659
      bep = self.cfg.GetClusterInfo().FillBE(instance)
5660

    
5661
      # check bridges existence
5662
      _CheckInstanceBridgesExist(self, instance)
5663

    
5664
      remote_info = self.rpc.call_instance_info(instance.primary_node,
5665
                                                instance.name,
5666
                                                instance.hypervisor)
5667
      remote_info.Raise("Error checking node %s" % instance.primary_node,
5668
                        prereq=True, ecode=errors.ECODE_ENVIRON)
5669
      if not remote_info.payload: # not running already
5670
        _CheckNodeFreeMemory(self, instance.primary_node,
5671
                             "starting instance %s" % instance.name,
5672
                             bep[constants.BE_MEMORY], instance.hypervisor)
5673

    
5674
  def Exec(self, feedback_fn):
5675
    """Start the instance.
5676

5677
    """
5678
    instance = self.instance
5679
    force = self.op.force
5680

    
5681
    if not self.op.no_remember:
5682
      self.cfg.MarkInstanceUp(instance.name)
5683

    
5684
    if self.primary_offline:
5685
      assert self.op.ignore_offline_nodes
5686
      self.proc.LogInfo("Primary node offline, marked instance as started")
5687
    else:
5688
      node_current = instance.primary_node
5689

    
5690
      _StartInstanceDisks(self, instance, force)
5691

    
5692
      result = self.rpc.call_instance_start(node_current, instance,
5693
                                            self.op.hvparams, self.op.beparams)
5694
      msg = result.fail_msg
5695
      if msg:
5696
        _ShutdownInstanceDisks(self, instance)
5697
        raise errors.OpExecError("Could not start instance: %s" % msg)
5698

    
5699

    
5700
class LUInstanceReboot(LogicalUnit):
5701
  """Reboot an instance.
5702

5703
  """
5704
  HPATH = "instance-reboot"
5705
  HTYPE = constants.HTYPE_INSTANCE
5706
  REQ_BGL = False
5707

    
5708
  def ExpandNames(self):
5709
    self._ExpandAndLockInstance()
5710

    
5711
  def BuildHooksEnv(self):
5712
    """Build hooks env.
5713

5714
    This runs on master, primary and secondary nodes of the instance.
5715

5716
    """
5717
    env = {
5718
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
5719
      "REBOOT_TYPE": self.op.reboot_type,
5720
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5721
      }
5722

    
5723
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5724

    
5725
    return env
5726

    
5727
  def BuildHooksNodes(self):
5728
    """Build hooks nodes.
5729

5730
    """
5731
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5732
    return (nl, nl)
5733

    
5734
  def CheckPrereq(self):
5735
    """Check prerequisites.
5736

5737
    This checks that the instance is in the cluster.
5738

5739
    """
5740
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5741
    assert self.instance is not None, \
5742
      "Cannot retrieve locked instance %s" % self.op.instance_name
5743

    
5744
    _CheckNodeOnline(self, instance.primary_node)
5745

    
5746
    # check bridges existence
5747
    _CheckInstanceBridgesExist(self, instance)
5748

    
5749
  def Exec(self, feedback_fn):
5750
    """Reboot the instance.
5751

5752
    """
5753
    instance = self.instance
5754
    ignore_secondaries = self.op.ignore_secondaries
5755
    reboot_type = self.op.reboot_type
5756

    
5757
    remote_info = self.rpc.call_instance_info(instance.primary_node,
5758
                                              instance.name,
5759
                                              instance.hypervisor)
5760
    remote_info.Raise("Error checking node %s" % instance.primary_node)
5761
    instance_running = bool(remote_info.payload)
5762

    
5763
    node_current = instance.primary_node
5764

    
5765
    if instance_running and reboot_type in [constants.INSTANCE_REBOOT_SOFT,
5766
                                            constants.INSTANCE_REBOOT_HARD]:
5767
      for disk in instance.disks:
5768
        self.cfg.SetDiskID(disk, node_current)
5769
      result = self.rpc.call_instance_reboot(node_current, instance,
5770
                                             reboot_type,
5771
                                             self.op.shutdown_timeout)
5772
      result.Raise("Could not reboot instance")
5773
    else:
5774
      if instance_running:
5775
        result = self.rpc.call_instance_shutdown(node_current, instance,
5776
                                                 self.op.shutdown_timeout)
5777
        result.Raise("Could not shutdown instance for full reboot")
5778
        _ShutdownInstanceDisks(self, instance)
5779
      else:
5780
        self.LogInfo("Instance %s was already stopped, starting now",
5781
                     instance.name)
5782
      _StartInstanceDisks(self, instance, ignore_secondaries)
5783
      result = self.rpc.call_instance_start(node_current, instance, None, None)
5784
      msg = result.fail_msg
5785
      if msg:
5786
        _ShutdownInstanceDisks(self, instance)
5787
        raise errors.OpExecError("Could not start instance for"
5788
                                 " full reboot: %s" % msg)
5789

    
5790
    self.cfg.MarkInstanceUp(instance.name)
5791

    
5792

    
5793
class LUInstanceShutdown(LogicalUnit):
5794
  """Shutdown an instance.
5795

5796
  """
5797
  HPATH = "instance-stop"
5798
  HTYPE = constants.HTYPE_INSTANCE
5799
  REQ_BGL = False
5800

    
5801
  def ExpandNames(self):
5802
    self._ExpandAndLockInstance()
5803

    
5804
  def BuildHooksEnv(self):
5805
    """Build hooks env.
5806

5807
    This runs on master, primary and secondary nodes of the instance.
5808

5809
    """
5810
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5811
    env["TIMEOUT"] = self.op.timeout
5812
    return env
5813

    
5814
  def BuildHooksNodes(self):
5815
    """Build hooks nodes.
5816

5817
    """
5818
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5819
    return (nl, nl)
5820

    
5821
  def CheckPrereq(self):
5822
    """Check prerequisites.
5823

5824
    This checks that the instance is in the cluster.
5825

5826
    """
5827
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5828
    assert self.instance is not None, \
5829
      "Cannot retrieve locked instance %s" % self.op.instance_name
5830

    
5831
    self.primary_offline = \
5832
      self.cfg.GetNodeInfo(self.instance.primary_node).offline
5833

    
5834
    if self.primary_offline and self.op.ignore_offline_nodes:
5835
      self.proc.LogWarning("Ignoring offline primary node")
5836
    else:
5837
      _CheckNodeOnline(self, self.instance.primary_node)
5838

    
5839
  def Exec(self, feedback_fn):
5840
    """Shutdown the instance.
5841

5842
    """
5843
    instance = self.instance
5844
    node_current = instance.primary_node
5845
    timeout = self.op.timeout
5846

    
5847
    if not self.op.no_remember:
5848
      self.cfg.MarkInstanceDown(instance.name)
5849

    
5850
    if self.primary_offline:
5851
      assert self.op.ignore_offline_nodes
5852
      self.proc.LogInfo("Primary node offline, marked instance as stopped")
5853
    else:
5854
      result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
5855
      msg = result.fail_msg
5856
      if msg:
5857
        self.proc.LogWarning("Could not shutdown instance: %s" % msg)
5858

    
5859
      _ShutdownInstanceDisks(self, instance)
5860

    
5861

    
5862
class LUInstanceReinstall(LogicalUnit):
5863
  """Reinstall an instance.
5864

5865
  """
5866
  HPATH = "instance-reinstall"
5867
  HTYPE = constants.HTYPE_INSTANCE
5868
  REQ_BGL = False
5869

    
5870
  def ExpandNames(self):
5871
    self._ExpandAndLockInstance()
5872

    
5873
  def BuildHooksEnv(self):
5874
    """Build hooks env.
5875

5876
    This runs on master, primary and secondary nodes of the instance.
5877

5878
    """
5879
    return _BuildInstanceHookEnvByObject(self, self.instance)
5880

    
5881
  def BuildHooksNodes(self):
5882
    """Build hooks nodes.
5883

5884
    """
5885
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5886
    return (nl, nl)
5887

    
5888
  def CheckPrereq(self):
5889
    """Check prerequisites.
5890

5891
    This checks that the instance is in the cluster and is not running.
5892

5893
    """
5894
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5895
    assert instance is not None, \
5896
      "Cannot retrieve locked instance %s" % self.op.instance_name
5897
    _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
5898
                     " offline, cannot reinstall")
5899
    for node in instance.secondary_nodes:
5900
      _CheckNodeOnline(self, node, "Instance secondary node offline,"
5901
                       " cannot reinstall")
5902

    
5903
    if instance.disk_template == constants.DT_DISKLESS:
5904
      raise errors.OpPrereqError("Instance '%s' has no disks" %
5905
                                 self.op.instance_name,
5906
                                 errors.ECODE_INVAL)
5907
    _CheckInstanceDown(self, instance, "cannot reinstall")
5908

    
5909
    if self.op.os_type is not None:
5910
      # OS verification
5911
      pnode = _ExpandNodeName(self.cfg, instance.primary_node)
5912
      _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
5913
      instance_os = self.op.os_type
5914
    else:
5915
      instance_os = instance.os
5916

    
5917
    nodelist = list(instance.all_nodes)
5918

    
5919
    if self.op.osparams:
5920
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
5921
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
5922
      self.os_inst = i_osdict # the new dict (without defaults)
5923
    else:
5924
      self.os_inst = None
5925

    
5926
    self.instance = instance
5927

    
5928
  def Exec(self, feedback_fn):
5929
    """Reinstall the instance.
5930

5931
    """
5932
    inst = self.instance
5933

    
5934
    if self.op.os_type is not None:
5935
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
5936
      inst.os = self.op.os_type
5937
      # Write to configuration
5938
      self.cfg.Update(inst, feedback_fn)
5939

    
5940
    _StartInstanceDisks(self, inst, None)
5941
    try:
5942
      feedback_fn("Running the instance OS create scripts...")
5943
      # FIXME: pass debug option from opcode to backend
5944
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True,
5945
                                             self.op.debug_level,
5946
                                             osparams=self.os_inst)
5947
      result.Raise("Could not install OS for instance %s on node %s" %
5948
                   (inst.name, inst.primary_node))
5949
    finally:
5950
      _ShutdownInstanceDisks(self, inst)
5951

    
5952

    
5953
class LUInstanceRecreateDisks(LogicalUnit):
5954
  """Recreate an instance's missing disks.
5955

5956
  """
5957
  HPATH = "instance-recreate-disks"
5958
  HTYPE = constants.HTYPE_INSTANCE
5959
  REQ_BGL = False
5960

    
5961
  def CheckArguments(self):
5962
    # normalise the disk list
5963
    self.op.disks = sorted(frozenset(self.op.disks))
5964

    
5965
  def ExpandNames(self):
5966
    self._ExpandAndLockInstance()
5967
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5968
    if self.op.nodes:
5969
      self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
5970
      self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
5971
    else:
5972
      self.needed_locks[locking.LEVEL_NODE] = []
5973

    
5974
  def DeclareLocks(self, level):
5975
    if level == locking.LEVEL_NODE:
5976
      # if we replace the nodes, we only need to lock the old primary,
5977
      # otherwise we need to lock all nodes for disk re-creation
5978
      primary_only = bool(self.op.nodes)
5979
      self._LockInstancesNodes(primary_only=primary_only)
5980

    
5981
  def BuildHooksEnv(self):
5982
    """Build hooks env.
5983

5984
    This runs on master, primary and secondary nodes of the instance.
5985

5986
    """
5987
    return _BuildInstanceHookEnvByObject(self, self.instance)
5988

    
5989
  def BuildHooksNodes(self):
5990
    """Build hooks nodes.
5991

5992
    """
5993
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5994
    return (nl, nl)
5995

    
5996
  def CheckPrereq(self):
5997
    """Check prerequisites.
5998

5999
    This checks that the instance is in the cluster and is not running.
6000

6001
    """
6002
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6003
    assert instance is not None, \
6004
      "Cannot retrieve locked instance %s" % self.op.instance_name
6005
    if self.op.nodes:
6006
      if len(self.op.nodes) != len(instance.all_nodes):
6007
        raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
6008
                                   " %d replacement nodes were specified" %
6009
                                   (instance.name, len(instance.all_nodes),
6010
                                    len(self.op.nodes)),
6011
                                   errors.ECODE_INVAL)
6012
      assert instance.disk_template != constants.DT_DRBD8 or \
6013
          len(self.op.nodes) == 2
6014
      assert instance.disk_template != constants.DT_PLAIN or \
6015
          len(self.op.nodes) == 1
6016
      primary_node = self.op.nodes[0]
6017
    else:
6018
      primary_node = instance.primary_node
6019
    _CheckNodeOnline(self, primary_node)
6020

    
6021
    if instance.disk_template == constants.DT_DISKLESS:
6022
      raise errors.OpPrereqError("Instance '%s' has no disks" %
6023
                                 self.op.instance_name, errors.ECODE_INVAL)
6024
    # if we replace nodes *and* the old primary is offline, we don't
6025
    # check
6026
    assert instance.primary_node in self.needed_locks[locking.LEVEL_NODE]
6027
    old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
6028
    if not (self.op.nodes and old_pnode.offline):
6029
      _CheckInstanceDown(self, instance, "cannot recreate disks")
6030

    
6031
    if not self.op.disks:
6032
      self.op.disks = range(len(instance.disks))
6033
    else:
6034
      for idx in self.op.disks:
6035
        if idx >= len(instance.disks):
6036
          raise errors.OpPrereqError("Invalid disk index '%s'" % idx,
6037
                                     errors.ECODE_INVAL)
6038
    if self.op.disks != range(len(instance.disks)) and self.op.nodes:
6039
      raise errors.OpPrereqError("Can't recreate disks partially and"
6040
                                 " change the nodes at the same time",
6041
                                 errors.ECODE_INVAL)
6042
    self.instance = instance
6043

    
6044
  def Exec(self, feedback_fn):
6045
    """Recreate the disks.
6046

6047
    """
6048
    # change primary node, if needed
6049
    if self.op.nodes:
6050
      self.instance.primary_node = self.op.nodes[0]
6051
      self.LogWarning("Changing the instance's nodes, you will have to"
6052
                      " remove any disks left on the older nodes manually")
6053

    
6054
    to_skip = []
6055
    for idx, disk in enumerate(self.instance.disks):
6056
      if idx not in self.op.disks: # disk idx has not been passed in
6057
        to_skip.append(idx)
6058
        continue
6059
      # update secondaries for disks, if needed
6060
      if self.op.nodes:
6061
        if disk.dev_type == constants.LD_DRBD8:
6062
          # need to update the nodes
6063
          assert len(self.op.nodes) == 2
6064
          logical_id = list(disk.logical_id)
6065
          logical_id[0] = self.op.nodes[0]
6066
          logical_id[1] = self.op.nodes[1]
6067
          disk.logical_id = tuple(logical_id)
6068

    
6069
    if self.op.nodes:
6070
      self.cfg.Update(self.instance, feedback_fn)
6071

    
6072
    _CreateDisks(self, self.instance, to_skip=to_skip)
6073

    
6074

    
6075
class LUInstanceRename(LogicalUnit):
6076
  """Rename an instance.
6077

6078
  """
6079
  HPATH = "instance-rename"
6080
  HTYPE = constants.HTYPE_INSTANCE
6081

    
6082
  def CheckArguments(self):
6083
    """Check arguments.
6084

6085
    """
6086
    if self.op.ip_check and not self.op.name_check:
6087
      # TODO: make the ip check more flexible and not depend on the name check
6088
      raise errors.OpPrereqError("IP address check requires a name check",
6089
                                 errors.ECODE_INVAL)
6090

    
6091
  def BuildHooksEnv(self):
6092
    """Build hooks env.
6093

6094
    This runs on master, primary and secondary nodes of the instance.
6095

6096
    """
6097
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6098
    env["INSTANCE_NEW_NAME"] = self.op.new_name
6099
    return env
6100

    
6101
  def BuildHooksNodes(self):
6102
    """Build hooks nodes.
6103

6104
    """
6105
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6106
    return (nl, nl)
6107

    
6108
  def CheckPrereq(self):
6109
    """Check prerequisites.
6110

6111
    This checks that the instance is in the cluster and is not running.
6112

6113
    """
6114
    self.op.instance_name = _ExpandInstanceName(self.cfg,
6115
                                                self.op.instance_name)
6116
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6117
    assert instance is not None
6118
    _CheckNodeOnline(self, instance.primary_node)
6119
    _CheckInstanceDown(self, instance, "cannot rename")
6120
    self.instance = instance
6121

    
6122
    new_name = self.op.new_name
6123
    if self.op.name_check:
6124
      hostname = netutils.GetHostname(name=new_name)
6125
      if hostname != new_name:
6126
        self.LogInfo("Resolved given name '%s' to '%s'", new_name,
6127
                     hostname.name)
6128
      if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
6129
        raise errors.OpPrereqError(("Resolved hostname '%s' does not look the"
6130
                                    " same as given hostname '%s'") %
6131
                                    (hostname.name, self.op.new_name),
6132
                                    errors.ECODE_INVAL)
6133
      new_name = self.op.new_name = hostname.name
6134
      if (self.op.ip_check and
6135
          netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
6136
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
6137
                                   (hostname.ip, new_name),
6138
                                   errors.ECODE_NOTUNIQUE)
6139

    
6140
    instance_list = self.cfg.GetInstanceList()
6141
    if new_name in instance_list and new_name != instance.name:
6142
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
6143
                                 new_name, errors.ECODE_EXISTS)
6144

    
6145
  def Exec(self, feedback_fn):
6146
    """Rename the instance.
6147

6148
    """
6149
    inst = self.instance
6150
    old_name = inst.name
6151

    
6152
    rename_file_storage = False
6153
    if (inst.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE) and
6154
        self.op.new_name != inst.name):
6155
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
6156
      rename_file_storage = True
6157

    
6158
    self.cfg.RenameInstance(inst.name, self.op.new_name)
6159
    # Change the instance lock. This is definitely safe while we hold the BGL.
6160
    # Otherwise the new lock would have to be added in acquired mode.
6161
    assert self.REQ_BGL
6162
    self.glm.remove(locking.LEVEL_INSTANCE, old_name)
6163
    self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
6164

    
6165
    # re-read the instance from the configuration after rename
6166
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
6167

    
6168
    if rename_file_storage:
6169
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
6170
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
6171
                                                     old_file_storage_dir,
6172
                                                     new_file_storage_dir)
6173
      result.Raise("Could not rename on node %s directory '%s' to '%s'"
6174
                   " (but the instance has been renamed in Ganeti)" %
6175
                   (inst.primary_node, old_file_storage_dir,
6176
                    new_file_storage_dir))
6177

    
6178
    _StartInstanceDisks(self, inst, None)
6179
    try:
6180
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
6181
                                                 old_name, self.op.debug_level)
6182
      msg = result.fail_msg
6183
      if msg:
6184
        msg = ("Could not run OS rename script for instance %s on node %s"
6185
               " (but the instance has been renamed in Ganeti): %s" %
6186
               (inst.name, inst.primary_node, msg))
6187
        self.proc.LogWarning(msg)
6188
    finally:
6189
      _ShutdownInstanceDisks(self, inst)
6190

    
6191
    return inst.name
6192

    
6193

    
6194
class LUInstanceRemove(LogicalUnit):
6195
  """Remove an instance.
6196

6197
  """
6198
  HPATH = "instance-remove"
6199
  HTYPE = constants.HTYPE_INSTANCE
6200
  REQ_BGL = False
6201

    
6202
  def ExpandNames(self):
6203
    self._ExpandAndLockInstance()
6204
    self.needed_locks[locking.LEVEL_NODE] = []
6205
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6206

    
6207
  def DeclareLocks(self, level):
6208
    if level == locking.LEVEL_NODE:
6209
      self._LockInstancesNodes()
6210

    
6211
  def BuildHooksEnv(self):
6212
    """Build hooks env.
6213

6214
    This runs on master, primary and secondary nodes of the instance.
6215

6216
    """
6217
    env = _BuildInstanceHookEnvByObject(self, self.instance)
6218
    env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
6219
    return env
6220

    
6221
  def BuildHooksNodes(self):
6222
    """Build hooks nodes.
6223

6224
    """
6225
    nl = [self.cfg.GetMasterNode()]
6226
    nl_post = list(self.instance.all_nodes) + nl
6227
    return (nl, nl_post)
6228

    
6229
  def CheckPrereq(self):
6230
    """Check prerequisites.
6231

6232
    This checks that the instance is in the cluster.
6233

6234
    """
6235
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6236
    assert self.instance is not None, \
6237
      "Cannot retrieve locked instance %s" % self.op.instance_name
6238

    
6239
  def Exec(self, feedback_fn):
6240
    """Remove the instance.
6241

6242
    """
6243
    instance = self.instance
6244
    logging.info("Shutting down instance %s on node %s",
6245
                 instance.name, instance.primary_node)
6246

    
6247
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
6248
                                             self.op.shutdown_timeout)
6249
    msg = result.fail_msg
6250
    if msg:
6251
      if self.op.ignore_failures:
6252
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
6253
      else:
6254
        raise errors.OpExecError("Could not shutdown instance %s on"
6255
                                 " node %s: %s" %
6256
                                 (instance.name, instance.primary_node, msg))
6257

    
6258
    _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
6259

    
6260

    
6261
def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
6262
  """Utility function to remove an instance.
6263

6264
  """
6265
  logging.info("Removing block devices for instance %s", instance.name)
6266

    
6267
  if not _RemoveDisks(lu, instance):
6268
    if not ignore_failures:
6269
      raise errors.OpExecError("Can't remove instance's disks")
6270
    feedback_fn("Warning: can't remove instance's disks")
6271

    
6272
  logging.info("Removing instance %s out of cluster config", instance.name)
6273

    
6274
  lu.cfg.RemoveInstance(instance.name)
6275

    
6276
  assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
6277
    "Instance lock removal conflict"
6278

    
6279
  # Remove lock for the instance
6280
  lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
6281

    
6282

    
6283
class LUInstanceQuery(NoHooksLU):
6284
  """Logical unit for querying instances.
6285

6286
  """
6287
  # pylint: disable-msg=W0142
6288
  REQ_BGL = False
6289

    
6290
  def CheckArguments(self):
6291
    self.iq = _InstanceQuery(qlang.MakeSimpleFilter("name", self.op.names),
6292
                             self.op.output_fields, self.op.use_locking)
6293

    
6294
  def ExpandNames(self):
6295
    self.iq.ExpandNames(self)
6296

    
6297
  def DeclareLocks(self, level):
6298
    self.iq.DeclareLocks(self, level)
6299

    
6300
  def Exec(self, feedback_fn):
6301
    return self.iq.OldStyleQuery(self)
6302

    
6303

    
6304
class LUInstanceFailover(LogicalUnit):
6305
  """Failover an instance.
6306

6307
  """
6308
  HPATH = "instance-failover"
6309
  HTYPE = constants.HTYPE_INSTANCE
6310
  REQ_BGL = False
6311

    
6312
  def CheckArguments(self):
6313
    """Check the arguments.
6314

6315
    """
6316
    self.iallocator = getattr(self.op, "iallocator", None)
6317
    self.target_node = getattr(self.op, "target_node", None)
6318

    
6319
  def ExpandNames(self):
6320
    self._ExpandAndLockInstance()
6321

    
6322
    if self.op.target_node is not None:
6323
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6324

    
6325
    self.needed_locks[locking.LEVEL_NODE] = []
6326
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6327

    
6328
    ignore_consistency = self.op.ignore_consistency
6329
    shutdown_timeout = self.op.shutdown_timeout
6330
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
6331
                                       cleanup=False,
6332
                                       failover=True,
6333
                                       ignore_consistency=ignore_consistency,
6334
                                       shutdown_timeout=shutdown_timeout)
6335
    self.tasklets = [self._migrater]
6336

    
6337
  def DeclareLocks(self, level):
6338
    if level == locking.LEVEL_NODE:
6339
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
6340
      if instance.disk_template in constants.DTS_EXT_MIRROR:
6341
        if self.op.target_node is None:
6342
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6343
        else:
6344
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
6345
                                                   self.op.target_node]
6346
        del self.recalculate_locks[locking.LEVEL_NODE]
6347
      else:
6348
        self._LockInstancesNodes()
6349

    
6350
  def BuildHooksEnv(self):
6351
    """Build hooks env.
6352

6353
    This runs on master, primary and secondary nodes of the instance.
6354

6355
    """
6356
    instance = self._migrater.instance
6357
    source_node = instance.primary_node
6358
    target_node = self.op.target_node
6359
    env = {
6360
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
6361
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6362
      "OLD_PRIMARY": source_node,
6363
      "NEW_PRIMARY": target_node,
6364
      }
6365

    
6366
    if instance.disk_template in constants.DTS_INT_MIRROR:
6367
      env["OLD_SECONDARY"] = instance.secondary_nodes[0]
6368
      env["NEW_SECONDARY"] = source_node
6369
    else:
6370
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = ""
6371

    
6372
    env.update(_BuildInstanceHookEnvByObject(self, instance))
6373

    
6374
    return env
6375

    
6376
  def BuildHooksNodes(self):
6377
    """Build hooks nodes.
6378

6379
    """
6380
    instance = self._migrater.instance
6381
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
6382
    return (nl, nl + [instance.primary_node])
6383

    
6384

    
6385
class LUInstanceMigrate(LogicalUnit):
6386
  """Migrate an instance.
6387

6388
  This is migration without shutting down, compared to the failover,
6389
  which is done with shutdown.
6390

6391
  """
6392
  HPATH = "instance-migrate"
6393
  HTYPE = constants.HTYPE_INSTANCE
6394
  REQ_BGL = False
6395

    
6396
  def ExpandNames(self):
6397
    self._ExpandAndLockInstance()
6398

    
6399
    if self.op.target_node is not None:
6400
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6401

    
6402
    self.needed_locks[locking.LEVEL_NODE] = []
6403
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6404

    
6405
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
6406
                                       cleanup=self.op.cleanup,
6407
                                       failover=False,
6408
                                       fallback=self.op.allow_failover)
6409
    self.tasklets = [self._migrater]
6410

    
6411
  def DeclareLocks(self, level):
6412
    if level == locking.LEVEL_NODE:
6413
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
6414
      if instance.disk_template in constants.DTS_EXT_MIRROR:
6415
        if self.op.target_node is None:
6416
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6417
        else:
6418
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
6419
                                                   self.op.target_node]
6420
        del self.recalculate_locks[locking.LEVEL_NODE]
6421
      else:
6422
        self._LockInstancesNodes()
6423

    
6424
  def BuildHooksEnv(self):
6425
    """Build hooks env.
6426

6427
    This runs on master, primary and secondary nodes of the instance.
6428

6429
    """
6430
    instance = self._migrater.instance
6431
    source_node = instance.primary_node
6432
    target_node = self.op.target_node
6433
    env = _BuildInstanceHookEnvByObject(self, instance)
6434
    env.update({
6435
      "MIGRATE_LIVE": self._migrater.live,
6436
      "MIGRATE_CLEANUP": self.op.cleanup,
6437
      "OLD_PRIMARY": source_node,
6438
      "NEW_PRIMARY": target_node,
6439
      })
6440

    
6441
    if instance.disk_template in constants.DTS_INT_MIRROR:
6442
      env["OLD_SECONDARY"] = target_node
6443
      env["NEW_SECONDARY"] = source_node
6444
    else:
6445
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
6446

    
6447
    return env
6448

    
6449
  def BuildHooksNodes(self):
6450
    """Build hooks nodes.
6451

6452
    """
6453
    instance = self._migrater.instance
6454
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
6455
    return (nl, nl + [instance.primary_node])
6456

    
6457

    
6458
class LUInstanceMove(LogicalUnit):
6459
  """Move an instance by data-copying.
6460

6461
  """
6462
  HPATH = "instance-move"
6463
  HTYPE = constants.HTYPE_INSTANCE
6464
  REQ_BGL = False
6465

    
6466
  def ExpandNames(self):
6467
    self._ExpandAndLockInstance()
6468
    target_node = _ExpandNodeName(self.cfg, self.op.target_node)
6469
    self.op.target_node = target_node
6470
    self.needed_locks[locking.LEVEL_NODE] = [target_node]
6471
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6472

    
6473
  def DeclareLocks(self, level):
6474
    if level == locking.LEVEL_NODE:
6475
      self._LockInstancesNodes(primary_only=True)
6476

    
6477
  def BuildHooksEnv(self):
6478
    """Build hooks env.
6479

6480
    This runs on master, primary and secondary nodes of the instance.
6481

6482
    """
6483
    env = {
6484
      "TARGET_NODE": self.op.target_node,
6485
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
6486
      }
6487
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6488
    return env
6489

    
6490
  def BuildHooksNodes(self):
6491
    """Build hooks nodes.
6492

6493
    """
6494
    nl = [
6495
      self.cfg.GetMasterNode(),
6496
      self.instance.primary_node,
6497
      self.op.target_node,
6498
      ]
6499
    return (nl, nl)
6500

    
6501
  def CheckPrereq(self):
6502
    """Check prerequisites.
6503

6504
    This checks that the instance is in the cluster.
6505

6506
    """
6507
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6508
    assert self.instance is not None, \
6509
      "Cannot retrieve locked instance %s" % self.op.instance_name
6510

    
6511
    node = self.cfg.GetNodeInfo(self.op.target_node)
6512
    assert node is not None, \
6513
      "Cannot retrieve locked node %s" % self.op.target_node
6514

    
6515
    self.target_node = target_node = node.name
6516

    
6517
    if target_node == instance.primary_node:
6518
      raise errors.OpPrereqError("Instance %s is already on the node %s" %
6519
                                 (instance.name, target_node),
6520
                                 errors.ECODE_STATE)
6521

    
6522
    bep = self.cfg.GetClusterInfo().FillBE(instance)
6523

    
6524
    for idx, dsk in enumerate(instance.disks):
6525
      if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
6526
        raise errors.OpPrereqError("Instance disk %d has a complex layout,"
6527
                                   " cannot copy" % idx, errors.ECODE_STATE)
6528

    
6529
    _CheckNodeOnline(self, target_node)
6530
    _CheckNodeNotDrained(self, target_node)
6531
    _CheckNodeVmCapable(self, target_node)
6532

    
6533
    if instance.admin_up:
6534
      # check memory requirements on the secondary node
6535
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
6536
                           instance.name, bep[constants.BE_MEMORY],
6537
                           instance.hypervisor)
6538
    else:
6539
      self.LogInfo("Not checking memory on the secondary node as"
6540
                   " instance will not be started")
6541

    
6542
    # check bridge existance
6543
    _CheckInstanceBridgesExist(self, instance, node=target_node)
6544

    
6545
  def Exec(self, feedback_fn):
6546
    """Move an instance.
6547

6548
    The move is done by shutting it down on its present node, copying
6549
    the data over (slow) and starting it on the new node.
6550

6551
    """
6552
    instance = self.instance
6553

    
6554
    source_node = instance.primary_node
6555
    target_node = self.target_node
6556

    
6557
    self.LogInfo("Shutting down instance %s on source node %s",
6558
                 instance.name, source_node)
6559

    
6560
    result = self.rpc.call_instance_shutdown(source_node, instance,
6561
                                             self.op.shutdown_timeout)
6562
    msg = result.fail_msg
6563
    if msg:
6564
      if self.op.ignore_consistency:
6565
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
6566
                             " Proceeding anyway. Please make sure node"
6567
                             " %s is down. Error details: %s",
6568
                             instance.name, source_node, source_node, msg)
6569
      else:
6570
        raise errors.OpExecError("Could not shutdown instance %s on"
6571
                                 " node %s: %s" %
6572
                                 (instance.name, source_node, msg))
6573

    
6574
    # create the target disks
6575
    try:
6576
      _CreateDisks(self, instance, target_node=target_node)
6577
    except errors.OpExecError:
6578
      self.LogWarning("Device creation failed, reverting...")
6579
      try:
6580
        _RemoveDisks(self, instance, target_node=target_node)
6581
      finally:
6582
        self.cfg.ReleaseDRBDMinors(instance.name)
6583
        raise
6584

    
6585
    cluster_name = self.cfg.GetClusterInfo().cluster_name
6586

    
6587
    errs = []
6588
    # activate, get path, copy the data over
6589
    for idx, disk in enumerate(instance.disks):
6590
      self.LogInfo("Copying data for disk %d", idx)
6591
      result = self.rpc.call_blockdev_assemble(target_node, disk,
6592
                                               instance.name, True, idx)
6593
      if result.fail_msg:
6594
        self.LogWarning("Can't assemble newly created disk %d: %s",
6595
                        idx, result.fail_msg)
6596
        errs.append(result.fail_msg)
6597
        break
6598
      dev_path = result.payload
6599
      result = self.rpc.call_blockdev_export(source_node, disk,
6600
                                             target_node, dev_path,
6601
                                             cluster_name)
6602
      if result.fail_msg:
6603
        self.LogWarning("Can't copy data over for disk %d: %s",
6604
                        idx, result.fail_msg)
6605
        errs.append(result.fail_msg)
6606
        break
6607

    
6608
    if errs:
6609
      self.LogWarning("Some disks failed to copy, aborting")
6610
      try:
6611
        _RemoveDisks(self, instance, target_node=target_node)
6612
      finally:
6613
        self.cfg.ReleaseDRBDMinors(instance.name)
6614
        raise errors.OpExecError("Errors during disk copy: %s" %
6615
                                 (",".join(errs),))
6616

    
6617
    instance.primary_node = target_node
6618
    self.cfg.Update(instance, feedback_fn)
6619

    
6620
    self.LogInfo("Removing the disks on the original node")
6621
    _RemoveDisks(self, instance, target_node=source_node)
6622

    
6623
    # Only start the instance if it's marked as up
6624
    if instance.admin_up:
6625
      self.LogInfo("Starting instance %s on node %s",
6626
                   instance.name, target_node)
6627

    
6628
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
6629
                                           ignore_secondaries=True)
6630
      if not disks_ok:
6631
        _ShutdownInstanceDisks(self, instance)
6632
        raise errors.OpExecError("Can't activate the instance's disks")
6633

    
6634
      result = self.rpc.call_instance_start(target_node, instance, None, None)
6635
      msg = result.fail_msg
6636
      if msg:
6637
        _ShutdownInstanceDisks(self, instance)
6638
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
6639
                                 (instance.name, target_node, msg))
6640

    
6641

    
6642
class LUNodeMigrate(LogicalUnit):
6643
  """Migrate all instances from a node.
6644

6645
  """
6646
  HPATH = "node-migrate"
6647
  HTYPE = constants.HTYPE_NODE
6648
  REQ_BGL = False
6649

    
6650
  def CheckArguments(self):
6651
    pass
6652

    
6653
  def ExpandNames(self):
6654
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6655

    
6656
    self.needed_locks = {}
6657

    
6658
    # Create tasklets for migrating instances for all instances on this node
6659
    names = []
6660
    tasklets = []
6661

    
6662
    self.lock_all_nodes = False
6663

    
6664
    for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name):
6665
      logging.debug("Migrating instance %s", inst.name)
6666
      names.append(inst.name)
6667

    
6668
      tasklets.append(TLMigrateInstance(self, inst.name, cleanup=False))
6669

    
6670
      if inst.disk_template in constants.DTS_EXT_MIRROR:
6671
        # We need to lock all nodes, as the iallocator will choose the
6672
        # destination nodes afterwards
6673
        self.lock_all_nodes = True
6674

    
6675
    self.tasklets = tasklets
6676

    
6677
    # Declare node locks
6678
    if self.lock_all_nodes:
6679
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6680
    else:
6681
      self.needed_locks[locking.LEVEL_NODE] = [self.op.node_name]
6682
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6683

    
6684
    # Declare instance locks
6685
    self.needed_locks[locking.LEVEL_INSTANCE] = names
6686

    
6687
  def DeclareLocks(self, level):
6688
    if level == locking.LEVEL_NODE and not self.lock_all_nodes:
6689
      self._LockInstancesNodes()
6690

    
6691
  def BuildHooksEnv(self):
6692
    """Build hooks env.
6693

6694
    This runs on the master, the primary and all the secondaries.
6695

6696
    """
6697
    return {
6698
      "NODE_NAME": self.op.node_name,
6699
      }
6700

    
6701
  def BuildHooksNodes(self):
6702
    """Build hooks nodes.
6703

6704
    """
6705
    nl = [self.cfg.GetMasterNode()]
6706
    return (nl, nl)
6707

    
6708

    
6709
class TLMigrateInstance(Tasklet):
6710
  """Tasklet class for instance migration.
6711

6712
  @type live: boolean
6713
  @ivar live: whether the migration will be done live or non-live;
6714
      this variable is initalized only after CheckPrereq has run
6715
  @type cleanup: boolean
6716
  @ivar cleanup: Wheater we cleanup from a failed migration
6717
  @type iallocator: string
6718
  @ivar iallocator: The iallocator used to determine target_node
6719
  @type target_node: string
6720
  @ivar target_node: If given, the target_node to reallocate the instance to
6721
  @type failover: boolean
6722
  @ivar failover: Whether operation results in failover or migration
6723
  @type fallback: boolean
6724
  @ivar fallback: Whether fallback to failover is allowed if migration not
6725
                  possible
6726
  @type ignore_consistency: boolean
6727
  @ivar ignore_consistency: Wheter we should ignore consistency between source
6728
                            and target node
6729
  @type shutdown_timeout: int
6730
  @ivar shutdown_timeout: In case of failover timeout of the shutdown
6731

6732
  """
6733
  def __init__(self, lu, instance_name, cleanup=False,
6734
               failover=False, fallback=False,
6735
               ignore_consistency=False,
6736
               shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT):
6737
    """Initializes this class.
6738

6739
    """
6740
    Tasklet.__init__(self, lu)
6741

    
6742
    # Parameters
6743
    self.instance_name = instance_name
6744
    self.cleanup = cleanup
6745
    self.live = False # will be overridden later
6746
    self.failover = failover
6747
    self.fallback = fallback
6748
    self.ignore_consistency = ignore_consistency
6749
    self.shutdown_timeout = shutdown_timeout
6750

    
6751
  def CheckPrereq(self):
6752
    """Check prerequisites.
6753

6754
    This checks that the instance is in the cluster.
6755

6756
    """
6757
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
6758
    instance = self.cfg.GetInstanceInfo(instance_name)
6759
    assert instance is not None
6760
    self.instance = instance
6761

    
6762
    if (not self.cleanup and not instance.admin_up and not self.failover and
6763
        self.fallback):
6764
      self.lu.LogInfo("Instance is marked down, fallback allowed, switching"
6765
                      " to failover")
6766
      self.failover = True
6767

    
6768
    if instance.disk_template not in constants.DTS_MIRRORED:
6769
      if self.failover:
6770
        text = "failovers"
6771
      else:
6772
        text = "migrations"
6773
      raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
6774
                                 " %s" % (instance.disk_template, text),
6775
                                 errors.ECODE_STATE)
6776

    
6777
    if instance.disk_template in constants.DTS_EXT_MIRROR:
6778
      _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
6779

    
6780
      if self.lu.op.iallocator:
6781
        self._RunAllocator()
6782
      else:
6783
        # We set set self.target_node as it is required by
6784
        # BuildHooksEnv
6785
        self.target_node = self.lu.op.target_node
6786

    
6787
      # self.target_node is already populated, either directly or by the
6788
      # iallocator run
6789
      target_node = self.target_node
6790
      if self.target_node == instance.primary_node:
6791
        raise errors.OpPrereqError("Cannot migrate instance %s"
6792
                                   " to its primary (%s)" %
6793
                                   (instance.name, instance.primary_node))
6794

    
6795
      if len(self.lu.tasklets) == 1:
6796
        # It is safe to release locks only when we're the only tasklet
6797
        # in the LU
6798
        _ReleaseLocks(self.lu, locking.LEVEL_NODE,
6799
                      keep=[instance.primary_node, self.target_node])
6800

    
6801
    else:
6802
      secondary_nodes = instance.secondary_nodes
6803
      if not secondary_nodes:
6804
        raise errors.ConfigurationError("No secondary node but using"
6805
                                        " %s disk template" %
6806
                                        instance.disk_template)
6807
      target_node = secondary_nodes[0]
6808
      if self.lu.op.iallocator or (self.lu.op.target_node and
6809
                                   self.lu.op.target_node != target_node):
6810
        if self.failover:
6811
          text = "failed over"
6812
        else:
6813
          text = "migrated"
6814
        raise errors.OpPrereqError("Instances with disk template %s cannot"
6815
                                   " be %s to arbitrary nodes"
6816
                                   " (neither an iallocator nor a target"
6817
                                   " node can be passed)" %
6818
                                   (instance.disk_template, text),
6819
                                   errors.ECODE_INVAL)
6820

    
6821
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
6822

    
6823
    # check memory requirements on the secondary node
6824
    if not self.failover or instance.admin_up:
6825
      _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
6826
                           instance.name, i_be[constants.BE_MEMORY],
6827
                           instance.hypervisor)
6828
    else:
6829
      self.lu.LogInfo("Not checking memory on the secondary node as"
6830
                      " instance will not be started")
6831

    
6832
    # check bridge existance
6833
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
6834

    
6835
    if not self.cleanup:
6836
      _CheckNodeNotDrained(self.lu, target_node)
6837
      if not self.failover:
6838
        result = self.rpc.call_instance_migratable(instance.primary_node,
6839
                                                   instance)
6840
        if result.fail_msg and self.fallback:
6841
          self.lu.LogInfo("Can't migrate, instance offline, fallback to"
6842
                          " failover")
6843
          self.failover = True
6844
        else:
6845
          result.Raise("Can't migrate, please use failover",
6846
                       prereq=True, ecode=errors.ECODE_STATE)
6847

    
6848
    assert not (self.failover and self.cleanup)
6849

    
6850
    if not self.failover:
6851
      if self.lu.op.live is not None and self.lu.op.mode is not None:
6852
        raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
6853
                                   " parameters are accepted",
6854
                                   errors.ECODE_INVAL)
6855
      if self.lu.op.live is not None:
6856
        if self.lu.op.live:
6857
          self.lu.op.mode = constants.HT_MIGRATION_LIVE
6858
        else:
6859
          self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
6860
        # reset the 'live' parameter to None so that repeated
6861
        # invocations of CheckPrereq do not raise an exception
6862
        self.lu.op.live = None
6863
      elif self.lu.op.mode is None:
6864
        # read the default value from the hypervisor
6865
        i_hv = self.cfg.GetClusterInfo().FillHV(self.instance,
6866
                                                skip_globals=False)
6867
        self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
6868

    
6869
      self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
6870
    else:
6871
      # Failover is never live
6872
      self.live = False
6873

    
6874
  def _RunAllocator(self):
6875
    """Run the allocator based on input opcode.
6876

6877
    """
6878
    ial = IAllocator(self.cfg, self.rpc,
6879
                     mode=constants.IALLOCATOR_MODE_RELOC,
6880
                     name=self.instance_name,
6881
                     # TODO See why hail breaks with a single node below
6882
                     relocate_from=[self.instance.primary_node,
6883
                                    self.instance.primary_node],
6884
                     )
6885

    
6886
    ial.Run(self.lu.op.iallocator)
6887

    
6888
    if not ial.success:
6889
      raise errors.OpPrereqError("Can't compute nodes using"
6890
                                 " iallocator '%s': %s" %
6891
                                 (self.lu.op.iallocator, ial.info),
6892
                                 errors.ECODE_NORES)
6893
    if len(ial.result) != ial.required_nodes:
6894
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
6895
                                 " of nodes (%s), required %s" %
6896
                                 (self.lu.op.iallocator, len(ial.result),
6897
                                  ial.required_nodes), errors.ECODE_FAULT)
6898
    self.target_node = ial.result[0]
6899
    self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
6900
                 self.instance_name, self.lu.op.iallocator,
6901
                 utils.CommaJoin(ial.result))
6902

    
6903
  def _WaitUntilSync(self):
6904
    """Poll with custom rpc for disk sync.
6905

6906
    This uses our own step-based rpc call.
6907

6908
    """
6909
    self.feedback_fn("* wait until resync is done")
6910
    all_done = False
6911
    while not all_done:
6912
      all_done = True
6913
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
6914
                                            self.nodes_ip,
6915
                                            self.instance.disks)
6916
      min_percent = 100
6917
      for node, nres in result.items():
6918
        nres.Raise("Cannot resync disks on node %s" % node)
6919
        node_done, node_percent = nres.payload
6920
        all_done = all_done and node_done
6921
        if node_percent is not None:
6922
          min_percent = min(min_percent, node_percent)
6923
      if not all_done:
6924
        if min_percent < 100:
6925
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
6926
        time.sleep(2)
6927

    
6928
  def _EnsureSecondary(self, node):
6929
    """Demote a node to secondary.
6930

6931
    """
6932
    self.feedback_fn("* switching node %s to secondary mode" % node)
6933

    
6934
    for dev in self.instance.disks:
6935
      self.cfg.SetDiskID(dev, node)
6936

    
6937
    result = self.rpc.call_blockdev_close(node, self.instance.name,
6938
                                          self.instance.disks)
6939
    result.Raise("Cannot change disk to secondary on node %s" % node)
6940

    
6941
  def _GoStandalone(self):
6942
    """Disconnect from the network.
6943

6944
    """
6945
    self.feedback_fn("* changing into standalone mode")
6946
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
6947
                                               self.instance.disks)
6948
    for node, nres in result.items():
6949
      nres.Raise("Cannot disconnect disks node %s" % node)
6950

    
6951
  def _GoReconnect(self, multimaster):
6952
    """Reconnect to the network.
6953

6954
    """
6955
    if multimaster:
6956
      msg = "dual-master"
6957
    else:
6958
      msg = "single-master"
6959
    self.feedback_fn("* changing disks into %s mode" % msg)
6960
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
6961
                                           self.instance.disks,
6962
                                           self.instance.name, multimaster)
6963
    for node, nres in result.items():
6964
      nres.Raise("Cannot change disks config on node %s" % node)
6965

    
6966
  def _ExecCleanup(self):
6967
    """Try to cleanup after a failed migration.
6968

6969
    The cleanup is done by:
6970
      - check that the instance is running only on one node
6971
        (and update the config if needed)
6972
      - change disks on its secondary node to secondary
6973
      - wait until disks are fully synchronized
6974
      - disconnect from the network
6975
      - change disks into single-master mode
6976
      - wait again until disks are fully synchronized
6977

6978
    """
6979
    instance = self.instance
6980
    target_node = self.target_node
6981
    source_node = self.source_node
6982

    
6983
    # check running on only one node
6984
    self.feedback_fn("* checking where the instance actually runs"
6985
                     " (if this hangs, the hypervisor might be in"
6986
                     " a bad state)")
6987
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
6988
    for node, result in ins_l.items():
6989
      result.Raise("Can't contact node %s" % node)
6990

    
6991
    runningon_source = instance.name in ins_l[source_node].payload
6992
    runningon_target = instance.name in ins_l[target_node].payload
6993

    
6994
    if runningon_source and runningon_target:
6995
      raise errors.OpExecError("Instance seems to be running on two nodes,"
6996
                               " or the hypervisor is confused; you will have"
6997
                               " to ensure manually that it runs only on one"
6998
                               " and restart this operation")
6999

    
7000
    if not (runningon_source or runningon_target):
7001
      raise errors.OpExecError("Instance does not seem to be running at all;"
7002
                               " in this case it's safer to repair by"
7003
                               " running 'gnt-instance stop' to ensure disk"
7004
                               " shutdown, and then restarting it")
7005

    
7006
    if runningon_target:
7007
      # the migration has actually succeeded, we need to update the config
7008
      self.feedback_fn("* instance running on secondary node (%s),"
7009
                       " updating config" % target_node)
7010
      instance.primary_node = target_node
7011
      self.cfg.Update(instance, self.feedback_fn)
7012
      demoted_node = source_node
7013
    else:
7014
      self.feedback_fn("* instance confirmed to be running on its"
7015
                       " primary node (%s)" % source_node)
7016
      demoted_node = target_node
7017

    
7018
    if instance.disk_template in constants.DTS_INT_MIRROR:
7019
      self._EnsureSecondary(demoted_node)
7020
      try:
7021
        self._WaitUntilSync()
7022
      except errors.OpExecError:
7023
        # we ignore here errors, since if the device is standalone, it
7024
        # won't be able to sync
7025
        pass
7026
      self._GoStandalone()
7027
      self._GoReconnect(False)
7028
      self._WaitUntilSync()
7029

    
7030
    self.feedback_fn("* done")
7031

    
7032
  def _RevertDiskStatus(self):
7033
    """Try to revert the disk status after a failed migration.
7034

7035
    """
7036
    target_node = self.target_node
7037
    if self.instance.disk_template in constants.DTS_EXT_MIRROR:
7038
      return
7039

    
7040
    try:
7041
      self._EnsureSecondary(target_node)
7042
      self._GoStandalone()
7043
      self._GoReconnect(False)
7044
      self._WaitUntilSync()
7045
    except errors.OpExecError, err:
7046
      self.lu.LogWarning("Migration failed and I can't reconnect the drives,"
7047
                         " please try to recover the instance manually;"
7048
                         " error '%s'" % str(err))
7049

    
7050
  def _AbortMigration(self):
7051
    """Call the hypervisor code to abort a started migration.
7052

7053
    """
7054
    instance = self.instance
7055
    target_node = self.target_node
7056
    migration_info = self.migration_info
7057

    
7058
    abort_result = self.rpc.call_finalize_migration(target_node,
7059
                                                    instance,
7060
                                                    migration_info,
7061
                                                    False)
7062
    abort_msg = abort_result.fail_msg
7063
    if abort_msg:
7064
      logging.error("Aborting migration failed on target node %s: %s",
7065
                    target_node, abort_msg)
7066
      # Don't raise an exception here, as we stil have to try to revert the
7067
      # disk status, even if this step failed.
7068

    
7069
  def _ExecMigration(self):
7070
    """Migrate an instance.
7071

7072
    The migrate is done by:
7073
      - change the disks into dual-master mode
7074
      - wait until disks are fully synchronized again
7075
      - migrate the instance
7076
      - change disks on the new secondary node (the old primary) to secondary
7077
      - wait until disks are fully synchronized
7078
      - change disks into single-master mode
7079

7080
    """
7081
    instance = self.instance
7082
    target_node = self.target_node
7083
    source_node = self.source_node
7084

    
7085
    self.feedback_fn("* checking disk consistency between source and target")
7086
    for dev in instance.disks:
7087
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
7088
        raise errors.OpExecError("Disk %s is degraded or not fully"
7089
                                 " synchronized on target node,"
7090
                                 " aborting migration" % dev.iv_name)
7091

    
7092
    # First get the migration information from the remote node
7093
    result = self.rpc.call_migration_info(source_node, instance)
7094
    msg = result.fail_msg
7095
    if msg:
7096
      log_err = ("Failed fetching source migration information from %s: %s" %
7097
                 (source_node, msg))
7098
      logging.error(log_err)
7099
      raise errors.OpExecError(log_err)
7100

    
7101
    self.migration_info = migration_info = result.payload
7102

    
7103
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
7104
      # Then switch the disks to master/master mode
7105
      self._EnsureSecondary(target_node)
7106
      self._GoStandalone()
7107
      self._GoReconnect(True)
7108
      self._WaitUntilSync()
7109

    
7110
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
7111
    result = self.rpc.call_accept_instance(target_node,
7112
                                           instance,
7113
                                           migration_info,
7114
                                           self.nodes_ip[target_node])
7115

    
7116
    msg = result.fail_msg
7117
    if msg:
7118
      logging.error("Instance pre-migration failed, trying to revert"
7119
                    " disk status: %s", msg)
7120
      self.feedback_fn("Pre-migration failed, aborting")
7121
      self._AbortMigration()
7122
      self._RevertDiskStatus()
7123
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
7124
                               (instance.name, msg))
7125

    
7126
    self.feedback_fn("* migrating instance to %s" % target_node)
7127
    result = self.rpc.call_instance_migrate(source_node, instance,
7128
                                            self.nodes_ip[target_node],
7129
                                            self.live)
7130
    msg = result.fail_msg
7131
    if msg:
7132
      logging.error("Instance migration failed, trying to revert"
7133
                    " disk status: %s", msg)
7134
      self.feedback_fn("Migration failed, aborting")
7135
      self._AbortMigration()
7136
      self._RevertDiskStatus()
7137
      raise errors.OpExecError("Could not migrate instance %s: %s" %
7138
                               (instance.name, msg))
7139

    
7140
    instance.primary_node = target_node
7141
    # distribute new instance config to the other nodes
7142
    self.cfg.Update(instance, self.feedback_fn)
7143

    
7144
    result = self.rpc.call_finalize_migration(target_node,
7145
                                              instance,
7146
                                              migration_info,
7147
                                              True)
7148
    msg = result.fail_msg
7149
    if msg:
7150
      logging.error("Instance migration succeeded, but finalization failed:"
7151
                    " %s", msg)
7152
      raise errors.OpExecError("Could not finalize instance migration: %s" %
7153
                               msg)
7154

    
7155
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
7156
      self._EnsureSecondary(source_node)
7157
      self._WaitUntilSync()
7158
      self._GoStandalone()
7159
      self._GoReconnect(False)
7160
      self._WaitUntilSync()
7161

    
7162
    self.feedback_fn("* done")
7163

    
7164
  def _ExecFailover(self):
7165
    """Failover an instance.
7166

7167
    The failover is done by shutting it down on its present node and
7168
    starting it on the secondary.
7169

7170
    """
7171
    instance = self.instance
7172
    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
7173

    
7174
    source_node = instance.primary_node
7175
    target_node = self.target_node
7176

    
7177
    if instance.admin_up:
7178
      self.feedback_fn("* checking disk consistency between source and target")
7179
      for dev in instance.disks:
7180
        # for drbd, these are drbd over lvm
7181
        if not _CheckDiskConsistency(self, dev, target_node, False):
7182
          if not self.ignore_consistency:
7183
            raise errors.OpExecError("Disk %s is degraded on target node,"
7184
                                     " aborting failover" % dev.iv_name)
7185
    else:
7186
      self.feedback_fn("* not checking disk consistency as instance is not"
7187
                       " running")
7188

    
7189
    self.feedback_fn("* shutting down instance on source node")
7190
    logging.info("Shutting down instance %s on node %s",
7191
                 instance.name, source_node)
7192

    
7193
    result = self.rpc.call_instance_shutdown(source_node, instance,
7194
                                             self.shutdown_timeout)
7195
    msg = result.fail_msg
7196
    if msg:
7197
      if self.ignore_consistency or primary_node.offline:
7198
        self.lu.LogWarning("Could not shutdown instance %s on node %s,"
7199
                           " proceeding anyway; please make sure node"
7200
                           " %s is down; error details: %s",
7201
                           instance.name, source_node, source_node, msg)
7202
      else:
7203
        raise errors.OpExecError("Could not shutdown instance %s on"
7204
                                 " node %s: %s" %
7205
                                 (instance.name, source_node, msg))
7206

    
7207
    self.feedback_fn("* deactivating the instance's disks on source node")
7208
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
7209
      raise errors.OpExecError("Can't shut down the instance's disks.")
7210

    
7211
    instance.primary_node = target_node
7212
    # distribute new instance config to the other nodes
7213
    self.cfg.Update(instance, self.feedback_fn)
7214

    
7215
    # Only start the instance if it's marked as up
7216
    if instance.admin_up:
7217
      self.feedback_fn("* activating the instance's disks on target node")
7218
      logging.info("Starting instance %s on node %s",
7219
                   instance.name, target_node)
7220

    
7221
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
7222
                                           ignore_secondaries=True)
7223
      if not disks_ok:
7224
        _ShutdownInstanceDisks(self, instance)
7225
        raise errors.OpExecError("Can't activate the instance's disks")
7226

    
7227
      self.feedback_fn("* starting the instance on the target node")
7228
      result = self.rpc.call_instance_start(target_node, instance, None, None)
7229
      msg = result.fail_msg
7230
      if msg:
7231
        _ShutdownInstanceDisks(self, instance)
7232
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
7233
                                 (instance.name, target_node, msg))
7234

    
7235
  def Exec(self, feedback_fn):
7236
    """Perform the migration.
7237

7238
    """
7239
    self.feedback_fn = feedback_fn
7240
    self.source_node = self.instance.primary_node
7241

    
7242
    # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
7243
    if self.instance.disk_template in constants.DTS_INT_MIRROR:
7244
      self.target_node = self.instance.secondary_nodes[0]
7245
      # Otherwise self.target_node has been populated either
7246
      # directly, or through an iallocator.
7247

    
7248
    self.all_nodes = [self.source_node, self.target_node]
7249
    self.nodes_ip = {
7250
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
7251
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
7252
      }
7253

    
7254
    if self.failover:
7255
      feedback_fn("Failover instance %s" % self.instance.name)
7256
      self._ExecFailover()
7257
    else:
7258
      feedback_fn("Migrating instance %s" % self.instance.name)
7259

    
7260
      if self.cleanup:
7261
        return self._ExecCleanup()
7262
      else:
7263
        return self._ExecMigration()
7264

    
7265

    
7266
def _CreateBlockDev(lu, node, instance, device, force_create,
7267
                    info, force_open):
7268
  """Create a tree of block devices on a given node.
7269

7270
  If this device type has to be created on secondaries, create it and
7271
  all its children.
7272

7273
  If not, just recurse to children keeping the same 'force' value.
7274

7275
  @param lu: the lu on whose behalf we execute
7276
  @param node: the node on which to create the device
7277
  @type instance: L{objects.Instance}
7278
  @param instance: the instance which owns the device
7279
  @type device: L{objects.Disk}
7280
  @param device: the device to create
7281
  @type force_create: boolean
7282
  @param force_create: whether to force creation of this device; this
7283
      will be change to True whenever we find a device which has
7284
      CreateOnSecondary() attribute
7285
  @param info: the extra 'metadata' we should attach to the device
7286
      (this will be represented as a LVM tag)
7287
  @type force_open: boolean
7288
  @param force_open: this parameter will be passes to the
7289
      L{backend.BlockdevCreate} function where it specifies
7290
      whether we run on primary or not, and it affects both
7291
      the child assembly and the device own Open() execution
7292

7293
  """
7294
  if device.CreateOnSecondary():
7295
    force_create = True
7296

    
7297
  if device.children:
7298
    for child in device.children:
7299
      _CreateBlockDev(lu, node, instance, child, force_create,
7300
                      info, force_open)
7301

    
7302
  if not force_create:
7303
    return
7304

    
7305
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
7306

    
7307

    
7308
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
7309
  """Create a single block device on a given node.
7310

7311
  This will not recurse over children of the device, so they must be
7312
  created in advance.
7313

7314
  @param lu: the lu on whose behalf we execute
7315
  @param node: the node on which to create the device
7316
  @type instance: L{objects.Instance}
7317
  @param instance: the instance which owns the device
7318
  @type device: L{objects.Disk}
7319
  @param device: the device to create
7320
  @param info: the extra 'metadata' we should attach to the device
7321
      (this will be represented as a LVM tag)
7322
  @type force_open: boolean
7323
  @param force_open: this parameter will be passes to the
7324
      L{backend.BlockdevCreate} function where it specifies
7325
      whether we run on primary or not, and it affects both
7326
      the child assembly and the device own Open() execution
7327

7328
  """
7329
  lu.cfg.SetDiskID(device, node)
7330
  result = lu.rpc.call_blockdev_create(node, device, device.size,
7331
                                       instance.name, force_open, info)
7332
  result.Raise("Can't create block device %s on"
7333
               " node %s for instance %s" % (device, node, instance.name))
7334
  if device.physical_id is None:
7335
    device.physical_id = result.payload
7336

    
7337

    
7338
def _GenerateUniqueNames(lu, exts):
7339
  """Generate a suitable LV name.
7340

7341
  This will generate a logical volume name for the given instance.
7342

7343
  """
7344
  results = []
7345
  for val in exts:
7346
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
7347
    results.append("%s%s" % (new_id, val))
7348
  return results
7349

    
7350

    
7351
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
7352
                         iv_name, p_minor, s_minor):
7353
  """Generate a drbd8 device complete with its children.
7354

7355
  """
7356
  assert len(vgnames) == len(names) == 2
7357
  port = lu.cfg.AllocatePort()
7358
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
7359
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
7360
                          logical_id=(vgnames[0], names[0]))
7361
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
7362
                          logical_id=(vgnames[1], names[1]))
7363
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
7364
                          logical_id=(primary, secondary, port,
7365
                                      p_minor, s_minor,
7366
                                      shared_secret),
7367
                          children=[dev_data, dev_meta],
7368
                          iv_name=iv_name)
7369
  return drbd_dev
7370

    
7371

    
7372
def _GenerateDiskTemplate(lu, template_name,
7373
                          instance_name, primary_node,
7374
                          secondary_nodes, disk_info,
7375
                          file_storage_dir, file_driver,
7376
                          base_index, feedback_fn):
7377
  """Generate the entire disk layout for a given template type.
7378

7379
  """
7380
  #TODO: compute space requirements
7381

    
7382
  vgname = lu.cfg.GetVGName()
7383
  disk_count = len(disk_info)
7384
  disks = []
7385
  if template_name == constants.DT_DISKLESS:
7386
    pass
7387
  elif template_name == constants.DT_PLAIN:
7388
    if len(secondary_nodes) != 0:
7389
      raise errors.ProgrammerError("Wrong template configuration")
7390

    
7391
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
7392
                                      for i in range(disk_count)])
7393
    for idx, disk in enumerate(disk_info):
7394
      disk_index = idx + base_index
7395
      vg = disk.get(constants.IDISK_VG, vgname)
7396
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
7397
      disk_dev = objects.Disk(dev_type=constants.LD_LV,
7398
                              size=disk[constants.IDISK_SIZE],
7399
                              logical_id=(vg, names[idx]),
7400
                              iv_name="disk/%d" % disk_index,
7401
                              mode=disk[constants.IDISK_MODE])
7402
      disks.append(disk_dev)
7403
  elif template_name == constants.DT_DRBD8:
7404
    if len(secondary_nodes) != 1:
7405
      raise errors.ProgrammerError("Wrong template configuration")
7406
    remote_node = secondary_nodes[0]
7407
    minors = lu.cfg.AllocateDRBDMinor(
7408
      [primary_node, remote_node] * len(disk_info), instance_name)
7409

    
7410
    names = []
7411
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
7412
                                               for i in range(disk_count)]):
7413
      names.append(lv_prefix + "_data")
7414
      names.append(lv_prefix + "_meta")
7415
    for idx, disk in enumerate(disk_info):
7416
      disk_index = idx + base_index
7417
      data_vg = disk.get(constants.IDISK_VG, vgname)
7418
      meta_vg = disk.get(constants.IDISK_METAVG, data_vg)
7419
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
7420
                                      disk[constants.IDISK_SIZE],
7421
                                      [data_vg, meta_vg],
7422
                                      names[idx * 2:idx * 2 + 2],
7423
                                      "disk/%d" % disk_index,
7424
                                      minors[idx * 2], minors[idx * 2 + 1])
7425
      disk_dev.mode = disk[constants.IDISK_MODE]
7426
      disks.append(disk_dev)
7427
  elif template_name == constants.DT_FILE:
7428
    if len(secondary_nodes) != 0:
7429
      raise errors.ProgrammerError("Wrong template configuration")
7430

    
7431
    opcodes.RequireFileStorage()
7432

    
7433
    for idx, disk in enumerate(disk_info):
7434
      disk_index = idx + base_index
7435
      disk_dev = objects.Disk(dev_type=constants.LD_FILE,
7436
                              size=disk[constants.IDISK_SIZE],
7437
                              iv_name="disk/%d" % disk_index,
7438
                              logical_id=(file_driver,
7439
                                          "%s/disk%d" % (file_storage_dir,
7440
                                                         disk_index)),
7441
                              mode=disk[constants.IDISK_MODE])
7442
      disks.append(disk_dev)
7443
  elif template_name == constants.DT_SHARED_FILE:
7444
    if len(secondary_nodes) != 0:
7445
      raise errors.ProgrammerError("Wrong template configuration")
7446

    
7447
    opcodes.RequireSharedFileStorage()
7448

    
7449
    for idx, disk in enumerate(disk_info):
7450
      disk_index = idx + base_index
7451
      disk_dev = objects.Disk(dev_type=constants.LD_FILE,
7452
                              size=disk[constants.IDISK_SIZE],
7453
                              iv_name="disk/%d" % disk_index,
7454
                              logical_id=(file_driver,
7455
                                          "%s/disk%d" % (file_storage_dir,
7456
                                                         disk_index)),
7457
                              mode=disk[constants.IDISK_MODE])
7458
      disks.append(disk_dev)
7459
  elif template_name == constants.DT_BLOCK:
7460
    if len(secondary_nodes) != 0:
7461
      raise errors.ProgrammerError("Wrong template configuration")
7462

    
7463
    for idx, disk in enumerate(disk_info):
7464
      disk_index = idx + base_index
7465
      disk_dev = objects.Disk(dev_type=constants.LD_BLOCKDEV,
7466
                              size=disk[constants.IDISK_SIZE],
7467
                              logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
7468
                                          disk[constants.IDISK_ADOPT]),
7469
                              iv_name="disk/%d" % disk_index,
7470
                              mode=disk[constants.IDISK_MODE])
7471
      disks.append(disk_dev)
7472

    
7473
  else:
7474
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
7475
  return disks
7476

    
7477

    
7478
def _GetInstanceInfoText(instance):
7479
  """Compute that text that should be added to the disk's metadata.
7480

7481
  """
7482
  return "originstname+%s" % instance.name
7483

    
7484

    
7485
def _CalcEta(time_taken, written, total_size):
7486
  """Calculates the ETA based on size written and total size.
7487

7488
  @param time_taken: The time taken so far
7489
  @param written: amount written so far
7490
  @param total_size: The total size of data to be written
7491
  @return: The remaining time in seconds
7492

7493
  """
7494
  avg_time = time_taken / float(written)
7495
  return (total_size - written) * avg_time
7496

    
7497

    
7498
def _WipeDisks(lu, instance):
7499
  """Wipes instance disks.
7500

7501
  @type lu: L{LogicalUnit}
7502
  @param lu: the logical unit on whose behalf we execute
7503
  @type instance: L{objects.Instance}
7504
  @param instance: the instance whose disks we should create
7505
  @return: the success of the wipe
7506

7507
  """
7508
  node = instance.primary_node
7509

    
7510
  for device in instance.disks:
7511
    lu.cfg.SetDiskID(device, node)
7512

    
7513
  logging.info("Pause sync of instance %s disks", instance.name)
7514
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
7515

    
7516
  for idx, success in enumerate(result.payload):
7517
    if not success:
7518
      logging.warn("pause-sync of instance %s for disks %d failed",
7519
                   instance.name, idx)
7520

    
7521
  try:
7522
    for idx, device in enumerate(instance.disks):
7523
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
7524
      # MAX_WIPE_CHUNK at max
7525
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
7526
                            constants.MIN_WIPE_CHUNK_PERCENT)
7527
      # we _must_ make this an int, otherwise rounding errors will
7528
      # occur
7529
      wipe_chunk_size = int(wipe_chunk_size)
7530

    
7531
      lu.LogInfo("* Wiping disk %d", idx)
7532
      logging.info("Wiping disk %d for instance %s, node %s using"
7533
                   " chunk size %s", idx, instance.name, node, wipe_chunk_size)
7534

    
7535
      offset = 0
7536
      size = device.size
7537
      last_output = 0
7538
      start_time = time.time()
7539

    
7540
      while offset < size:
7541
        wipe_size = min(wipe_chunk_size, size - offset)
7542
        logging.debug("Wiping disk %d, offset %s, chunk %s",
7543
                      idx, offset, wipe_size)
7544
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
7545
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
7546
                     (idx, offset, wipe_size))
7547
        now = time.time()
7548
        offset += wipe_size
7549
        if now - last_output >= 60:
7550
          eta = _CalcEta(now - start_time, offset, size)
7551
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
7552
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
7553
          last_output = now
7554
  finally:
7555
    logging.info("Resume sync of instance %s disks", instance.name)
7556

    
7557
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
7558

    
7559
    for idx, success in enumerate(result.payload):
7560
      if not success:
7561
        lu.LogWarning("Resume sync of disk %d failed, please have a"
7562
                      " look at the status and troubleshoot the issue", idx)
7563
        logging.warn("resume-sync of instance %s for disks %d failed",
7564
                     instance.name, idx)
7565

    
7566

    
7567
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
7568
  """Create all disks for an instance.
7569

7570
  This abstracts away some work from AddInstance.
7571

7572
  @type lu: L{LogicalUnit}
7573
  @param lu: the logical unit on whose behalf we execute
7574
  @type instance: L{objects.Instance}
7575
  @param instance: the instance whose disks we should create
7576
  @type to_skip: list
7577
  @param to_skip: list of indices to skip
7578
  @type target_node: string
7579
  @param target_node: if passed, overrides the target node for creation
7580
  @rtype: boolean
7581
  @return: the success of the creation
7582

7583
  """
7584
  info = _GetInstanceInfoText(instance)
7585
  if target_node is None:
7586
    pnode = instance.primary_node
7587
    all_nodes = instance.all_nodes
7588
  else:
7589
    pnode = target_node
7590
    all_nodes = [pnode]
7591

    
7592
  if instance.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE):
7593
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
7594
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
7595

    
7596
    result.Raise("Failed to create directory '%s' on"
7597
                 " node %s" % (file_storage_dir, pnode))
7598

    
7599
  # Note: this needs to be kept in sync with adding of disks in
7600
  # LUInstanceSetParams
7601
  for idx, device in enumerate(instance.disks):
7602
    if to_skip and idx in to_skip:
7603
      continue
7604
    logging.info("Creating volume %s for instance %s",
7605
                 device.iv_name, instance.name)
7606
    #HARDCODE
7607
    for node in all_nodes:
7608
      f_create = node == pnode
7609
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
7610

    
7611

    
7612
def _RemoveDisks(lu, instance, target_node=None):
7613
  """Remove all disks for an instance.
7614

7615
  This abstracts away some work from `AddInstance()` and
7616
  `RemoveInstance()`. Note that in case some of the devices couldn't
7617
  be removed, the removal will continue with the other ones (compare
7618
  with `_CreateDisks()`).
7619

7620
  @type lu: L{LogicalUnit}
7621
  @param lu: the logical unit on whose behalf we execute
7622
  @type instance: L{objects.Instance}
7623
  @param instance: the instance whose disks we should remove
7624
  @type target_node: string
7625
  @param target_node: used to override the node on which to remove the disks
7626
  @rtype: boolean
7627
  @return: the success of the removal
7628

7629
  """
7630
  logging.info("Removing block devices for instance %s", instance.name)
7631

    
7632
  all_result = True
7633
  for device in instance.disks:
7634
    if target_node:
7635
      edata = [(target_node, device)]
7636
    else:
7637
      edata = device.ComputeNodeTree(instance.primary_node)
7638
    for node, disk in edata:
7639
      lu.cfg.SetDiskID(disk, node)
7640
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
7641
      if msg:
7642
        lu.LogWarning("Could not remove block device %s on node %s,"
7643
                      " continuing anyway: %s", device.iv_name, node, msg)
7644
        all_result = False
7645

    
7646
  if instance.disk_template == constants.DT_FILE:
7647
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
7648
    if target_node:
7649
      tgt = target_node
7650
    else:
7651
      tgt = instance.primary_node
7652
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
7653
    if result.fail_msg:
7654
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
7655
                    file_storage_dir, instance.primary_node, result.fail_msg)
7656
      all_result = False
7657

    
7658
  return all_result
7659

    
7660

    
7661
def _ComputeDiskSizePerVG(disk_template, disks):
7662
  """Compute disk size requirements in the volume group
7663

7664
  """
7665
  def _compute(disks, payload):
7666
    """Universal algorithm.
7667

7668
    """
7669
    vgs = {}
7670
    for disk in disks:
7671
      vgs[disk[constants.IDISK_VG]] = \
7672
        vgs.get(constants.IDISK_VG, 0) + disk[constants.IDISK_SIZE] + payload
7673

    
7674
    return vgs
7675

    
7676
  # Required free disk space as a function of disk and swap space
7677
  req_size_dict = {
7678
    constants.DT_DISKLESS: {},
7679
    constants.DT_PLAIN: _compute(disks, 0),
7680
    # 128 MB are added for drbd metadata for each disk
7681
    constants.DT_DRBD8: _compute(disks, 128),
7682
    constants.DT_FILE: {},
7683
    constants.DT_SHARED_FILE: {},
7684
  }
7685

    
7686
  if disk_template not in req_size_dict:
7687
    raise errors.ProgrammerError("Disk template '%s' size requirement"
7688
                                 " is unknown" %  disk_template)
7689

    
7690
  return req_size_dict[disk_template]
7691

    
7692

    
7693
def _ComputeDiskSize(disk_template, disks):
7694
  """Compute disk size requirements in the volume group
7695

7696
  """
7697
  # Required free disk space as a function of disk and swap space
7698
  req_size_dict = {
7699
    constants.DT_DISKLESS: None,
7700
    constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
7701
    # 128 MB are added for drbd metadata for each disk
7702
    constants.DT_DRBD8: sum(d[constants.IDISK_SIZE] + 128 for d in disks),
7703
    constants.DT_FILE: None,
7704
    constants.DT_SHARED_FILE: 0,
7705
    constants.DT_BLOCK: 0,
7706
  }
7707

    
7708
  if disk_template not in req_size_dict:
7709
    raise errors.ProgrammerError("Disk template '%s' size requirement"
7710
                                 " is unknown" %  disk_template)
7711

    
7712
  return req_size_dict[disk_template]
7713

    
7714

    
7715
def _FilterVmNodes(lu, nodenames):
7716
  """Filters out non-vm_capable nodes from a list.
7717

7718
  @type lu: L{LogicalUnit}
7719
  @param lu: the logical unit for which we check
7720
  @type nodenames: list
7721
  @param nodenames: the list of nodes on which we should check
7722
  @rtype: list
7723
  @return: the list of vm-capable nodes
7724

7725
  """
7726
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
7727
  return [name for name in nodenames if name not in vm_nodes]
7728

    
7729

    
7730
def _CheckHVParams(lu, nodenames, hvname, hvparams):
7731
  """Hypervisor parameter validation.
7732

7733
  This function abstract the hypervisor parameter validation to be
7734
  used in both instance create and instance modify.
7735

7736
  @type lu: L{LogicalUnit}
7737
  @param lu: the logical unit for which we check
7738
  @type nodenames: list
7739
  @param nodenames: the list of nodes on which we should check
7740
  @type hvname: string
7741
  @param hvname: the name of the hypervisor we should use
7742
  @type hvparams: dict
7743
  @param hvparams: the parameters which we need to check
7744
  @raise errors.OpPrereqError: if the parameters are not valid
7745

7746
  """
7747
  nodenames = _FilterVmNodes(lu, nodenames)
7748
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
7749
                                                  hvname,
7750
                                                  hvparams)
7751
  for node in nodenames:
7752
    info = hvinfo[node]
7753
    if info.offline:
7754
      continue
7755
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
7756

    
7757

    
7758
def _CheckOSParams(lu, required, nodenames, osname, osparams):
7759
  """OS parameters validation.
7760

7761
  @type lu: L{LogicalUnit}
7762
  @param lu: the logical unit for which we check
7763
  @type required: boolean
7764
  @param required: whether the validation should fail if the OS is not
7765
      found
7766
  @type nodenames: list
7767
  @param nodenames: the list of nodes on which we should check
7768
  @type osname: string
7769
  @param osname: the name of the hypervisor we should use
7770
  @type osparams: dict
7771
  @param osparams: the parameters which we need to check
7772
  @raise errors.OpPrereqError: if the parameters are not valid
7773

7774
  """
7775
  nodenames = _FilterVmNodes(lu, nodenames)
7776
  result = lu.rpc.call_os_validate(required, nodenames, osname,
7777
                                   [constants.OS_VALIDATE_PARAMETERS],
7778
                                   osparams)
7779
  for node, nres in result.items():
7780
    # we don't check for offline cases since this should be run only
7781
    # against the master node and/or an instance's nodes
7782
    nres.Raise("OS Parameters validation failed on node %s" % node)
7783
    if not nres.payload:
7784
      lu.LogInfo("OS %s not found on node %s, validation skipped",
7785
                 osname, node)
7786

    
7787

    
7788
class LUInstanceCreate(LogicalUnit):
7789
  """Create an instance.
7790

7791
  """
7792
  HPATH = "instance-add"
7793
  HTYPE = constants.HTYPE_INSTANCE
7794
  REQ_BGL = False
7795

    
7796
  def CheckArguments(self):
7797
    """Check arguments.
7798

7799
    """
7800
    # do not require name_check to ease forward/backward compatibility
7801
    # for tools
7802
    if self.op.no_install and self.op.start:
7803
      self.LogInfo("No-installation mode selected, disabling startup")
7804
      self.op.start = False
7805
    # validate/normalize the instance name
7806
    self.op.instance_name = \
7807
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
7808

    
7809
    if self.op.ip_check and not self.op.name_check:
7810
      # TODO: make the ip check more flexible and not depend on the name check
7811
      raise errors.OpPrereqError("Cannot do IP address check without a name"
7812
                                 " check", errors.ECODE_INVAL)
7813

    
7814
    # check nics' parameter names
7815
    for nic in self.op.nics:
7816
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
7817

    
7818
    # check disks. parameter names and consistent adopt/no-adopt strategy
7819
    has_adopt = has_no_adopt = False
7820
    for disk in self.op.disks:
7821
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
7822
      if constants.IDISK_ADOPT in disk:
7823
        has_adopt = True
7824
      else:
7825
        has_no_adopt = True
7826
    if has_adopt and has_no_adopt:
7827
      raise errors.OpPrereqError("Either all disks are adopted or none is",
7828
                                 errors.ECODE_INVAL)
7829
    if has_adopt:
7830
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
7831
        raise errors.OpPrereqError("Disk adoption is not supported for the"
7832
                                   " '%s' disk template" %
7833
                                   self.op.disk_template,
7834
                                   errors.ECODE_INVAL)
7835
      if self.op.iallocator is not None:
7836
        raise errors.OpPrereqError("Disk adoption not allowed with an"
7837
                                   " iallocator script", errors.ECODE_INVAL)
7838
      if self.op.mode == constants.INSTANCE_IMPORT:
7839
        raise errors.OpPrereqError("Disk adoption not allowed for"
7840
                                   " instance import", errors.ECODE_INVAL)
7841
    else:
7842
      if self.op.disk_template in constants.DTS_MUST_ADOPT:
7843
        raise errors.OpPrereqError("Disk template %s requires disk adoption,"
7844
                                   " but no 'adopt' parameter given" %
7845
                                   self.op.disk_template,
7846
                                   errors.ECODE_INVAL)
7847

    
7848
    self.adopt_disks = has_adopt
7849

    
7850
    # instance name verification
7851
    if self.op.name_check:
7852
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
7853
      self.op.instance_name = self.hostname1.name
7854
      # used in CheckPrereq for ip ping check
7855
      self.check_ip = self.hostname1.ip
7856
    else:
7857
      self.check_ip = None
7858

    
7859
    # file storage checks
7860
    if (self.op.file_driver and
7861
        not self.op.file_driver in constants.FILE_DRIVER):
7862
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
7863
                                 self.op.file_driver, errors.ECODE_INVAL)
7864

    
7865
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
7866
      raise errors.OpPrereqError("File storage directory path not absolute",
7867
                                 errors.ECODE_INVAL)
7868

    
7869
    ### Node/iallocator related checks
7870
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
7871

    
7872
    if self.op.pnode is not None:
7873
      if self.op.disk_template in constants.DTS_INT_MIRROR:
7874
        if self.op.snode is None:
7875
          raise errors.OpPrereqError("The networked disk templates need"
7876
                                     " a mirror node", errors.ECODE_INVAL)
7877
      elif self.op.snode:
7878
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
7879
                        " template")
7880
        self.op.snode = None
7881

    
7882
    self._cds = _GetClusterDomainSecret()
7883

    
7884
    if self.op.mode == constants.INSTANCE_IMPORT:
7885
      # On import force_variant must be True, because if we forced it at
7886
      # initial install, our only chance when importing it back is that it
7887
      # works again!
7888
      self.op.force_variant = True
7889

    
7890
      if self.op.no_install:
7891
        self.LogInfo("No-installation mode has no effect during import")
7892

    
7893
    elif self.op.mode == constants.INSTANCE_CREATE:
7894
      if self.op.os_type is None:
7895
        raise errors.OpPrereqError("No guest OS specified",
7896
                                   errors.ECODE_INVAL)
7897
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
7898
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
7899
                                   " installation" % self.op.os_type,
7900
                                   errors.ECODE_STATE)
7901
      if self.op.disk_template is None:
7902
        raise errors.OpPrereqError("No disk template specified",
7903
                                   errors.ECODE_INVAL)
7904

    
7905
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7906
      # Check handshake to ensure both clusters have the same domain secret
7907
      src_handshake = self.op.source_handshake
7908
      if not src_handshake:
7909
        raise errors.OpPrereqError("Missing source handshake",
7910
                                   errors.ECODE_INVAL)
7911

    
7912
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
7913
                                                           src_handshake)
7914
      if errmsg:
7915
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
7916
                                   errors.ECODE_INVAL)
7917

    
7918
      # Load and check source CA
7919
      self.source_x509_ca_pem = self.op.source_x509_ca
7920
      if not self.source_x509_ca_pem:
7921
        raise errors.OpPrereqError("Missing source X509 CA",
7922
                                   errors.ECODE_INVAL)
7923

    
7924
      try:
7925
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
7926
                                                    self._cds)
7927
      except OpenSSL.crypto.Error, err:
7928
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
7929
                                   (err, ), errors.ECODE_INVAL)
7930

    
7931
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
7932
      if errcode is not None:
7933
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
7934
                                   errors.ECODE_INVAL)
7935

    
7936
      self.source_x509_ca = cert
7937

    
7938
      src_instance_name = self.op.source_instance_name
7939
      if not src_instance_name:
7940
        raise errors.OpPrereqError("Missing source instance name",
7941
                                   errors.ECODE_INVAL)
7942

    
7943
      self.source_instance_name = \
7944
          netutils.GetHostname(name=src_instance_name).name
7945

    
7946
    else:
7947
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
7948
                                 self.op.mode, errors.ECODE_INVAL)
7949

    
7950
  def ExpandNames(self):
7951
    """ExpandNames for CreateInstance.
7952

7953
    Figure out the right locks for instance creation.
7954

7955
    """
7956
    self.needed_locks = {}
7957

    
7958
    instance_name = self.op.instance_name
7959
    # this is just a preventive check, but someone might still add this
7960
    # instance in the meantime, and creation will fail at lock-add time
7961
    if instance_name in self.cfg.GetInstanceList():
7962
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7963
                                 instance_name, errors.ECODE_EXISTS)
7964

    
7965
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
7966

    
7967
    if self.op.iallocator:
7968
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7969
    else:
7970
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
7971
      nodelist = [self.op.pnode]
7972
      if self.op.snode is not None:
7973
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
7974
        nodelist.append(self.op.snode)
7975
      self.needed_locks[locking.LEVEL_NODE] = nodelist
7976

    
7977
    # in case of import lock the source node too
7978
    if self.op.mode == constants.INSTANCE_IMPORT:
7979
      src_node = self.op.src_node
7980
      src_path = self.op.src_path
7981

    
7982
      if src_path is None:
7983
        self.op.src_path = src_path = self.op.instance_name
7984

    
7985
      if src_node is None:
7986
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7987
        self.op.src_node = None
7988
        if os.path.isabs(src_path):
7989
          raise errors.OpPrereqError("Importing an instance from an absolute"
7990
                                     " path requires a source node option",
7991
                                     errors.ECODE_INVAL)
7992
      else:
7993
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
7994
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
7995
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
7996
        if not os.path.isabs(src_path):
7997
          self.op.src_path = src_path = \
7998
            utils.PathJoin(constants.EXPORT_DIR, src_path)
7999

    
8000
  def _RunAllocator(self):
8001
    """Run the allocator based on input opcode.
8002

8003
    """
8004
    nics = [n.ToDict() for n in self.nics]
8005
    ial = IAllocator(self.cfg, self.rpc,
8006
                     mode=constants.IALLOCATOR_MODE_ALLOC,
8007
                     name=self.op.instance_name,
8008
                     disk_template=self.op.disk_template,
8009
                     tags=[],
8010
                     os=self.op.os_type,
8011
                     vcpus=self.be_full[constants.BE_VCPUS],
8012
                     mem_size=self.be_full[constants.BE_MEMORY],
8013
                     disks=self.disks,
8014
                     nics=nics,
8015
                     hypervisor=self.op.hypervisor,
8016
                     )
8017

    
8018
    ial.Run(self.op.iallocator)
8019

    
8020
    if not ial.success:
8021
      raise errors.OpPrereqError("Can't compute nodes using"
8022
                                 " iallocator '%s': %s" %
8023
                                 (self.op.iallocator, ial.info),
8024
                                 errors.ECODE_NORES)
8025
    if len(ial.result) != ial.required_nodes:
8026
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8027
                                 " of nodes (%s), required %s" %
8028
                                 (self.op.iallocator, len(ial.result),
8029
                                  ial.required_nodes), errors.ECODE_FAULT)
8030
    self.op.pnode = ial.result[0]
8031
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
8032
                 self.op.instance_name, self.op.iallocator,
8033
                 utils.CommaJoin(ial.result))
8034
    if ial.required_nodes == 2:
8035
      self.op.snode = ial.result[1]
8036

    
8037
  def BuildHooksEnv(self):
8038
    """Build hooks env.
8039

8040
    This runs on master, primary and secondary nodes of the instance.
8041

8042
    """
8043
    env = {
8044
      "ADD_MODE": self.op.mode,
8045
      }
8046
    if self.op.mode == constants.INSTANCE_IMPORT:
8047
      env["SRC_NODE"] = self.op.src_node
8048
      env["SRC_PATH"] = self.op.src_path
8049
      env["SRC_IMAGES"] = self.src_images
8050

    
8051
    env.update(_BuildInstanceHookEnv(
8052
      name=self.op.instance_name,
8053
      primary_node=self.op.pnode,
8054
      secondary_nodes=self.secondaries,
8055
      status=self.op.start,
8056
      os_type=self.op.os_type,
8057
      memory=self.be_full[constants.BE_MEMORY],
8058
      vcpus=self.be_full[constants.BE_VCPUS],
8059
      nics=_NICListToTuple(self, self.nics),
8060
      disk_template=self.op.disk_template,
8061
      disks=[(d[constants.IDISK_SIZE], d[constants.IDISK_MODE])
8062
             for d in self.disks],
8063
      bep=self.be_full,
8064
      hvp=self.hv_full,
8065
      hypervisor_name=self.op.hypervisor,
8066
    ))
8067

    
8068
    return env
8069

    
8070
  def BuildHooksNodes(self):
8071
    """Build hooks nodes.
8072

8073
    """
8074
    nl = [self.cfg.GetMasterNode(), self.op.pnode] + self.secondaries
8075
    return nl, nl
8076

    
8077
  def _ReadExportInfo(self):
8078
    """Reads the export information from disk.
8079

8080
    It will override the opcode source node and path with the actual
8081
    information, if these two were not specified before.
8082

8083
    @return: the export information
8084

8085
    """
8086
    assert self.op.mode == constants.INSTANCE_IMPORT
8087

    
8088
    src_node = self.op.src_node
8089
    src_path = self.op.src_path
8090

    
8091
    if src_node is None:
8092
      locked_nodes = self.glm.list_owned(locking.LEVEL_NODE)
8093
      exp_list = self.rpc.call_export_list(locked_nodes)
8094
      found = False
8095
      for node in exp_list:
8096
        if exp_list[node].fail_msg:
8097
          continue
8098
        if src_path in exp_list[node].payload:
8099
          found = True
8100
          self.op.src_node = src_node = node
8101
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
8102
                                                       src_path)
8103
          break
8104
      if not found:
8105
        raise errors.OpPrereqError("No export found for relative path %s" %
8106
                                    src_path, errors.ECODE_INVAL)
8107

    
8108
    _CheckNodeOnline(self, src_node)
8109
    result = self.rpc.call_export_info(src_node, src_path)
8110
    result.Raise("No export or invalid export found in dir %s" % src_path)
8111

    
8112
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
8113
    if not export_info.has_section(constants.INISECT_EXP):
8114
      raise errors.ProgrammerError("Corrupted export config",
8115
                                   errors.ECODE_ENVIRON)
8116

    
8117
    ei_version = export_info.get(constants.INISECT_EXP, "version")
8118
    if (int(ei_version) != constants.EXPORT_VERSION):
8119
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
8120
                                 (ei_version, constants.EXPORT_VERSION),
8121
                                 errors.ECODE_ENVIRON)
8122
    return export_info
8123

    
8124
  def _ReadExportParams(self, einfo):
8125
    """Use export parameters as defaults.
8126

8127
    In case the opcode doesn't specify (as in override) some instance
8128
    parameters, then try to use them from the export information, if
8129
    that declares them.
8130

8131
    """
8132
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
8133

    
8134
    if self.op.disk_template is None:
8135
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
8136
        self.op.disk_template = einfo.get(constants.INISECT_INS,
8137
                                          "disk_template")
8138
      else:
8139
        raise errors.OpPrereqError("No disk template specified and the export"
8140
                                   " is missing the disk_template information",
8141
                                   errors.ECODE_INVAL)
8142

    
8143
    if not self.op.disks:
8144
      if einfo.has_option(constants.INISECT_INS, "disk_count"):
8145
        disks = []
8146
        # TODO: import the disk iv_name too
8147
        for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
8148
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
8149
          disks.append({constants.IDISK_SIZE: disk_sz})
8150
        self.op.disks = disks
8151
      else:
8152
        raise errors.OpPrereqError("No disk info specified and the export"
8153
                                   " is missing the disk information",
8154
                                   errors.ECODE_INVAL)
8155

    
8156
    if (not self.op.nics and
8157
        einfo.has_option(constants.INISECT_INS, "nic_count")):
8158
      nics = []
8159
      for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
8160
        ndict = {}
8161
        for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
8162
          v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
8163
          ndict[name] = v
8164
        nics.append(ndict)
8165
      self.op.nics = nics
8166

    
8167
    if (self.op.hypervisor is None and
8168
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
8169
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
8170
    if einfo.has_section(constants.INISECT_HYP):
8171
      # use the export parameters but do not override the ones
8172
      # specified by the user
8173
      for name, value in einfo.items(constants.INISECT_HYP):
8174
        if name not in self.op.hvparams:
8175
          self.op.hvparams[name] = value
8176

    
8177
    if einfo.has_section(constants.INISECT_BEP):
8178
      # use the parameters, without overriding
8179
      for name, value in einfo.items(constants.INISECT_BEP):
8180
        if name not in self.op.beparams:
8181
          self.op.beparams[name] = value
8182
    else:
8183
      # try to read the parameters old style, from the main section
8184
      for name in constants.BES_PARAMETERS:
8185
        if (name not in self.op.beparams and
8186
            einfo.has_option(constants.INISECT_INS, name)):
8187
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
8188

    
8189
    if einfo.has_section(constants.INISECT_OSP):
8190
      # use the parameters, without overriding
8191
      for name, value in einfo.items(constants.INISECT_OSP):
8192
        if name not in self.op.osparams:
8193
          self.op.osparams[name] = value
8194

    
8195
  def _RevertToDefaults(self, cluster):
8196
    """Revert the instance parameters to the default values.
8197

8198
    """
8199
    # hvparams
8200
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
8201
    for name in self.op.hvparams.keys():
8202
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
8203
        del self.op.hvparams[name]
8204
    # beparams
8205
    be_defs = cluster.SimpleFillBE({})
8206
    for name in self.op.beparams.keys():
8207
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
8208
        del self.op.beparams[name]
8209
    # nic params
8210
    nic_defs = cluster.SimpleFillNIC({})
8211
    for nic in self.op.nics:
8212
      for name in constants.NICS_PARAMETERS:
8213
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
8214
          del nic[name]
8215
    # osparams
8216
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
8217
    for name in self.op.osparams.keys():
8218
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
8219
        del self.op.osparams[name]
8220

    
8221
  def CheckPrereq(self):
8222
    """Check prerequisites.
8223

8224
    """
8225
    if self.op.mode == constants.INSTANCE_IMPORT:
8226
      export_info = self._ReadExportInfo()
8227
      self._ReadExportParams(export_info)
8228

    
8229
    if (not self.cfg.GetVGName() and
8230
        self.op.disk_template not in constants.DTS_NOT_LVM):
8231
      raise errors.OpPrereqError("Cluster does not support lvm-based"
8232
                                 " instances", errors.ECODE_STATE)
8233

    
8234
    if self.op.hypervisor is None:
8235
      self.op.hypervisor = self.cfg.GetHypervisorType()
8236

    
8237
    cluster = self.cfg.GetClusterInfo()
8238
    enabled_hvs = cluster.enabled_hypervisors
8239
    if self.op.hypervisor not in enabled_hvs:
8240
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
8241
                                 " cluster (%s)" % (self.op.hypervisor,
8242
                                  ",".join(enabled_hvs)),
8243
                                 errors.ECODE_STATE)
8244

    
8245
    # check hypervisor parameter syntax (locally)
8246
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
8247
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
8248
                                      self.op.hvparams)
8249
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
8250
    hv_type.CheckParameterSyntax(filled_hvp)
8251
    self.hv_full = filled_hvp
8252
    # check that we don't specify global parameters on an instance
8253
    _CheckGlobalHvParams(self.op.hvparams)
8254

    
8255
    # fill and remember the beparams dict
8256
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
8257
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
8258

    
8259
    # build os parameters
8260
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
8261

    
8262
    # now that hvp/bep are in final format, let's reset to defaults,
8263
    # if told to do so
8264
    if self.op.identify_defaults:
8265
      self._RevertToDefaults(cluster)
8266

    
8267
    # NIC buildup
8268
    self.nics = []
8269
    for idx, nic in enumerate(self.op.nics):
8270
      nic_mode_req = nic.get(constants.INIC_MODE, None)
8271
      nic_mode = nic_mode_req
8272
      if nic_mode is None:
8273
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
8274

    
8275
      # in routed mode, for the first nic, the default ip is 'auto'
8276
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
8277
        default_ip_mode = constants.VALUE_AUTO
8278
      else:
8279
        default_ip_mode = constants.VALUE_NONE
8280

    
8281
      # ip validity checks
8282
      ip = nic.get(constants.INIC_IP, default_ip_mode)
8283
      if ip is None or ip.lower() == constants.VALUE_NONE:
8284
        nic_ip = None
8285
      elif ip.lower() == constants.VALUE_AUTO:
8286
        if not self.op.name_check:
8287
          raise errors.OpPrereqError("IP address set to auto but name checks"
8288
                                     " have been skipped",
8289
                                     errors.ECODE_INVAL)
8290
        nic_ip = self.hostname1.ip
8291
      else:
8292
        if not netutils.IPAddress.IsValid(ip):
8293
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
8294
                                     errors.ECODE_INVAL)
8295
        nic_ip = ip
8296

    
8297
      # TODO: check the ip address for uniqueness
8298
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
8299
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
8300
                                   errors.ECODE_INVAL)
8301

    
8302
      # MAC address verification
8303
      mac = nic.get(constants.INIC_MAC, constants.VALUE_AUTO)
8304
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
8305
        mac = utils.NormalizeAndValidateMac(mac)
8306

    
8307
        try:
8308
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
8309
        except errors.ReservationError:
8310
          raise errors.OpPrereqError("MAC address %s already in use"
8311
                                     " in cluster" % mac,
8312
                                     errors.ECODE_NOTUNIQUE)
8313

    
8314
      #  Build nic parameters
8315
      link = nic.get(constants.INIC_LINK, None)
8316
      nicparams = {}
8317
      if nic_mode_req:
8318
        nicparams[constants.NIC_MODE] = nic_mode_req
8319
      if link:
8320
        nicparams[constants.NIC_LINK] = link
8321

    
8322
      check_params = cluster.SimpleFillNIC(nicparams)
8323
      objects.NIC.CheckParameterSyntax(check_params)
8324
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
8325

    
8326
    # disk checks/pre-build
8327
    default_vg = self.cfg.GetVGName()
8328
    self.disks = []
8329
    for disk in self.op.disks:
8330
      mode = disk.get(constants.IDISK_MODE, constants.DISK_RDWR)
8331
      if mode not in constants.DISK_ACCESS_SET:
8332
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
8333
                                   mode, errors.ECODE_INVAL)
8334
      size = disk.get(constants.IDISK_SIZE, None)
8335
      if size is None:
8336
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
8337
      try:
8338
        size = int(size)
8339
      except (TypeError, ValueError):
8340
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
8341
                                   errors.ECODE_INVAL)
8342

    
8343
      data_vg = disk.get(constants.IDISK_VG, default_vg)
8344
      new_disk = {
8345
        constants.IDISK_SIZE: size,
8346
        constants.IDISK_MODE: mode,
8347
        constants.IDISK_VG: data_vg,
8348
        constants.IDISK_METAVG: disk.get(constants.IDISK_METAVG, data_vg),
8349
        }
8350
      if constants.IDISK_ADOPT in disk:
8351
        new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
8352
      self.disks.append(new_disk)
8353

    
8354
    if self.op.mode == constants.INSTANCE_IMPORT:
8355

    
8356
      # Check that the new instance doesn't have less disks than the export
8357
      instance_disks = len(self.disks)
8358
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
8359
      if instance_disks < export_disks:
8360
        raise errors.OpPrereqError("Not enough disks to import."
8361
                                   " (instance: %d, export: %d)" %
8362
                                   (instance_disks, export_disks),
8363
                                   errors.ECODE_INVAL)
8364

    
8365
      disk_images = []
8366
      for idx in range(export_disks):
8367
        option = 'disk%d_dump' % idx
8368
        if export_info.has_option(constants.INISECT_INS, option):
8369
          # FIXME: are the old os-es, disk sizes, etc. useful?
8370
          export_name = export_info.get(constants.INISECT_INS, option)
8371
          image = utils.PathJoin(self.op.src_path, export_name)
8372
          disk_images.append(image)
8373
        else:
8374
          disk_images.append(False)
8375

    
8376
      self.src_images = disk_images
8377

    
8378
      old_name = export_info.get(constants.INISECT_INS, 'name')
8379
      try:
8380
        exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
8381
      except (TypeError, ValueError), err:
8382
        raise errors.OpPrereqError("Invalid export file, nic_count is not"
8383
                                   " an integer: %s" % str(err),
8384
                                   errors.ECODE_STATE)
8385
      if self.op.instance_name == old_name:
8386
        for idx, nic in enumerate(self.nics):
8387
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
8388
            nic_mac_ini = 'nic%d_mac' % idx
8389
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
8390

    
8391
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
8392

    
8393
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
8394
    if self.op.ip_check:
8395
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
8396
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
8397
                                   (self.check_ip, self.op.instance_name),
8398
                                   errors.ECODE_NOTUNIQUE)
8399

    
8400
    #### mac address generation
8401
    # By generating here the mac address both the allocator and the hooks get
8402
    # the real final mac address rather than the 'auto' or 'generate' value.
8403
    # There is a race condition between the generation and the instance object
8404
    # creation, which means that we know the mac is valid now, but we're not
8405
    # sure it will be when we actually add the instance. If things go bad
8406
    # adding the instance will abort because of a duplicate mac, and the
8407
    # creation job will fail.
8408
    for nic in self.nics:
8409
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
8410
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
8411

    
8412
    #### allocator run
8413

    
8414
    if self.op.iallocator is not None:
8415
      self._RunAllocator()
8416

    
8417
    #### node related checks
8418

    
8419
    # check primary node
8420
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
8421
    assert self.pnode is not None, \
8422
      "Cannot retrieve locked node %s" % self.op.pnode
8423
    if pnode.offline:
8424
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
8425
                                 pnode.name, errors.ECODE_STATE)
8426
    if pnode.drained:
8427
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
8428
                                 pnode.name, errors.ECODE_STATE)
8429
    if not pnode.vm_capable:
8430
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
8431
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
8432

    
8433
    self.secondaries = []
8434

    
8435
    # mirror node verification
8436
    if self.op.disk_template in constants.DTS_INT_MIRROR:
8437
      if self.op.snode == pnode.name:
8438
        raise errors.OpPrereqError("The secondary node cannot be the"
8439
                                   " primary node", errors.ECODE_INVAL)
8440
      _CheckNodeOnline(self, self.op.snode)
8441
      _CheckNodeNotDrained(self, self.op.snode)
8442
      _CheckNodeVmCapable(self, self.op.snode)
8443
      self.secondaries.append(self.op.snode)
8444

    
8445
    nodenames = [pnode.name] + self.secondaries
8446

    
8447
    if not self.adopt_disks:
8448
      # Check lv size requirements, if not adopting
8449
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
8450
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
8451

    
8452
    elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
8453
      all_lvs = set(["%s/%s" % (disk[constants.IDISK_VG],
8454
                                disk[constants.IDISK_ADOPT])
8455
                     for disk in self.disks])
8456
      if len(all_lvs) != len(self.disks):
8457
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
8458
                                   errors.ECODE_INVAL)
8459
      for lv_name in all_lvs:
8460
        try:
8461
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
8462
          # to ReserveLV uses the same syntax
8463
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
8464
        except errors.ReservationError:
8465
          raise errors.OpPrereqError("LV named %s used by another instance" %
8466
                                     lv_name, errors.ECODE_NOTUNIQUE)
8467

    
8468
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
8469
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
8470

    
8471
      node_lvs = self.rpc.call_lv_list([pnode.name],
8472
                                       vg_names.payload.keys())[pnode.name]
8473
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
8474
      node_lvs = node_lvs.payload
8475

    
8476
      delta = all_lvs.difference(node_lvs.keys())
8477
      if delta:
8478
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
8479
                                   utils.CommaJoin(delta),
8480
                                   errors.ECODE_INVAL)
8481
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
8482
      if online_lvs:
8483
        raise errors.OpPrereqError("Online logical volumes found, cannot"
8484
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
8485
                                   errors.ECODE_STATE)
8486
      # update the size of disk based on what is found
8487
      for dsk in self.disks:
8488
        dsk[constants.IDISK_SIZE] = \
8489
          int(float(node_lvs["%s/%s" % (dsk[constants.IDISK_VG],
8490
                                        dsk[constants.IDISK_ADOPT])][0]))
8491

    
8492
    elif self.op.disk_template == constants.DT_BLOCK:
8493
      # Normalize and de-duplicate device paths
8494
      all_disks = set([os.path.abspath(disk[constants.IDISK_ADOPT])
8495
                       for disk in self.disks])
8496
      if len(all_disks) != len(self.disks):
8497
        raise errors.OpPrereqError("Duplicate disk names given for adoption",
8498
                                   errors.ECODE_INVAL)
8499
      baddisks = [d for d in all_disks
8500
                  if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
8501
      if baddisks:
8502
        raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
8503
                                   " cannot be adopted" %
8504
                                   (", ".join(baddisks),
8505
                                    constants.ADOPTABLE_BLOCKDEV_ROOT),
8506
                                   errors.ECODE_INVAL)
8507

    
8508
      node_disks = self.rpc.call_bdev_sizes([pnode.name],
8509
                                            list(all_disks))[pnode.name]
8510
      node_disks.Raise("Cannot get block device information from node %s" %
8511
                       pnode.name)
8512
      node_disks = node_disks.payload
8513
      delta = all_disks.difference(node_disks.keys())
8514
      if delta:
8515
        raise errors.OpPrereqError("Missing block device(s): %s" %
8516
                                   utils.CommaJoin(delta),
8517
                                   errors.ECODE_INVAL)
8518
      for dsk in self.disks:
8519
        dsk[constants.IDISK_SIZE] = \
8520
          int(float(node_disks[dsk[constants.IDISK_ADOPT]]))
8521

    
8522
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
8523

    
8524
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
8525
    # check OS parameters (remotely)
8526
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
8527

    
8528
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
8529

    
8530
    # memory check on primary node
8531
    if self.op.start:
8532
      _CheckNodeFreeMemory(self, self.pnode.name,
8533
                           "creating instance %s" % self.op.instance_name,
8534
                           self.be_full[constants.BE_MEMORY],
8535
                           self.op.hypervisor)
8536

    
8537
    self.dry_run_result = list(nodenames)
8538

    
8539
  def Exec(self, feedback_fn):
8540
    """Create and add the instance to the cluster.
8541

8542
    """
8543
    instance = self.op.instance_name
8544
    pnode_name = self.pnode.name
8545

    
8546
    ht_kind = self.op.hypervisor
8547
    if ht_kind in constants.HTS_REQ_PORT:
8548
      network_port = self.cfg.AllocatePort()
8549
    else:
8550
      network_port = None
8551

    
8552
    if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE:
8553
      # this is needed because os.path.join does not accept None arguments
8554
      if self.op.file_storage_dir is None:
8555
        string_file_storage_dir = ""
8556
      else:
8557
        string_file_storage_dir = self.op.file_storage_dir
8558

    
8559
      # build the full file storage dir path
8560
      if self.op.disk_template == constants.DT_SHARED_FILE:
8561
        get_fsd_fn = self.cfg.GetSharedFileStorageDir
8562
      else:
8563
        get_fsd_fn = self.cfg.GetFileStorageDir
8564

    
8565
      file_storage_dir = utils.PathJoin(get_fsd_fn(),
8566
                                        string_file_storage_dir, instance)
8567
    else:
8568
      file_storage_dir = ""
8569

    
8570
    disks = _GenerateDiskTemplate(self,
8571
                                  self.op.disk_template,
8572
                                  instance, pnode_name,
8573
                                  self.secondaries,
8574
                                  self.disks,
8575
                                  file_storage_dir,
8576
                                  self.op.file_driver,
8577
                                  0,
8578
                                  feedback_fn)
8579

    
8580
    iobj = objects.Instance(name=instance, os=self.op.os_type,
8581
                            primary_node=pnode_name,
8582
                            nics=self.nics, disks=disks,
8583
                            disk_template=self.op.disk_template,
8584
                            admin_up=False,
8585
                            network_port=network_port,
8586
                            beparams=self.op.beparams,
8587
                            hvparams=self.op.hvparams,
8588
                            hypervisor=self.op.hypervisor,
8589
                            osparams=self.op.osparams,
8590
                            )
8591

    
8592
    if self.adopt_disks:
8593
      if self.op.disk_template == constants.DT_PLAIN:
8594
        # rename LVs to the newly-generated names; we need to construct
8595
        # 'fake' LV disks with the old data, plus the new unique_id
8596
        tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
8597
        rename_to = []
8598
        for t_dsk, a_dsk in zip (tmp_disks, self.disks):
8599
          rename_to.append(t_dsk.logical_id)
8600
          t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
8601
          self.cfg.SetDiskID(t_dsk, pnode_name)
8602
        result = self.rpc.call_blockdev_rename(pnode_name,
8603
                                               zip(tmp_disks, rename_to))
8604
        result.Raise("Failed to rename adoped LVs")
8605
    else:
8606
      feedback_fn("* creating instance disks...")
8607
      try:
8608
        _CreateDisks(self, iobj)
8609
      except errors.OpExecError:
8610
        self.LogWarning("Device creation failed, reverting...")
8611
        try:
8612
          _RemoveDisks(self, iobj)
8613
        finally:
8614
          self.cfg.ReleaseDRBDMinors(instance)
8615
          raise
8616

    
8617
    feedback_fn("adding instance %s to cluster config" % instance)
8618

    
8619
    self.cfg.AddInstance(iobj, self.proc.GetECId())
8620

    
8621
    # Declare that we don't want to remove the instance lock anymore, as we've
8622
    # added the instance to the config
8623
    del self.remove_locks[locking.LEVEL_INSTANCE]
8624

    
8625
    if self.op.mode == constants.INSTANCE_IMPORT:
8626
      # Release unused nodes
8627
      _ReleaseLocks(self, locking.LEVEL_NODE, keep=[self.op.src_node])
8628
    else:
8629
      # Release all nodes
8630
      _ReleaseLocks(self, locking.LEVEL_NODE)
8631

    
8632
    disk_abort = False
8633
    if not self.adopt_disks and self.cfg.GetClusterInfo().prealloc_wipe_disks:
8634
      feedback_fn("* wiping instance disks...")
8635
      try:
8636
        _WipeDisks(self, iobj)
8637
      except errors.OpExecError, err:
8638
        logging.exception("Wiping disks failed")
8639
        self.LogWarning("Wiping instance disks failed (%s)", err)
8640
        disk_abort = True
8641

    
8642
    if disk_abort:
8643
      # Something is already wrong with the disks, don't do anything else
8644
      pass
8645
    elif self.op.wait_for_sync:
8646
      disk_abort = not _WaitForSync(self, iobj)
8647
    elif iobj.disk_template in constants.DTS_INT_MIRROR:
8648
      # make sure the disks are not degraded (still sync-ing is ok)
8649
      time.sleep(15)
8650
      feedback_fn("* checking mirrors status")
8651
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
8652
    else:
8653
      disk_abort = False
8654

    
8655
    if disk_abort:
8656
      _RemoveDisks(self, iobj)
8657
      self.cfg.RemoveInstance(iobj.name)
8658
      # Make sure the instance lock gets removed
8659
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
8660
      raise errors.OpExecError("There are some degraded disks for"
8661
                               " this instance")
8662

    
8663
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
8664
      if self.op.mode == constants.INSTANCE_CREATE:
8665
        if not self.op.no_install:
8666
          feedback_fn("* running the instance OS create scripts...")
8667
          # FIXME: pass debug option from opcode to backend
8668
          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
8669
                                                 self.op.debug_level)
8670
          result.Raise("Could not add os for instance %s"
8671
                       " on node %s" % (instance, pnode_name))
8672

    
8673
      elif self.op.mode == constants.INSTANCE_IMPORT:
8674
        feedback_fn("* running the instance OS import scripts...")
8675

    
8676
        transfers = []
8677

    
8678
        for idx, image in enumerate(self.src_images):
8679
          if not image:
8680
            continue
8681

    
8682
          # FIXME: pass debug option from opcode to backend
8683
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
8684
                                             constants.IEIO_FILE, (image, ),
8685
                                             constants.IEIO_SCRIPT,
8686
                                             (iobj.disks[idx], idx),
8687
                                             None)
8688
          transfers.append(dt)
8689

    
8690
        import_result = \
8691
          masterd.instance.TransferInstanceData(self, feedback_fn,
8692
                                                self.op.src_node, pnode_name,
8693
                                                self.pnode.secondary_ip,
8694
                                                iobj, transfers)
8695
        if not compat.all(import_result):
8696
          self.LogWarning("Some disks for instance %s on node %s were not"
8697
                          " imported successfully" % (instance, pnode_name))
8698

    
8699
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
8700
        feedback_fn("* preparing remote import...")
8701
        # The source cluster will stop the instance before attempting to make a
8702
        # connection. In some cases stopping an instance can take a long time,
8703
        # hence the shutdown timeout is added to the connection timeout.
8704
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
8705
                           self.op.source_shutdown_timeout)
8706
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
8707

    
8708
        assert iobj.primary_node == self.pnode.name
8709
        disk_results = \
8710
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
8711
                                        self.source_x509_ca,
8712
                                        self._cds, timeouts)
8713
        if not compat.all(disk_results):
8714
          # TODO: Should the instance still be started, even if some disks
8715
          # failed to import (valid for local imports, too)?
8716
          self.LogWarning("Some disks for instance %s on node %s were not"
8717
                          " imported successfully" % (instance, pnode_name))
8718

    
8719
        # Run rename script on newly imported instance
8720
        assert iobj.name == instance
8721
        feedback_fn("Running rename script for %s" % instance)
8722
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
8723
                                                   self.source_instance_name,
8724
                                                   self.op.debug_level)
8725
        if result.fail_msg:
8726
          self.LogWarning("Failed to run rename script for %s on node"
8727
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
8728

    
8729
      else:
8730
        # also checked in the prereq part
8731
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
8732
                                     % self.op.mode)
8733

    
8734
    if self.op.start:
8735
      iobj.admin_up = True
8736
      self.cfg.Update(iobj, feedback_fn)
8737
      logging.info("Starting instance %s on node %s", instance, pnode_name)
8738
      feedback_fn("* starting instance...")
8739
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
8740
      result.Raise("Could not start instance")
8741

    
8742
    return list(iobj.all_nodes)
8743

    
8744

    
8745
class LUInstanceConsole(NoHooksLU):
8746
  """Connect to an instance's console.
8747

8748
  This is somewhat special in that it returns the command line that
8749
  you need to run on the master node in order to connect to the
8750
  console.
8751

8752
  """
8753
  REQ_BGL = False
8754

    
8755
  def ExpandNames(self):
8756
    self._ExpandAndLockInstance()
8757

    
8758
  def CheckPrereq(self):
8759
    """Check prerequisites.
8760

8761
    This checks that the instance is in the cluster.
8762

8763
    """
8764
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8765
    assert self.instance is not None, \
8766
      "Cannot retrieve locked instance %s" % self.op.instance_name
8767
    _CheckNodeOnline(self, self.instance.primary_node)
8768

    
8769
  def Exec(self, feedback_fn):
8770
    """Connect to the console of an instance
8771

8772
    """
8773
    instance = self.instance
8774
    node = instance.primary_node
8775

    
8776
    node_insts = self.rpc.call_instance_list([node],
8777
                                             [instance.hypervisor])[node]
8778
    node_insts.Raise("Can't get node information from %s" % node)
8779

    
8780
    if instance.name not in node_insts.payload:
8781
      if instance.admin_up:
8782
        state = constants.INSTST_ERRORDOWN
8783
      else:
8784
        state = constants.INSTST_ADMINDOWN
8785
      raise errors.OpExecError("Instance %s is not running (state %s)" %
8786
                               (instance.name, state))
8787

    
8788
    logging.debug("Connecting to console of %s on %s", instance.name, node)
8789

    
8790
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
8791

    
8792

    
8793
def _GetInstanceConsole(cluster, instance):
8794
  """Returns console information for an instance.
8795

8796
  @type cluster: L{objects.Cluster}
8797
  @type instance: L{objects.Instance}
8798
  @rtype: dict
8799

8800
  """
8801
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
8802
  # beparams and hvparams are passed separately, to avoid editing the
8803
  # instance and then saving the defaults in the instance itself.
8804
  hvparams = cluster.FillHV(instance)
8805
  beparams = cluster.FillBE(instance)
8806
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
8807

    
8808
  assert console.instance == instance.name
8809
  assert console.Validate()
8810

    
8811
  return console.ToDict()
8812

    
8813

    
8814
class LUInstanceReplaceDisks(LogicalUnit):
8815
  """Replace the disks of an instance.
8816

8817
  """
8818
  HPATH = "mirrors-replace"
8819
  HTYPE = constants.HTYPE_INSTANCE
8820
  REQ_BGL = False
8821

    
8822
  def CheckArguments(self):
8823
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
8824
                                  self.op.iallocator)
8825

    
8826
  def ExpandNames(self):
8827
    self._ExpandAndLockInstance()
8828

    
8829
    assert locking.LEVEL_NODE not in self.needed_locks
8830
    assert locking.LEVEL_NODEGROUP not in self.needed_locks
8831

    
8832
    assert self.op.iallocator is None or self.op.remote_node is None, \
8833
      "Conflicting options"
8834

    
8835
    if self.op.remote_node is not None:
8836
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8837

    
8838
      # Warning: do not remove the locking of the new secondary here
8839
      # unless DRBD8.AddChildren is changed to work in parallel;
8840
      # currently it doesn't since parallel invocations of
8841
      # FindUnusedMinor will conflict
8842
      self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
8843
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
8844
    else:
8845
      self.needed_locks[locking.LEVEL_NODE] = []
8846
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8847

    
8848
      if self.op.iallocator is not None:
8849
        # iallocator will select a new node in the same group
8850
        self.needed_locks[locking.LEVEL_NODEGROUP] = []
8851

    
8852
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
8853
                                   self.op.iallocator, self.op.remote_node,
8854
                                   self.op.disks, False, self.op.early_release)
8855

    
8856
    self.tasklets = [self.replacer]
8857

    
8858
  def DeclareLocks(self, level):
8859
    if level == locking.LEVEL_NODEGROUP:
8860
      assert self.op.remote_node is None
8861
      assert self.op.iallocator is not None
8862
      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
8863

    
8864
      self.share_locks[locking.LEVEL_NODEGROUP] = 1
8865
      self.needed_locks[locking.LEVEL_NODEGROUP] = \
8866
        self.cfg.GetInstanceNodeGroups(self.op.instance_name)
8867

    
8868
    elif level == locking.LEVEL_NODE:
8869
      if self.op.iallocator is not None:
8870
        assert self.op.remote_node is None
8871
        assert not self.needed_locks[locking.LEVEL_NODE]
8872

    
8873
        # Lock member nodes of all locked groups
8874
        self.needed_locks[locking.LEVEL_NODE] = [node_name
8875
          for group_uuid in self.glm.list_owned(locking.LEVEL_NODEGROUP)
8876
          for node_name in self.cfg.GetNodeGroup(group_uuid).members]
8877
      else:
8878
        self._LockInstancesNodes()
8879

    
8880
  def BuildHooksEnv(self):
8881
    """Build hooks env.
8882

8883
    This runs on the master, the primary and all the secondaries.
8884

8885
    """
8886
    instance = self.replacer.instance
8887
    env = {
8888
      "MODE": self.op.mode,
8889
      "NEW_SECONDARY": self.op.remote_node,
8890
      "OLD_SECONDARY": instance.secondary_nodes[0],
8891
      }
8892
    env.update(_BuildInstanceHookEnvByObject(self, instance))
8893
    return env
8894

    
8895
  def BuildHooksNodes(self):
8896
    """Build hooks nodes.
8897

8898
    """
8899
    instance = self.replacer.instance
8900
    nl = [
8901
      self.cfg.GetMasterNode(),
8902
      instance.primary_node,
8903
      ]
8904
    if self.op.remote_node is not None:
8905
      nl.append(self.op.remote_node)
8906
    return nl, nl
8907

    
8908
  def CheckPrereq(self):
8909
    """Check prerequisites.
8910

8911
    """
8912
    assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
8913
            self.op.iallocator is None)
8914

    
8915
    owned_groups = self.glm.list_owned(locking.LEVEL_NODEGROUP)
8916
    if owned_groups:
8917
      groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
8918
      if owned_groups != groups:
8919
        raise errors.OpExecError("Node groups used by instance '%s' changed"
8920
                                 " since lock was acquired, current list is %r,"
8921
                                 " used to be '%s'" %
8922
                                 (self.op.instance_name,
8923
                                  utils.CommaJoin(groups),
8924
                                  utils.CommaJoin(owned_groups)))
8925

    
8926
    return LogicalUnit.CheckPrereq(self)
8927

    
8928

    
8929
class TLReplaceDisks(Tasklet):
8930
  """Replaces disks for an instance.
8931

8932
  Note: Locking is not within the scope of this class.
8933

8934
  """
8935
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
8936
               disks, delay_iallocator, early_release):
8937
    """Initializes this class.
8938

8939
    """
8940
    Tasklet.__init__(self, lu)
8941

    
8942
    # Parameters
8943
    self.instance_name = instance_name
8944
    self.mode = mode
8945
    self.iallocator_name = iallocator_name
8946
    self.remote_node = remote_node
8947
    self.disks = disks
8948
    self.delay_iallocator = delay_iallocator
8949
    self.early_release = early_release
8950

    
8951
    # Runtime data
8952
    self.instance = None
8953
    self.new_node = None
8954
    self.target_node = None
8955
    self.other_node = None
8956
    self.remote_node_info = None
8957
    self.node_secondary_ip = None
8958

    
8959
  @staticmethod
8960
  def CheckArguments(mode, remote_node, iallocator):
8961
    """Helper function for users of this class.
8962

8963
    """
8964
    # check for valid parameter combination
8965
    if mode == constants.REPLACE_DISK_CHG:
8966
      if remote_node is None and iallocator is None:
8967
        raise errors.OpPrereqError("When changing the secondary either an"
8968
                                   " iallocator script must be used or the"
8969
                                   " new node given", errors.ECODE_INVAL)
8970

    
8971
      if remote_node is not None and iallocator is not None:
8972
        raise errors.OpPrereqError("Give either the iallocator or the new"
8973
                                   " secondary, not both", errors.ECODE_INVAL)
8974

    
8975
    elif remote_node is not None or iallocator is not None:
8976
      # Not replacing the secondary
8977
      raise errors.OpPrereqError("The iallocator and new node options can"
8978
                                 " only be used when changing the"
8979
                                 " secondary node", errors.ECODE_INVAL)
8980

    
8981
  @staticmethod
8982
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
8983
    """Compute a new secondary node using an IAllocator.
8984

8985
    """
8986
    ial = IAllocator(lu.cfg, lu.rpc,
8987
                     mode=constants.IALLOCATOR_MODE_RELOC,
8988
                     name=instance_name,
8989
                     relocate_from=relocate_from)
8990

    
8991
    ial.Run(iallocator_name)
8992

    
8993
    if not ial.success:
8994
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
8995
                                 " %s" % (iallocator_name, ial.info),
8996
                                 errors.ECODE_NORES)
8997

    
8998
    if len(ial.result) != ial.required_nodes:
8999
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
9000
                                 " of nodes (%s), required %s" %
9001
                                 (iallocator_name,
9002
                                  len(ial.result), ial.required_nodes),
9003
                                 errors.ECODE_FAULT)
9004

    
9005
    remote_node_name = ial.result[0]
9006

    
9007
    lu.LogInfo("Selected new secondary for instance '%s': %s",
9008
               instance_name, remote_node_name)
9009

    
9010
    return remote_node_name
9011

    
9012
  def _FindFaultyDisks(self, node_name):
9013
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
9014
                                    node_name, True)
9015

    
9016
  def _CheckDisksActivated(self, instance):
9017
    """Checks if the instance disks are activated.
9018

9019
    @param instance: The instance to check disks
9020
    @return: True if they are activated, False otherwise
9021

9022
    """
9023
    nodes = instance.all_nodes
9024

    
9025
    for idx, dev in enumerate(instance.disks):
9026
      for node in nodes:
9027
        self.lu.LogInfo("Checking disk/%d on %s", idx, node)
9028
        self.cfg.SetDiskID(dev, node)
9029

    
9030
        result = self.rpc.call_blockdev_find(node, dev)
9031

    
9032
        if result.offline:
9033
          continue
9034
        elif result.fail_msg or not result.payload:
9035
          return False
9036

    
9037
    return True
9038

    
9039
  def CheckPrereq(self):
9040
    """Check prerequisites.
9041

9042
    This checks that the instance is in the cluster.
9043

9044
    """
9045
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
9046
    assert instance is not None, \
9047
      "Cannot retrieve locked instance %s" % self.instance_name
9048

    
9049
    if instance.disk_template != constants.DT_DRBD8:
9050
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
9051
                                 " instances", errors.ECODE_INVAL)
9052

    
9053
    if len(instance.secondary_nodes) != 1:
9054
      raise errors.OpPrereqError("The instance has a strange layout,"
9055
                                 " expected one secondary but found %d" %
9056
                                 len(instance.secondary_nodes),
9057
                                 errors.ECODE_FAULT)
9058

    
9059
    if not self.delay_iallocator:
9060
      self._CheckPrereq2()
9061

    
9062
  def _CheckPrereq2(self):
9063
    """Check prerequisites, second part.
9064

9065
    This function should always be part of CheckPrereq. It was separated and is
9066
    now called from Exec because during node evacuation iallocator was only
9067
    called with an unmodified cluster model, not taking planned changes into
9068
    account.
9069

9070
    """
9071
    instance = self.instance
9072
    secondary_node = instance.secondary_nodes[0]
9073

    
9074
    if self.iallocator_name is None:
9075
      remote_node = self.remote_node
9076
    else:
9077
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
9078
                                       instance.name, instance.secondary_nodes)
9079

    
9080
    if remote_node is None:
9081
      self.remote_node_info = None
9082
    else:
9083
      assert remote_node in self.lu.glm.list_owned(locking.LEVEL_NODE), \
9084
             "Remote node '%s' is not locked" % remote_node
9085

    
9086
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
9087
      assert self.remote_node_info is not None, \
9088
        "Cannot retrieve locked node %s" % remote_node
9089

    
9090
    if remote_node == self.instance.primary_node:
9091
      raise errors.OpPrereqError("The specified node is the primary node of"
9092
                                 " the instance", errors.ECODE_INVAL)
9093

    
9094
    if remote_node == secondary_node:
9095
      raise errors.OpPrereqError("The specified node is already the"
9096
                                 " secondary node of the instance",
9097
                                 errors.ECODE_INVAL)
9098

    
9099
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
9100
                                    constants.REPLACE_DISK_CHG):
9101
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
9102
                                 errors.ECODE_INVAL)
9103

    
9104
    if self.mode == constants.REPLACE_DISK_AUTO:
9105
      if not self._CheckDisksActivated(instance):
9106
        raise errors.OpPrereqError("Please run activate-disks on instance %s"
9107
                                   " first" % self.instance_name,
9108
                                   errors.ECODE_STATE)
9109
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
9110
      faulty_secondary = self._FindFaultyDisks(secondary_node)
9111

    
9112
      if faulty_primary and faulty_secondary:
9113
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
9114
                                   " one node and can not be repaired"
9115
                                   " automatically" % self.instance_name,
9116
                                   errors.ECODE_STATE)
9117

    
9118
      if faulty_primary:
9119
        self.disks = faulty_primary
9120
        self.target_node = instance.primary_node
9121
        self.other_node = secondary_node
9122
        check_nodes = [self.target_node, self.other_node]
9123
      elif faulty_secondary:
9124
        self.disks = faulty_secondary
9125
        self.target_node = secondary_node
9126
        self.other_node = instance.primary_node
9127
        check_nodes = [self.target_node, self.other_node]
9128
      else:
9129
        self.disks = []
9130
        check_nodes = []
9131

    
9132
    else:
9133
      # Non-automatic modes
9134
      if self.mode == constants.REPLACE_DISK_PRI:
9135
        self.target_node = instance.primary_node
9136
        self.other_node = secondary_node
9137
        check_nodes = [self.target_node, self.other_node]
9138

    
9139
      elif self.mode == constants.REPLACE_DISK_SEC:
9140
        self.target_node = secondary_node
9141
        self.other_node = instance.primary_node
9142
        check_nodes = [self.target_node, self.other_node]
9143

    
9144
      elif self.mode == constants.REPLACE_DISK_CHG:
9145
        self.new_node = remote_node
9146
        self.other_node = instance.primary_node
9147
        self.target_node = secondary_node
9148
        check_nodes = [self.new_node, self.other_node]
9149

    
9150
        _CheckNodeNotDrained(self.lu, remote_node)
9151
        _CheckNodeVmCapable(self.lu, remote_node)
9152

    
9153
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
9154
        assert old_node_info is not None
9155
        if old_node_info.offline and not self.early_release:
9156
          # doesn't make sense to delay the release
9157
          self.early_release = True
9158
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
9159
                          " early-release mode", secondary_node)
9160

    
9161
      else:
9162
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
9163
                                     self.mode)
9164

    
9165
      # If not specified all disks should be replaced
9166
      if not self.disks:
9167
        self.disks = range(len(self.instance.disks))
9168

    
9169
    for node in check_nodes:
9170
      _CheckNodeOnline(self.lu, node)
9171

    
9172
    touched_nodes = frozenset(node_name for node_name in [self.new_node,
9173
                                                          self.other_node,
9174
                                                          self.target_node]
9175
                              if node_name is not None)
9176

    
9177
    # Release unneeded node locks
9178
    _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
9179

    
9180
    # Release any owned node group
9181
    if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
9182
      _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
9183

    
9184
    # Check whether disks are valid
9185
    for disk_idx in self.disks:
9186
      instance.FindDisk(disk_idx)
9187

    
9188
    # Get secondary node IP addresses
9189
    self.node_secondary_ip = \
9190
      dict((node_name, self.cfg.GetNodeInfo(node_name).secondary_ip)
9191
           for node_name in touched_nodes)
9192

    
9193
  def Exec(self, feedback_fn):
9194
    """Execute disk replacement.
9195

9196
    This dispatches the disk replacement to the appropriate handler.
9197

9198
    """
9199
    if self.delay_iallocator:
9200
      self._CheckPrereq2()
9201

    
9202
    if __debug__:
9203
      # Verify owned locks before starting operation
9204
      owned_locks = self.lu.glm.list_owned(locking.LEVEL_NODE)
9205
      assert set(owned_locks) == set(self.node_secondary_ip), \
9206
          ("Incorrect node locks, owning %s, expected %s" %
9207
           (owned_locks, self.node_secondary_ip.keys()))
9208

    
9209
      owned_locks = self.lu.glm.list_owned(locking.LEVEL_INSTANCE)
9210
      assert list(owned_locks) == [self.instance_name], \
9211
          "Instance '%s' not locked" % self.instance_name
9212

    
9213
      assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
9214
          "Should not own any node group lock at this point"
9215

    
9216
    if not self.disks:
9217
      feedback_fn("No disks need replacement")
9218
      return
9219

    
9220
    feedback_fn("Replacing disk(s) %s for %s" %
9221
                (utils.CommaJoin(self.disks), self.instance.name))
9222

    
9223
    activate_disks = (not self.instance.admin_up)
9224

    
9225
    # Activate the instance disks if we're replacing them on a down instance
9226
    if activate_disks:
9227
      _StartInstanceDisks(self.lu, self.instance, True)
9228

    
9229
    try:
9230
      # Should we replace the secondary node?
9231
      if self.new_node is not None:
9232
        fn = self._ExecDrbd8Secondary
9233
      else:
9234
        fn = self._ExecDrbd8DiskOnly
9235

    
9236
      result = fn(feedback_fn)
9237
    finally:
9238
      # Deactivate the instance disks if we're replacing them on a
9239
      # down instance
9240
      if activate_disks:
9241
        _SafeShutdownInstanceDisks(self.lu, self.instance)
9242

    
9243
    if __debug__:
9244
      # Verify owned locks
9245
      owned_locks = self.lu.glm.list_owned(locking.LEVEL_NODE)
9246
      nodes = frozenset(self.node_secondary_ip)
9247
      assert ((self.early_release and not owned_locks) or
9248
              (not self.early_release and not (set(owned_locks) - nodes))), \
9249
        ("Not owning the correct locks, early_release=%s, owned=%r,"
9250
         " nodes=%r" % (self.early_release, owned_locks, nodes))
9251

    
9252
    return result
9253

    
9254
  def _CheckVolumeGroup(self, nodes):
9255
    self.lu.LogInfo("Checking volume groups")
9256

    
9257
    vgname = self.cfg.GetVGName()
9258

    
9259
    # Make sure volume group exists on all involved nodes
9260
    results = self.rpc.call_vg_list(nodes)
9261
    if not results:
9262
      raise errors.OpExecError("Can't list volume groups on the nodes")
9263

    
9264
    for node in nodes:
9265
      res = results[node]
9266
      res.Raise("Error checking node %s" % node)
9267
      if vgname not in res.payload:
9268
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
9269
                                 (vgname, node))
9270

    
9271
  def _CheckDisksExistence(self, nodes):
9272
    # Check disk existence
9273
    for idx, dev in enumerate(self.instance.disks):
9274
      if idx not in self.disks:
9275
        continue
9276

    
9277
      for node in nodes:
9278
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
9279
        self.cfg.SetDiskID(dev, node)
9280

    
9281
        result = self.rpc.call_blockdev_find(node, dev)
9282

    
9283
        msg = result.fail_msg
9284
        if msg or not result.payload:
9285
          if not msg:
9286
            msg = "disk not found"
9287
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
9288
                                   (idx, node, msg))
9289

    
9290
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
9291
    for idx, dev in enumerate(self.instance.disks):
9292
      if idx not in self.disks:
9293
        continue
9294

    
9295
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
9296
                      (idx, node_name))
9297

    
9298
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
9299
                                   ldisk=ldisk):
9300
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
9301
                                 " replace disks for instance %s" %
9302
                                 (node_name, self.instance.name))
9303

    
9304
  def _CreateNewStorage(self, node_name):
9305
    iv_names = {}
9306

    
9307
    for idx, dev in enumerate(self.instance.disks):
9308
      if idx not in self.disks:
9309
        continue
9310

    
9311
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
9312

    
9313
      self.cfg.SetDiskID(dev, node_name)
9314

    
9315
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
9316
      names = _GenerateUniqueNames(self.lu, lv_names)
9317

    
9318
      vg_data = dev.children[0].logical_id[0]
9319
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
9320
                             logical_id=(vg_data, names[0]))
9321
      vg_meta = dev.children[1].logical_id[0]
9322
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
9323
                             logical_id=(vg_meta, names[1]))
9324

    
9325
      new_lvs = [lv_data, lv_meta]
9326
      old_lvs = dev.children
9327
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
9328

    
9329
      # we pass force_create=True to force the LVM creation
9330
      for new_lv in new_lvs:
9331
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
9332
                        _GetInstanceInfoText(self.instance), False)
9333

    
9334
    return iv_names
9335

    
9336
  def _CheckDevices(self, node_name, iv_names):
9337
    for name, (dev, _, _) in iv_names.iteritems():
9338
      self.cfg.SetDiskID(dev, node_name)
9339

    
9340
      result = self.rpc.call_blockdev_find(node_name, dev)
9341

    
9342
      msg = result.fail_msg
9343
      if msg or not result.payload:
9344
        if not msg:
9345
          msg = "disk not found"
9346
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
9347
                                 (name, msg))
9348

    
9349
      if result.payload.is_degraded:
9350
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
9351

    
9352
  def _RemoveOldStorage(self, node_name, iv_names):
9353
    for name, (_, old_lvs, _) in iv_names.iteritems():
9354
      self.lu.LogInfo("Remove logical volumes for %s" % name)
9355

    
9356
      for lv in old_lvs:
9357
        self.cfg.SetDiskID(lv, node_name)
9358

    
9359
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
9360
        if msg:
9361
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
9362
                             hint="remove unused LVs manually")
9363

    
9364
  def _ExecDrbd8DiskOnly(self, feedback_fn):
9365
    """Replace a disk on the primary or secondary for DRBD 8.
9366

9367
    The algorithm for replace is quite complicated:
9368

9369
      1. for each disk to be replaced:
9370

9371
        1. create new LVs on the target node with unique names
9372
        1. detach old LVs from the drbd device
9373
        1. rename old LVs to name_replaced.<time_t>
9374
        1. rename new LVs to old LVs
9375
        1. attach the new LVs (with the old names now) to the drbd device
9376

9377
      1. wait for sync across all devices
9378

9379
      1. for each modified disk:
9380

9381
        1. remove old LVs (which have the name name_replaces.<time_t>)
9382

9383
    Failures are not very well handled.
9384

9385
    """
9386
    steps_total = 6
9387

    
9388
    # Step: check device activation
9389
    self.lu.LogStep(1, steps_total, "Check device existence")
9390
    self._CheckDisksExistence([self.other_node, self.target_node])
9391
    self._CheckVolumeGroup([self.target_node, self.other_node])
9392

    
9393
    # Step: check other node consistency
9394
    self.lu.LogStep(2, steps_total, "Check peer consistency")
9395
    self._CheckDisksConsistency(self.other_node,
9396
                                self.other_node == self.instance.primary_node,
9397
                                False)
9398

    
9399
    # Step: create new storage
9400
    self.lu.LogStep(3, steps_total, "Allocate new storage")
9401
    iv_names = self._CreateNewStorage(self.target_node)
9402

    
9403
    # Step: for each lv, detach+rename*2+attach
9404
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
9405
    for dev, old_lvs, new_lvs in iv_names.itervalues():
9406
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
9407

    
9408
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
9409
                                                     old_lvs)
9410
      result.Raise("Can't detach drbd from local storage on node"
9411
                   " %s for device %s" % (self.target_node, dev.iv_name))
9412
      #dev.children = []
9413
      #cfg.Update(instance)
9414

    
9415
      # ok, we created the new LVs, so now we know we have the needed
9416
      # storage; as such, we proceed on the target node to rename
9417
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
9418
      # using the assumption that logical_id == physical_id (which in
9419
      # turn is the unique_id on that node)
9420

    
9421
      # FIXME(iustin): use a better name for the replaced LVs
9422
      temp_suffix = int(time.time())
9423
      ren_fn = lambda d, suff: (d.physical_id[0],
9424
                                d.physical_id[1] + "_replaced-%s" % suff)
9425

    
9426
      # Build the rename list based on what LVs exist on the node
9427
      rename_old_to_new = []
9428
      for to_ren in old_lvs:
9429
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
9430
        if not result.fail_msg and result.payload:
9431
          # device exists
9432
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
9433

    
9434
      self.lu.LogInfo("Renaming the old LVs on the target node")
9435
      result = self.rpc.call_blockdev_rename(self.target_node,
9436
                                             rename_old_to_new)
9437
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
9438

    
9439
      # Now we rename the new LVs to the old LVs
9440
      self.lu.LogInfo("Renaming the new LVs on the target node")
9441
      rename_new_to_old = [(new, old.physical_id)
9442
                           for old, new in zip(old_lvs, new_lvs)]
9443
      result = self.rpc.call_blockdev_rename(self.target_node,
9444
                                             rename_new_to_old)
9445
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
9446

    
9447
      for old, new in zip(old_lvs, new_lvs):
9448
        new.logical_id = old.logical_id
9449
        self.cfg.SetDiskID(new, self.target_node)
9450

    
9451
      for disk in old_lvs:
9452
        disk.logical_id = ren_fn(disk, temp_suffix)
9453
        self.cfg.SetDiskID(disk, self.target_node)
9454

    
9455
      # Now that the new lvs have the old name, we can add them to the device
9456
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
9457
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
9458
                                                  new_lvs)
9459
      msg = result.fail_msg
9460
      if msg:
9461
        for new_lv in new_lvs:
9462
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
9463
                                               new_lv).fail_msg
9464
          if msg2:
9465
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
9466
                               hint=("cleanup manually the unused logical"
9467
                                     "volumes"))
9468
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
9469

    
9470
      dev.children = new_lvs
9471

    
9472
      self.cfg.Update(self.instance, feedback_fn)
9473

    
9474
    cstep = 5
9475
    if self.early_release:
9476
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
9477
      cstep += 1
9478
      self._RemoveOldStorage(self.target_node, iv_names)
9479
      # WARNING: we release both node locks here, do not do other RPCs
9480
      # than WaitForSync to the primary node
9481
      _ReleaseLocks(self.lu, locking.LEVEL_NODE,
9482
                    names=[self.target_node, self.other_node])
9483

    
9484
    # Wait for sync
9485
    # This can fail as the old devices are degraded and _WaitForSync
9486
    # does a combined result over all disks, so we don't check its return value
9487
    self.lu.LogStep(cstep, steps_total, "Sync devices")
9488
    cstep += 1
9489
    _WaitForSync(self.lu, self.instance)
9490

    
9491
    # Check all devices manually
9492
    self._CheckDevices(self.instance.primary_node, iv_names)
9493

    
9494
    # Step: remove old storage
9495
    if not self.early_release:
9496
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
9497
      cstep += 1
9498
      self._RemoveOldStorage(self.target_node, iv_names)
9499

    
9500
  def _ExecDrbd8Secondary(self, feedback_fn):
9501
    """Replace the secondary node for DRBD 8.
9502

9503
    The algorithm for replace is quite complicated:
9504
      - for all disks of the instance:
9505
        - create new LVs on the new node with same names
9506
        - shutdown the drbd device on the old secondary
9507
        - disconnect the drbd network on the primary
9508
        - create the drbd device on the new secondary
9509
        - network attach the drbd on the primary, using an artifice:
9510
          the drbd code for Attach() will connect to the network if it
9511
          finds a device which is connected to the good local disks but
9512
          not network enabled
9513
      - wait for sync across all devices
9514
      - remove all disks from the old secondary
9515

9516
    Failures are not very well handled.
9517

9518
    """
9519
    steps_total = 6
9520

    
9521
    # Step: check device activation
9522
    self.lu.LogStep(1, steps_total, "Check device existence")
9523
    self._CheckDisksExistence([self.instance.primary_node])
9524
    self._CheckVolumeGroup([self.instance.primary_node])
9525

    
9526
    # Step: check other node consistency
9527
    self.lu.LogStep(2, steps_total, "Check peer consistency")
9528
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
9529

    
9530
    # Step: create new storage
9531
    self.lu.LogStep(3, steps_total, "Allocate new storage")
9532
    for idx, dev in enumerate(self.instance.disks):
9533
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
9534
                      (self.new_node, idx))
9535
      # we pass force_create=True to force LVM creation
9536
      for new_lv in dev.children:
9537
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
9538
                        _GetInstanceInfoText(self.instance), False)
9539

    
9540
    # Step 4: dbrd minors and drbd setups changes
9541
    # after this, we must manually remove the drbd minors on both the
9542
    # error and the success paths
9543
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
9544
    minors = self.cfg.AllocateDRBDMinor([self.new_node
9545
                                         for dev in self.instance.disks],
9546
                                        self.instance.name)
9547
    logging.debug("Allocated minors %r", minors)
9548

    
9549
    iv_names = {}
9550
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
9551
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
9552
                      (self.new_node, idx))
9553
      # create new devices on new_node; note that we create two IDs:
9554
      # one without port, so the drbd will be activated without
9555
      # networking information on the new node at this stage, and one
9556
      # with network, for the latter activation in step 4
9557
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
9558
      if self.instance.primary_node == o_node1:
9559
        p_minor = o_minor1
9560
      else:
9561
        assert self.instance.primary_node == o_node2, "Three-node instance?"
9562
        p_minor = o_minor2
9563

    
9564
      new_alone_id = (self.instance.primary_node, self.new_node, None,
9565
                      p_minor, new_minor, o_secret)
9566
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
9567
                    p_minor, new_minor, o_secret)
9568

    
9569
      iv_names[idx] = (dev, dev.children, new_net_id)
9570
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
9571
                    new_net_id)
9572
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
9573
                              logical_id=new_alone_id,
9574
                              children=dev.children,
9575
                              size=dev.size)
9576
      try:
9577
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
9578
                              _GetInstanceInfoText(self.instance), False)
9579
      except errors.GenericError:
9580
        self.cfg.ReleaseDRBDMinors(self.instance.name)
9581
        raise
9582

    
9583
    # We have new devices, shutdown the drbd on the old secondary
9584
    for idx, dev in enumerate(self.instance.disks):
9585
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
9586
      self.cfg.SetDiskID(dev, self.target_node)
9587
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
9588
      if msg:
9589
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
9590
                           "node: %s" % (idx, msg),
9591
                           hint=("Please cleanup this device manually as"
9592
                                 " soon as possible"))
9593

    
9594
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
9595
    result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
9596
                                               self.node_secondary_ip,
9597
                                               self.instance.disks)\
9598
                                              [self.instance.primary_node]
9599

    
9600
    msg = result.fail_msg
9601
    if msg:
9602
      # detaches didn't succeed (unlikely)
9603
      self.cfg.ReleaseDRBDMinors(self.instance.name)
9604
      raise errors.OpExecError("Can't detach the disks from the network on"
9605
                               " old node: %s" % (msg,))
9606

    
9607
    # if we managed to detach at least one, we update all the disks of
9608
    # the instance to point to the new secondary
9609
    self.lu.LogInfo("Updating instance configuration")
9610
    for dev, _, new_logical_id in iv_names.itervalues():
9611
      dev.logical_id = new_logical_id
9612
      self.cfg.SetDiskID(dev, self.instance.primary_node)
9613

    
9614
    self.cfg.Update(self.instance, feedback_fn)
9615

    
9616
    # and now perform the drbd attach
9617
    self.lu.LogInfo("Attaching primary drbds to new secondary"
9618
                    " (standalone => connected)")
9619
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
9620
                                            self.new_node],
9621
                                           self.node_secondary_ip,
9622
                                           self.instance.disks,
9623
                                           self.instance.name,
9624
                                           False)
9625
    for to_node, to_result in result.items():
9626
      msg = to_result.fail_msg
9627
      if msg:
9628
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
9629
                           to_node, msg,
9630
                           hint=("please do a gnt-instance info to see the"
9631
                                 " status of disks"))
9632
    cstep = 5
9633
    if self.early_release:
9634
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
9635
      cstep += 1
9636
      self._RemoveOldStorage(self.target_node, iv_names)
9637
      # WARNING: we release all node locks here, do not do other RPCs
9638
      # than WaitForSync to the primary node
9639
      _ReleaseLocks(self.lu, locking.LEVEL_NODE,
9640
                    names=[self.instance.primary_node,
9641
                           self.target_node,
9642
                           self.new_node])
9643

    
9644
    # Wait for sync
9645
    # This can fail as the old devices are degraded and _WaitForSync
9646
    # does a combined result over all disks, so we don't check its return value
9647
    self.lu.LogStep(cstep, steps_total, "Sync devices")
9648
    cstep += 1
9649
    _WaitForSync(self.lu, self.instance)
9650

    
9651
    # Check all devices manually
9652
    self._CheckDevices(self.instance.primary_node, iv_names)
9653

    
9654
    # Step: remove old storage
9655
    if not self.early_release:
9656
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
9657
      self._RemoveOldStorage(self.target_node, iv_names)
9658

    
9659

    
9660
class LURepairNodeStorage(NoHooksLU):
9661
  """Repairs the volume group on a node.
9662

9663
  """
9664
  REQ_BGL = False
9665

    
9666
  def CheckArguments(self):
9667
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
9668

    
9669
    storage_type = self.op.storage_type
9670

    
9671
    if (constants.SO_FIX_CONSISTENCY not in
9672
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
9673
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
9674
                                 " repaired" % storage_type,
9675
                                 errors.ECODE_INVAL)
9676

    
9677
  def ExpandNames(self):
9678
    self.needed_locks = {
9679
      locking.LEVEL_NODE: [self.op.node_name],
9680
      }
9681

    
9682
  def _CheckFaultyDisks(self, instance, node_name):
9683
    """Ensure faulty disks abort the opcode or at least warn."""
9684
    try:
9685
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
9686
                                  node_name, True):
9687
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
9688
                                   " node '%s'" % (instance.name, node_name),
9689
                                   errors.ECODE_STATE)
9690
    except errors.OpPrereqError, err:
9691
      if self.op.ignore_consistency:
9692
        self.proc.LogWarning(str(err.args[0]))
9693
      else:
9694
        raise
9695

    
9696
  def CheckPrereq(self):
9697
    """Check prerequisites.
9698

9699
    """
9700
    # Check whether any instance on this node has faulty disks
9701
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
9702
      if not inst.admin_up:
9703
        continue
9704
      check_nodes = set(inst.all_nodes)
9705
      check_nodes.discard(self.op.node_name)
9706
      for inst_node_name in check_nodes:
9707
        self._CheckFaultyDisks(inst, inst_node_name)
9708

    
9709
  def Exec(self, feedback_fn):
9710
    feedback_fn("Repairing storage unit '%s' on %s ..." %
9711
                (self.op.name, self.op.node_name))
9712

    
9713
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
9714
    result = self.rpc.call_storage_execute(self.op.node_name,
9715
                                           self.op.storage_type, st_args,
9716
                                           self.op.name,
9717
                                           constants.SO_FIX_CONSISTENCY)
9718
    result.Raise("Failed to repair storage unit '%s' on %s" %
9719
                 (self.op.name, self.op.node_name))
9720

    
9721

    
9722
class LUNodeEvacStrategy(NoHooksLU):
9723
  """Computes the node evacuation strategy.
9724

9725
  """
9726
  REQ_BGL = False
9727

    
9728
  def CheckArguments(self):
9729
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
9730

    
9731
  def ExpandNames(self):
9732
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
9733
    self.needed_locks = locks = {}
9734
    if self.op.remote_node is None:
9735
      locks[locking.LEVEL_NODE] = locking.ALL_SET
9736
    else:
9737
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9738
      locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
9739

    
9740
  def Exec(self, feedback_fn):
9741
    instances = []
9742
    for node in self.op.nodes:
9743
      instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
9744
    if not instances:
9745
      return []
9746

    
9747
    if self.op.remote_node is not None:
9748
      result = []
9749
      for i in instances:
9750
        if i.primary_node == self.op.remote_node:
9751
          raise errors.OpPrereqError("Node %s is the primary node of"
9752
                                     " instance %s, cannot use it as"
9753
                                     " secondary" %
9754
                                     (self.op.remote_node, i.name),
9755
                                     errors.ECODE_INVAL)
9756
        result.append([i.name, self.op.remote_node])
9757
    else:
9758
      ial = IAllocator(self.cfg, self.rpc,
9759
                       mode=constants.IALLOCATOR_MODE_MEVAC,
9760
                       evac_nodes=self.op.nodes)
9761
      ial.Run(self.op.iallocator, validate=True)
9762
      if not ial.success:
9763
        raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
9764
                                 errors.ECODE_NORES)
9765
      result = ial.result
9766
    return result
9767

    
9768

    
9769
class LUInstanceGrowDisk(LogicalUnit):
9770
  """Grow a disk of an instance.
9771

9772
  """
9773
  HPATH = "disk-grow"
9774
  HTYPE = constants.HTYPE_INSTANCE
9775
  REQ_BGL = False
9776

    
9777
  def ExpandNames(self):
9778
    self._ExpandAndLockInstance()
9779
    self.needed_locks[locking.LEVEL_NODE] = []
9780
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9781

    
9782
  def DeclareLocks(self, level):
9783
    if level == locking.LEVEL_NODE:
9784
      self._LockInstancesNodes()
9785

    
9786
  def BuildHooksEnv(self):
9787
    """Build hooks env.
9788

9789
    This runs on the master, the primary and all the secondaries.
9790

9791
    """
9792
    env = {
9793
      "DISK": self.op.disk,
9794
      "AMOUNT": self.op.amount,
9795
      }
9796
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
9797
    return env
9798

    
9799
  def BuildHooksNodes(self):
9800
    """Build hooks nodes.
9801

9802
    """
9803
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
9804
    return (nl, nl)
9805

    
9806
  def CheckPrereq(self):
9807
    """Check prerequisites.
9808

9809
    This checks that the instance is in the cluster.
9810

9811
    """
9812
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9813
    assert instance is not None, \
9814
      "Cannot retrieve locked instance %s" % self.op.instance_name
9815
    nodenames = list(instance.all_nodes)
9816
    for node in nodenames:
9817
      _CheckNodeOnline(self, node)
9818

    
9819
    self.instance = instance
9820

    
9821
    if instance.disk_template not in constants.DTS_GROWABLE:
9822
      raise errors.OpPrereqError("Instance's disk layout does not support"
9823
                                 " growing", errors.ECODE_INVAL)
9824

    
9825
    self.disk = instance.FindDisk(self.op.disk)
9826

    
9827
    if instance.disk_template not in (constants.DT_FILE,
9828
                                      constants.DT_SHARED_FILE):
9829
      # TODO: check the free disk space for file, when that feature will be
9830
      # supported
9831
      _CheckNodesFreeDiskPerVG(self, nodenames,
9832
                               self.disk.ComputeGrowth(self.op.amount))
9833

    
9834
  def Exec(self, feedback_fn):
9835
    """Execute disk grow.
9836

9837
    """
9838
    instance = self.instance
9839
    disk = self.disk
9840

    
9841
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
9842
    if not disks_ok:
9843
      raise errors.OpExecError("Cannot activate block device to grow")
9844

    
9845
    # First run all grow ops in dry-run mode
9846
    for node in instance.all_nodes:
9847
      self.cfg.SetDiskID(disk, node)
9848
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, True)
9849
      result.Raise("Grow request failed to node %s" % node)
9850

    
9851
    # We know that (as far as we can test) operations across different
9852
    # nodes will succeed, time to run it for real
9853
    for node in instance.all_nodes:
9854
      self.cfg.SetDiskID(disk, node)
9855
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount, False)
9856
      result.Raise("Grow request failed to node %s" % node)
9857

    
9858
      # TODO: Rewrite code to work properly
9859
      # DRBD goes into sync mode for a short amount of time after executing the
9860
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
9861
      # calling "resize" in sync mode fails. Sleeping for a short amount of
9862
      # time is a work-around.
9863
      time.sleep(5)
9864

    
9865
    disk.RecordGrow(self.op.amount)
9866
    self.cfg.Update(instance, feedback_fn)
9867
    if self.op.wait_for_sync:
9868
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
9869
      if disk_abort:
9870
        self.proc.LogWarning("Disk sync-ing has not returned a good"
9871
                             " status; please check the instance")
9872
      if not instance.admin_up:
9873
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
9874
    elif not instance.admin_up:
9875
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
9876
                           " not supposed to be running because no wait for"
9877
                           " sync mode was requested")
9878

    
9879

    
9880
class LUInstanceQueryData(NoHooksLU):
9881
  """Query runtime instance data.
9882

9883
  """
9884
  REQ_BGL = False
9885

    
9886
  def ExpandNames(self):
9887
    self.needed_locks = {}
9888

    
9889
    # Use locking if requested or when non-static information is wanted
9890
    if not (self.op.static or self.op.use_locking):
9891
      self.LogWarning("Non-static data requested, locks need to be acquired")
9892
      self.op.use_locking = True
9893

    
9894
    if self.op.instances or not self.op.use_locking:
9895
      # Expand instance names right here
9896
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
9897
    else:
9898
      # Will use acquired locks
9899
      self.wanted_names = None
9900

    
9901
    if self.op.use_locking:
9902
      self.share_locks = dict.fromkeys(locking.LEVELS, 1)
9903

    
9904
      if self.wanted_names is None:
9905
        self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
9906
      else:
9907
        self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
9908

    
9909
      self.needed_locks[locking.LEVEL_NODE] = []
9910
      self.share_locks = dict.fromkeys(locking.LEVELS, 1)
9911
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9912

    
9913
  def DeclareLocks(self, level):
9914
    if self.op.use_locking and level == locking.LEVEL_NODE:
9915
      self._LockInstancesNodes()
9916

    
9917
  def CheckPrereq(self):
9918
    """Check prerequisites.
9919

9920
    This only checks the optional instance list against the existing names.
9921

9922
    """
9923
    if self.wanted_names is None:
9924
      assert self.op.use_locking, "Locking was not used"
9925
      self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
9926

    
9927
    self.wanted_instances = [self.cfg.GetInstanceInfo(name)
9928
                             for name in self.wanted_names]
9929

    
9930
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
9931
    """Returns the status of a block device
9932

9933
    """
9934
    if self.op.static or not node:
9935
      return None
9936

    
9937
    self.cfg.SetDiskID(dev, node)
9938

    
9939
    result = self.rpc.call_blockdev_find(node, dev)
9940
    if result.offline:
9941
      return None
9942

    
9943
    result.Raise("Can't compute disk status for %s" % instance_name)
9944

    
9945
    status = result.payload
9946
    if status is None:
9947
      return None
9948

    
9949
    return (status.dev_path, status.major, status.minor,
9950
            status.sync_percent, status.estimated_time,
9951
            status.is_degraded, status.ldisk_status)
9952

    
9953
  def _ComputeDiskStatus(self, instance, snode, dev):
9954
    """Compute block device status.
9955

9956
    """
9957
    if dev.dev_type in constants.LDS_DRBD:
9958
      # we change the snode then (otherwise we use the one passed in)
9959
      if dev.logical_id[0] == instance.primary_node:
9960
        snode = dev.logical_id[1]
9961
      else:
9962
        snode = dev.logical_id[0]
9963

    
9964
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
9965
                                              instance.name, dev)
9966
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
9967

    
9968
    if dev.children:
9969
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
9970
                      for child in dev.children]
9971
    else:
9972
      dev_children = []
9973

    
9974
    return {
9975
      "iv_name": dev.iv_name,
9976
      "dev_type": dev.dev_type,
9977
      "logical_id": dev.logical_id,
9978
      "physical_id": dev.physical_id,
9979
      "pstatus": dev_pstatus,
9980
      "sstatus": dev_sstatus,
9981
      "children": dev_children,
9982
      "mode": dev.mode,
9983
      "size": dev.size,
9984
      }
9985

    
9986
  def Exec(self, feedback_fn):
9987
    """Gather and return data"""
9988
    result = {}
9989

    
9990
    cluster = self.cfg.GetClusterInfo()
9991

    
9992
    for instance in self.wanted_instances:
9993
      if not self.op.static:
9994
        remote_info = self.rpc.call_instance_info(instance.primary_node,
9995
                                                  instance.name,
9996
                                                  instance.hypervisor)
9997
        remote_info.Raise("Error checking node %s" % instance.primary_node)
9998
        remote_info = remote_info.payload
9999
        if remote_info and "state" in remote_info:
10000
          remote_state = "up"
10001
        else:
10002
          remote_state = "down"
10003
      else:
10004
        remote_state = None
10005
      if instance.admin_up:
10006
        config_state = "up"
10007
      else:
10008
        config_state = "down"
10009

    
10010
      disks = [self._ComputeDiskStatus(instance, None, device)
10011
               for device in instance.disks]
10012

    
10013
      result[instance.name] = {
10014
        "name": instance.name,
10015
        "config_state": config_state,
10016
        "run_state": remote_state,
10017
        "pnode": instance.primary_node,
10018
        "snodes": instance.secondary_nodes,
10019
        "os": instance.os,
10020
        # this happens to be the same format used for hooks
10021
        "nics": _NICListToTuple(self, instance.nics),
10022
        "disk_template": instance.disk_template,
10023
        "disks": disks,
10024
        "hypervisor": instance.hypervisor,
10025
        "network_port": instance.network_port,
10026
        "hv_instance": instance.hvparams,
10027
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
10028
        "be_instance": instance.beparams,
10029
        "be_actual": cluster.FillBE(instance),
10030
        "os_instance": instance.osparams,
10031
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
10032
        "serial_no": instance.serial_no,
10033
        "mtime": instance.mtime,
10034
        "ctime": instance.ctime,
10035
        "uuid": instance.uuid,
10036
        }
10037

    
10038
    return result
10039

    
10040

    
10041
class LUInstanceSetParams(LogicalUnit):
10042
  """Modifies an instances's parameters.
10043

10044
  """
10045
  HPATH = "instance-modify"
10046
  HTYPE = constants.HTYPE_INSTANCE
10047
  REQ_BGL = False
10048

    
10049
  def CheckArguments(self):
10050
    if not (self.op.nics or self.op.disks or self.op.disk_template or
10051
            self.op.hvparams or self.op.beparams or self.op.os_name):
10052
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
10053

    
10054
    if self.op.hvparams:
10055
      _CheckGlobalHvParams(self.op.hvparams)
10056

    
10057
    # Disk validation
10058
    disk_addremove = 0
10059
    for disk_op, disk_dict in self.op.disks:
10060
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
10061
      if disk_op == constants.DDM_REMOVE:
10062
        disk_addremove += 1
10063
        continue
10064
      elif disk_op == constants.DDM_ADD:
10065
        disk_addremove += 1
10066
      else:
10067
        if not isinstance(disk_op, int):
10068
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
10069
        if not isinstance(disk_dict, dict):
10070
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
10071
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
10072

    
10073
      if disk_op == constants.DDM_ADD:
10074
        mode = disk_dict.setdefault(constants.IDISK_MODE, constants.DISK_RDWR)
10075
        if mode not in constants.DISK_ACCESS_SET:
10076
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
10077
                                     errors.ECODE_INVAL)
10078
        size = disk_dict.get(constants.IDISK_SIZE, None)
10079
        if size is None:
10080
          raise errors.OpPrereqError("Required disk parameter size missing",
10081
                                     errors.ECODE_INVAL)
10082
        try:
10083
          size = int(size)
10084
        except (TypeError, ValueError), err:
10085
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
10086
                                     str(err), errors.ECODE_INVAL)
10087
        disk_dict[constants.IDISK_SIZE] = size
10088
      else:
10089
        # modification of disk
10090
        if constants.IDISK_SIZE in disk_dict:
10091
          raise errors.OpPrereqError("Disk size change not possible, use"
10092
                                     " grow-disk", errors.ECODE_INVAL)
10093

    
10094
    if disk_addremove > 1:
10095
      raise errors.OpPrereqError("Only one disk add or remove operation"
10096
                                 " supported at a time", errors.ECODE_INVAL)
10097

    
10098
    if self.op.disks and self.op.disk_template is not None:
10099
      raise errors.OpPrereqError("Disk template conversion and other disk"
10100
                                 " changes not supported at the same time",
10101
                                 errors.ECODE_INVAL)
10102

    
10103
    if (self.op.disk_template and
10104
        self.op.disk_template in constants.DTS_INT_MIRROR and
10105
        self.op.remote_node is None):
10106
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
10107
                                 " one requires specifying a secondary node",
10108
                                 errors.ECODE_INVAL)
10109

    
10110
    # NIC validation
10111
    nic_addremove = 0
10112
    for nic_op, nic_dict in self.op.nics:
10113
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
10114
      if nic_op == constants.DDM_REMOVE:
10115
        nic_addremove += 1
10116
        continue
10117
      elif nic_op == constants.DDM_ADD:
10118
        nic_addremove += 1
10119
      else:
10120
        if not isinstance(nic_op, int):
10121
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
10122
        if not isinstance(nic_dict, dict):
10123
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
10124
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
10125

    
10126
      # nic_dict should be a dict
10127
      nic_ip = nic_dict.get(constants.INIC_IP, None)
10128
      if nic_ip is not None:
10129
        if nic_ip.lower() == constants.VALUE_NONE:
10130
          nic_dict[constants.INIC_IP] = None
10131
        else:
10132
          if not netutils.IPAddress.IsValid(nic_ip):
10133
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
10134
                                       errors.ECODE_INVAL)
10135

    
10136
      nic_bridge = nic_dict.get('bridge', None)
10137
      nic_link = nic_dict.get(constants.INIC_LINK, None)
10138
      if nic_bridge and nic_link:
10139
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
10140
                                   " at the same time", errors.ECODE_INVAL)
10141
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
10142
        nic_dict['bridge'] = None
10143
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
10144
        nic_dict[constants.INIC_LINK] = None
10145

    
10146
      if nic_op == constants.DDM_ADD:
10147
        nic_mac = nic_dict.get(constants.INIC_MAC, None)
10148
        if nic_mac is None:
10149
          nic_dict[constants.INIC_MAC] = constants.VALUE_AUTO
10150

    
10151
      if constants.INIC_MAC in nic_dict:
10152
        nic_mac = nic_dict[constants.INIC_MAC]
10153
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
10154
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
10155

    
10156
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
10157
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
10158
                                     " modifying an existing nic",
10159
                                     errors.ECODE_INVAL)
10160

    
10161
    if nic_addremove > 1:
10162
      raise errors.OpPrereqError("Only one NIC add or remove operation"
10163
                                 " supported at a time", errors.ECODE_INVAL)
10164

    
10165
  def ExpandNames(self):
10166
    self._ExpandAndLockInstance()
10167
    self.needed_locks[locking.LEVEL_NODE] = []
10168
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
10169

    
10170
  def DeclareLocks(self, level):
10171
    if level == locking.LEVEL_NODE:
10172
      self._LockInstancesNodes()
10173
      if self.op.disk_template and self.op.remote_node:
10174
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
10175
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
10176

    
10177
  def BuildHooksEnv(self):
10178
    """Build hooks env.
10179

10180
    This runs on the master, primary and secondaries.
10181

10182
    """
10183
    args = dict()
10184
    if constants.BE_MEMORY in self.be_new:
10185
      args['memory'] = self.be_new[constants.BE_MEMORY]
10186
    if constants.BE_VCPUS in self.be_new:
10187
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
10188
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
10189
    # information at all.
10190
    if self.op.nics:
10191
      args['nics'] = []
10192
      nic_override = dict(self.op.nics)
10193
      for idx, nic in enumerate(self.instance.nics):
10194
        if idx in nic_override:
10195
          this_nic_override = nic_override[idx]
10196
        else:
10197
          this_nic_override = {}
10198
        if constants.INIC_IP in this_nic_override:
10199
          ip = this_nic_override[constants.INIC_IP]
10200
        else:
10201
          ip = nic.ip
10202
        if constants.INIC_MAC in this_nic_override:
10203
          mac = this_nic_override[constants.INIC_MAC]
10204
        else:
10205
          mac = nic.mac
10206
        if idx in self.nic_pnew:
10207
          nicparams = self.nic_pnew[idx]
10208
        else:
10209
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
10210
        mode = nicparams[constants.NIC_MODE]
10211
        link = nicparams[constants.NIC_LINK]
10212
        args['nics'].append((ip, mac, mode, link))
10213
      if constants.DDM_ADD in nic_override:
10214
        ip = nic_override[constants.DDM_ADD].get(constants.INIC_IP, None)
10215
        mac = nic_override[constants.DDM_ADD][constants.INIC_MAC]
10216
        nicparams = self.nic_pnew[constants.DDM_ADD]
10217
        mode = nicparams[constants.NIC_MODE]
10218
        link = nicparams[constants.NIC_LINK]
10219
        args['nics'].append((ip, mac, mode, link))
10220
      elif constants.DDM_REMOVE in nic_override:
10221
        del args['nics'][-1]
10222

    
10223
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
10224
    if self.op.disk_template:
10225
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
10226

    
10227
    return env
10228

    
10229
  def BuildHooksNodes(self):
10230
    """Build hooks nodes.
10231

10232
    """
10233
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
10234
    return (nl, nl)
10235

    
10236
  def CheckPrereq(self):
10237
    """Check prerequisites.
10238

10239
    This only checks the instance list against the existing names.
10240

10241
    """
10242
    # checking the new params on the primary/secondary nodes
10243

    
10244
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
10245
    cluster = self.cluster = self.cfg.GetClusterInfo()
10246
    assert self.instance is not None, \
10247
      "Cannot retrieve locked instance %s" % self.op.instance_name
10248
    pnode = instance.primary_node
10249
    nodelist = list(instance.all_nodes)
10250

    
10251
    # OS change
10252
    if self.op.os_name and not self.op.force:
10253
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
10254
                      self.op.force_variant)
10255
      instance_os = self.op.os_name
10256
    else:
10257
      instance_os = instance.os
10258

    
10259
    if self.op.disk_template:
10260
      if instance.disk_template == self.op.disk_template:
10261
        raise errors.OpPrereqError("Instance already has disk template %s" %
10262
                                   instance.disk_template, errors.ECODE_INVAL)
10263

    
10264
      if (instance.disk_template,
10265
          self.op.disk_template) not in self._DISK_CONVERSIONS:
10266
        raise errors.OpPrereqError("Unsupported disk template conversion from"
10267
                                   " %s to %s" % (instance.disk_template,
10268
                                                  self.op.disk_template),
10269
                                   errors.ECODE_INVAL)
10270
      _CheckInstanceDown(self, instance, "cannot change disk template")
10271
      if self.op.disk_template in constants.DTS_INT_MIRROR:
10272
        if self.op.remote_node == pnode:
10273
          raise errors.OpPrereqError("Given new secondary node %s is the same"
10274
                                     " as the primary node of the instance" %
10275
                                     self.op.remote_node, errors.ECODE_STATE)
10276
        _CheckNodeOnline(self, self.op.remote_node)
10277
        _CheckNodeNotDrained(self, self.op.remote_node)
10278
        # FIXME: here we assume that the old instance type is DT_PLAIN
10279
        assert instance.disk_template == constants.DT_PLAIN
10280
        disks = [{constants.IDISK_SIZE: d.size,
10281
                  constants.IDISK_VG: d.logical_id[0]}
10282
                 for d in instance.disks]
10283
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
10284
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
10285

    
10286
    # hvparams processing
10287
    if self.op.hvparams:
10288
      hv_type = instance.hypervisor
10289
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
10290
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
10291
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
10292

    
10293
      # local check
10294
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
10295
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
10296
      self.hv_new = hv_new # the new actual values
10297
      self.hv_inst = i_hvdict # the new dict (without defaults)
10298
    else:
10299
      self.hv_new = self.hv_inst = {}
10300

    
10301
    # beparams processing
10302
    if self.op.beparams:
10303
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
10304
                                   use_none=True)
10305
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
10306
      be_new = cluster.SimpleFillBE(i_bedict)
10307
      self.be_new = be_new # the new actual values
10308
      self.be_inst = i_bedict # the new dict (without defaults)
10309
    else:
10310
      self.be_new = self.be_inst = {}
10311
    be_old = cluster.FillBE(instance)
10312

    
10313
    # osparams processing
10314
    if self.op.osparams:
10315
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
10316
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
10317
      self.os_inst = i_osdict # the new dict (without defaults)
10318
    else:
10319
      self.os_inst = {}
10320

    
10321
    self.warn = []
10322

    
10323
    if (constants.BE_MEMORY in self.op.beparams and not self.op.force and
10324
        be_new[constants.BE_MEMORY] > be_old[constants.BE_MEMORY]):
10325
      mem_check_list = [pnode]
10326
      if be_new[constants.BE_AUTO_BALANCE]:
10327
        # either we changed auto_balance to yes or it was from before
10328
        mem_check_list.extend(instance.secondary_nodes)
10329
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
10330
                                                  instance.hypervisor)
10331
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
10332
                                         instance.hypervisor)
10333
      pninfo = nodeinfo[pnode]
10334
      msg = pninfo.fail_msg
10335
      if msg:
10336
        # Assume the primary node is unreachable and go ahead
10337
        self.warn.append("Can't get info from primary node %s: %s" %
10338
                         (pnode,  msg))
10339
      elif not isinstance(pninfo.payload.get('memory_free', None), int):
10340
        self.warn.append("Node data from primary node %s doesn't contain"
10341
                         " free memory information" % pnode)
10342
      elif instance_info.fail_msg:
10343
        self.warn.append("Can't get instance runtime information: %s" %
10344
                        instance_info.fail_msg)
10345
      else:
10346
        if instance_info.payload:
10347
          current_mem = int(instance_info.payload['memory'])
10348
        else:
10349
          # Assume instance not running
10350
          # (there is a slight race condition here, but it's not very probable,
10351
          # and we have no other way to check)
10352
          current_mem = 0
10353
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
10354
                    pninfo.payload['memory_free'])
10355
        if miss_mem > 0:
10356
          raise errors.OpPrereqError("This change will prevent the instance"
10357
                                     " from starting, due to %d MB of memory"
10358
                                     " missing on its primary node" % miss_mem,
10359
                                     errors.ECODE_NORES)
10360

    
10361
      if be_new[constants.BE_AUTO_BALANCE]:
10362
        for node, nres in nodeinfo.items():
10363
          if node not in instance.secondary_nodes:
10364
            continue
10365
          nres.Raise("Can't get info from secondary node %s" % node,
10366
                     prereq=True, ecode=errors.ECODE_STATE)
10367
          if not isinstance(nres.payload.get('memory_free', None), int):
10368
            raise errors.OpPrereqError("Secondary node %s didn't return free"
10369
                                       " memory information" % node,
10370
                                       errors.ECODE_STATE)
10371
          elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
10372
            raise errors.OpPrereqError("This change will prevent the instance"
10373
                                       " from failover to its secondary node"
10374
                                       " %s, due to not enough memory" % node,
10375
                                       errors.ECODE_STATE)
10376

    
10377
    # NIC processing
10378
    self.nic_pnew = {}
10379
    self.nic_pinst = {}
10380
    for nic_op, nic_dict in self.op.nics:
10381
      if nic_op == constants.DDM_REMOVE:
10382
        if not instance.nics:
10383
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
10384
                                     errors.ECODE_INVAL)
10385
        continue
10386
      if nic_op != constants.DDM_ADD:
10387
        # an existing nic
10388
        if not instance.nics:
10389
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
10390
                                     " no NICs" % nic_op,
10391
                                     errors.ECODE_INVAL)
10392
        if nic_op < 0 or nic_op >= len(instance.nics):
10393
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
10394
                                     " are 0 to %d" %
10395
                                     (nic_op, len(instance.nics) - 1),
10396
                                     errors.ECODE_INVAL)
10397
        old_nic_params = instance.nics[nic_op].nicparams
10398
        old_nic_ip = instance.nics[nic_op].ip
10399
      else:
10400
        old_nic_params = {}
10401
        old_nic_ip = None
10402

    
10403
      update_params_dict = dict([(key, nic_dict[key])
10404
                                 for key in constants.NICS_PARAMETERS
10405
                                 if key in nic_dict])
10406

    
10407
      if 'bridge' in nic_dict:
10408
        update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
10409

    
10410
      new_nic_params = _GetUpdatedParams(old_nic_params,
10411
                                         update_params_dict)
10412
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
10413
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
10414
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
10415
      self.nic_pinst[nic_op] = new_nic_params
10416
      self.nic_pnew[nic_op] = new_filled_nic_params
10417
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
10418

    
10419
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
10420
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
10421
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
10422
        if msg:
10423
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
10424
          if self.op.force:
10425
            self.warn.append(msg)
10426
          else:
10427
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
10428
      if new_nic_mode == constants.NIC_MODE_ROUTED:
10429
        if constants.INIC_IP in nic_dict:
10430
          nic_ip = nic_dict[constants.INIC_IP]
10431
        else:
10432
          nic_ip = old_nic_ip
10433
        if nic_ip is None:
10434
          raise errors.OpPrereqError('Cannot set the nic ip to None'
10435
                                     ' on a routed nic', errors.ECODE_INVAL)
10436
      if constants.INIC_MAC in nic_dict:
10437
        nic_mac = nic_dict[constants.INIC_MAC]
10438
        if nic_mac is None:
10439
          raise errors.OpPrereqError('Cannot set the nic mac to None',
10440
                                     errors.ECODE_INVAL)
10441
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
10442
          # otherwise generate the mac
10443
          nic_dict[constants.INIC_MAC] = \
10444
            self.cfg.GenerateMAC(self.proc.GetECId())
10445
        else:
10446
          # or validate/reserve the current one
10447
          try:
10448
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
10449
          except errors.ReservationError:
10450
            raise errors.OpPrereqError("MAC address %s already in use"
10451
                                       " in cluster" % nic_mac,
10452
                                       errors.ECODE_NOTUNIQUE)
10453

    
10454
    # DISK processing
10455
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
10456
      raise errors.OpPrereqError("Disk operations not supported for"
10457
                                 " diskless instances",
10458
                                 errors.ECODE_INVAL)
10459
    for disk_op, _ in self.op.disks:
10460
      if disk_op == constants.DDM_REMOVE:
10461
        if len(instance.disks) == 1:
10462
          raise errors.OpPrereqError("Cannot remove the last disk of"
10463
                                     " an instance", errors.ECODE_INVAL)
10464
        _CheckInstanceDown(self, instance, "cannot remove disks")
10465

    
10466
      if (disk_op == constants.DDM_ADD and
10467
          len(instance.disks) >= constants.MAX_DISKS):
10468
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
10469
                                   " add more" % constants.MAX_DISKS,
10470
                                   errors.ECODE_STATE)
10471
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
10472
        # an existing disk
10473
        if disk_op < 0 or disk_op >= len(instance.disks):
10474
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
10475
                                     " are 0 to %d" %
10476
                                     (disk_op, len(instance.disks)),
10477
                                     errors.ECODE_INVAL)
10478

    
10479
    return
10480

    
10481
  def _ConvertPlainToDrbd(self, feedback_fn):
10482
    """Converts an instance from plain to drbd.
10483

10484
    """
10485
    feedback_fn("Converting template to drbd")
10486
    instance = self.instance
10487
    pnode = instance.primary_node
10488
    snode = self.op.remote_node
10489

    
10490
    # create a fake disk info for _GenerateDiskTemplate
10491
    disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
10492
                  constants.IDISK_VG: d.logical_id[0]}
10493
                 for d in instance.disks]
10494
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
10495
                                      instance.name, pnode, [snode],
10496
                                      disk_info, None, None, 0, feedback_fn)
10497
    info = _GetInstanceInfoText(instance)
10498
    feedback_fn("Creating aditional volumes...")
10499
    # first, create the missing data and meta devices
10500
    for disk in new_disks:
10501
      # unfortunately this is... not too nice
10502
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
10503
                            info, True)
10504
      for child in disk.children:
10505
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
10506
    # at this stage, all new LVs have been created, we can rename the
10507
    # old ones
10508
    feedback_fn("Renaming original volumes...")
10509
    rename_list = [(o, n.children[0].logical_id)
10510
                   for (o, n) in zip(instance.disks, new_disks)]
10511
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
10512
    result.Raise("Failed to rename original LVs")
10513

    
10514
    feedback_fn("Initializing DRBD devices...")
10515
    # all child devices are in place, we can now create the DRBD devices
10516
    for disk in new_disks:
10517
      for node in [pnode, snode]:
10518
        f_create = node == pnode
10519
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
10520

    
10521
    # at this point, the instance has been modified
10522
    instance.disk_template = constants.DT_DRBD8
10523
    instance.disks = new_disks
10524
    self.cfg.Update(instance, feedback_fn)
10525

    
10526
    # disks are created, waiting for sync
10527
    disk_abort = not _WaitForSync(self, instance,
10528
                                  oneshot=not self.op.wait_for_sync)
10529
    if disk_abort:
10530
      raise errors.OpExecError("There are some degraded disks for"
10531
                               " this instance, please cleanup manually")
10532

    
10533
  def _ConvertDrbdToPlain(self, feedback_fn):
10534
    """Converts an instance from drbd to plain.
10535

10536
    """
10537
    instance = self.instance
10538
    assert len(instance.secondary_nodes) == 1
10539
    pnode = instance.primary_node
10540
    snode = instance.secondary_nodes[0]
10541
    feedback_fn("Converting template to plain")
10542

    
10543
    old_disks = instance.disks
10544
    new_disks = [d.children[0] for d in old_disks]
10545

    
10546
    # copy over size and mode
10547
    for parent, child in zip(old_disks, new_disks):
10548
      child.size = parent.size
10549
      child.mode = parent.mode
10550

    
10551
    # update instance structure
10552
    instance.disks = new_disks
10553
    instance.disk_template = constants.DT_PLAIN
10554
    self.cfg.Update(instance, feedback_fn)
10555

    
10556
    feedback_fn("Removing volumes on the secondary node...")
10557
    for disk in old_disks:
10558
      self.cfg.SetDiskID(disk, snode)
10559
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
10560
      if msg:
10561
        self.LogWarning("Could not remove block device %s on node %s,"
10562
                        " continuing anyway: %s", disk.iv_name, snode, msg)
10563

    
10564
    feedback_fn("Removing unneeded volumes on the primary node...")
10565
    for idx, disk in enumerate(old_disks):
10566
      meta = disk.children[1]
10567
      self.cfg.SetDiskID(meta, pnode)
10568
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
10569
      if msg:
10570
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
10571
                        " continuing anyway: %s", idx, pnode, msg)
10572

    
10573
  def Exec(self, feedback_fn):
10574
    """Modifies an instance.
10575

10576
    All parameters take effect only at the next restart of the instance.
10577

10578
    """
10579
    # Process here the warnings from CheckPrereq, as we don't have a
10580
    # feedback_fn there.
10581
    for warn in self.warn:
10582
      feedback_fn("WARNING: %s" % warn)
10583

    
10584
    result = []
10585
    instance = self.instance
10586
    # disk changes
10587
    for disk_op, disk_dict in self.op.disks:
10588
      if disk_op == constants.DDM_REMOVE:
10589
        # remove the last disk
10590
        device = instance.disks.pop()
10591
        device_idx = len(instance.disks)
10592
        for node, disk in device.ComputeNodeTree(instance.primary_node):
10593
          self.cfg.SetDiskID(disk, node)
10594
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
10595
          if msg:
10596
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
10597
                            " continuing anyway", device_idx, node, msg)
10598
        result.append(("disk/%d" % device_idx, "remove"))
10599
      elif disk_op == constants.DDM_ADD:
10600
        # add a new disk
10601
        if instance.disk_template in (constants.DT_FILE,
10602
                                        constants.DT_SHARED_FILE):
10603
          file_driver, file_path = instance.disks[0].logical_id
10604
          file_path = os.path.dirname(file_path)
10605
        else:
10606
          file_driver = file_path = None
10607
        disk_idx_base = len(instance.disks)
10608
        new_disk = _GenerateDiskTemplate(self,
10609
                                         instance.disk_template,
10610
                                         instance.name, instance.primary_node,
10611
                                         instance.secondary_nodes,
10612
                                         [disk_dict],
10613
                                         file_path,
10614
                                         file_driver,
10615
                                         disk_idx_base, feedback_fn)[0]
10616
        instance.disks.append(new_disk)
10617
        info = _GetInstanceInfoText(instance)
10618

    
10619
        logging.info("Creating volume %s for instance %s",
10620
                     new_disk.iv_name, instance.name)
10621
        # Note: this needs to be kept in sync with _CreateDisks
10622
        #HARDCODE
10623
        for node in instance.all_nodes:
10624
          f_create = node == instance.primary_node
10625
          try:
10626
            _CreateBlockDev(self, node, instance, new_disk,
10627
                            f_create, info, f_create)
10628
          except errors.OpExecError, err:
10629
            self.LogWarning("Failed to create volume %s (%s) on"
10630
                            " node %s: %s",
10631
                            new_disk.iv_name, new_disk, node, err)
10632
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
10633
                       (new_disk.size, new_disk.mode)))
10634
      else:
10635
        # change a given disk
10636
        instance.disks[disk_op].mode = disk_dict[constants.IDISK_MODE]
10637
        result.append(("disk.mode/%d" % disk_op,
10638
                       disk_dict[constants.IDISK_MODE]))
10639

    
10640
    if self.op.disk_template:
10641
      r_shut = _ShutdownInstanceDisks(self, instance)
10642
      if not r_shut:
10643
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
10644
                                 " proceed with disk template conversion")
10645
      mode = (instance.disk_template, self.op.disk_template)
10646
      try:
10647
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
10648
      except:
10649
        self.cfg.ReleaseDRBDMinors(instance.name)
10650
        raise
10651
      result.append(("disk_template", self.op.disk_template))
10652

    
10653
    # NIC changes
10654
    for nic_op, nic_dict in self.op.nics:
10655
      if nic_op == constants.DDM_REMOVE:
10656
        # remove the last nic
10657
        del instance.nics[-1]
10658
        result.append(("nic.%d" % len(instance.nics), "remove"))
10659
      elif nic_op == constants.DDM_ADD:
10660
        # mac and bridge should be set, by now
10661
        mac = nic_dict[constants.INIC_MAC]
10662
        ip = nic_dict.get(constants.INIC_IP, None)
10663
        nicparams = self.nic_pinst[constants.DDM_ADD]
10664
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
10665
        instance.nics.append(new_nic)
10666
        result.append(("nic.%d" % (len(instance.nics) - 1),
10667
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
10668
                       (new_nic.mac, new_nic.ip,
10669
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
10670
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
10671
                       )))
10672
      else:
10673
        for key in (constants.INIC_MAC, constants.INIC_IP):
10674
          if key in nic_dict:
10675
            setattr(instance.nics[nic_op], key, nic_dict[key])
10676
        if nic_op in self.nic_pinst:
10677
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
10678
        for key, val in nic_dict.iteritems():
10679
          result.append(("nic.%s/%d" % (key, nic_op), val))
10680

    
10681
    # hvparams changes
10682
    if self.op.hvparams:
10683
      instance.hvparams = self.hv_inst
10684
      for key, val in self.op.hvparams.iteritems():
10685
        result.append(("hv/%s" % key, val))
10686

    
10687
    # beparams changes
10688
    if self.op.beparams:
10689
      instance.beparams = self.be_inst
10690
      for key, val in self.op.beparams.iteritems():
10691
        result.append(("be/%s" % key, val))
10692

    
10693
    # OS change
10694
    if self.op.os_name:
10695
      instance.os = self.op.os_name
10696

    
10697
    # osparams changes
10698
    if self.op.osparams:
10699
      instance.osparams = self.os_inst
10700
      for key, val in self.op.osparams.iteritems():
10701
        result.append(("os/%s" % key, val))
10702

    
10703
    self.cfg.Update(instance, feedback_fn)
10704

    
10705
    return result
10706

    
10707
  _DISK_CONVERSIONS = {
10708
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
10709
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
10710
    }
10711

    
10712

    
10713
class LUBackupQuery(NoHooksLU):
10714
  """Query the exports list
10715

10716
  """
10717
  REQ_BGL = False
10718

    
10719
  def ExpandNames(self):
10720
    self.needed_locks = {}
10721
    self.share_locks[locking.LEVEL_NODE] = 1
10722
    if not self.op.nodes:
10723
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10724
    else:
10725
      self.needed_locks[locking.LEVEL_NODE] = \
10726
        _GetWantedNodes(self, self.op.nodes)
10727

    
10728
  def Exec(self, feedback_fn):
10729
    """Compute the list of all the exported system images.
10730

10731
    @rtype: dict
10732
    @return: a dictionary with the structure node->(export-list)
10733
        where export-list is a list of the instances exported on
10734
        that node.
10735

10736
    """
10737
    self.nodes = self.glm.list_owned(locking.LEVEL_NODE)
10738
    rpcresult = self.rpc.call_export_list(self.nodes)
10739
    result = {}
10740
    for node in rpcresult:
10741
      if rpcresult[node].fail_msg:
10742
        result[node] = False
10743
      else:
10744
        result[node] = rpcresult[node].payload
10745

    
10746
    return result
10747

    
10748

    
10749
class LUBackupPrepare(NoHooksLU):
10750
  """Prepares an instance for an export and returns useful information.
10751

10752
  """
10753
  REQ_BGL = False
10754

    
10755
  def ExpandNames(self):
10756
    self._ExpandAndLockInstance()
10757

    
10758
  def CheckPrereq(self):
10759
    """Check prerequisites.
10760

10761
    """
10762
    instance_name = self.op.instance_name
10763

    
10764
    self.instance = self.cfg.GetInstanceInfo(instance_name)
10765
    assert self.instance is not None, \
10766
          "Cannot retrieve locked instance %s" % self.op.instance_name
10767
    _CheckNodeOnline(self, self.instance.primary_node)
10768

    
10769
    self._cds = _GetClusterDomainSecret()
10770

    
10771
  def Exec(self, feedback_fn):
10772
    """Prepares an instance for an export.
10773

10774
    """
10775
    instance = self.instance
10776

    
10777
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
10778
      salt = utils.GenerateSecret(8)
10779

    
10780
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
10781
      result = self.rpc.call_x509_cert_create(instance.primary_node,
10782
                                              constants.RIE_CERT_VALIDITY)
10783
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
10784

    
10785
      (name, cert_pem) = result.payload
10786

    
10787
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
10788
                                             cert_pem)
10789

    
10790
      return {
10791
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
10792
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
10793
                          salt),
10794
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
10795
        }
10796

    
10797
    return None
10798

    
10799

    
10800
class LUBackupExport(LogicalUnit):
10801
  """Export an instance to an image in the cluster.
10802

10803
  """
10804
  HPATH = "instance-export"
10805
  HTYPE = constants.HTYPE_INSTANCE
10806
  REQ_BGL = False
10807

    
10808
  def CheckArguments(self):
10809
    """Check the arguments.
10810

10811
    """
10812
    self.x509_key_name = self.op.x509_key_name
10813
    self.dest_x509_ca_pem = self.op.destination_x509_ca
10814

    
10815
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
10816
      if not self.x509_key_name:
10817
        raise errors.OpPrereqError("Missing X509 key name for encryption",
10818
                                   errors.ECODE_INVAL)
10819

    
10820
      if not self.dest_x509_ca_pem:
10821
        raise errors.OpPrereqError("Missing destination X509 CA",
10822
                                   errors.ECODE_INVAL)
10823

    
10824
  def ExpandNames(self):
10825
    self._ExpandAndLockInstance()
10826

    
10827
    # Lock all nodes for local exports
10828
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10829
      # FIXME: lock only instance primary and destination node
10830
      #
10831
      # Sad but true, for now we have do lock all nodes, as we don't know where
10832
      # the previous export might be, and in this LU we search for it and
10833
      # remove it from its current node. In the future we could fix this by:
10834
      #  - making a tasklet to search (share-lock all), then create the
10835
      #    new one, then one to remove, after
10836
      #  - removing the removal operation altogether
10837
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10838

    
10839
  def DeclareLocks(self, level):
10840
    """Last minute lock declaration."""
10841
    # All nodes are locked anyway, so nothing to do here.
10842

    
10843
  def BuildHooksEnv(self):
10844
    """Build hooks env.
10845

10846
    This will run on the master, primary node and target node.
10847

10848
    """
10849
    env = {
10850
      "EXPORT_MODE": self.op.mode,
10851
      "EXPORT_NODE": self.op.target_node,
10852
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
10853
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
10854
      # TODO: Generic function for boolean env variables
10855
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
10856
      }
10857

    
10858
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
10859

    
10860
    return env
10861

    
10862
  def BuildHooksNodes(self):
10863
    """Build hooks nodes.
10864

10865
    """
10866
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
10867

    
10868
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10869
      nl.append(self.op.target_node)
10870

    
10871
    return (nl, nl)
10872

    
10873
  def CheckPrereq(self):
10874
    """Check prerequisites.
10875

10876
    This checks that the instance and node names are valid.
10877

10878
    """
10879
    instance_name = self.op.instance_name
10880

    
10881
    self.instance = self.cfg.GetInstanceInfo(instance_name)
10882
    assert self.instance is not None, \
10883
          "Cannot retrieve locked instance %s" % self.op.instance_name
10884
    _CheckNodeOnline(self, self.instance.primary_node)
10885

    
10886
    if (self.op.remove_instance and self.instance.admin_up and
10887
        not self.op.shutdown):
10888
      raise errors.OpPrereqError("Can not remove instance without shutting it"
10889
                                 " down before")
10890

    
10891
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10892
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
10893
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
10894
      assert self.dst_node is not None
10895

    
10896
      _CheckNodeOnline(self, self.dst_node.name)
10897
      _CheckNodeNotDrained(self, self.dst_node.name)
10898

    
10899
      self._cds = None
10900
      self.dest_disk_info = None
10901
      self.dest_x509_ca = None
10902

    
10903
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10904
      self.dst_node = None
10905

    
10906
      if len(self.op.target_node) != len(self.instance.disks):
10907
        raise errors.OpPrereqError(("Received destination information for %s"
10908
                                    " disks, but instance %s has %s disks") %
10909
                                   (len(self.op.target_node), instance_name,
10910
                                    len(self.instance.disks)),
10911
                                   errors.ECODE_INVAL)
10912

    
10913
      cds = _GetClusterDomainSecret()
10914

    
10915
      # Check X509 key name
10916
      try:
10917
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
10918
      except (TypeError, ValueError), err:
10919
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
10920

    
10921
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
10922
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
10923
                                   errors.ECODE_INVAL)
10924

    
10925
      # Load and verify CA
10926
      try:
10927
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
10928
      except OpenSSL.crypto.Error, err:
10929
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
10930
                                   (err, ), errors.ECODE_INVAL)
10931

    
10932
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
10933
      if errcode is not None:
10934
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
10935
                                   (msg, ), errors.ECODE_INVAL)
10936

    
10937
      self.dest_x509_ca = cert
10938

    
10939
      # Verify target information
10940
      disk_info = []
10941
      for idx, disk_data in enumerate(self.op.target_node):
10942
        try:
10943
          (host, port, magic) = \
10944
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
10945
        except errors.GenericError, err:
10946
          raise errors.OpPrereqError("Target info for disk %s: %s" %
10947
                                     (idx, err), errors.ECODE_INVAL)
10948

    
10949
        disk_info.append((host, port, magic))
10950

    
10951
      assert len(disk_info) == len(self.op.target_node)
10952
      self.dest_disk_info = disk_info
10953

    
10954
    else:
10955
      raise errors.ProgrammerError("Unhandled export mode %r" %
10956
                                   self.op.mode)
10957

    
10958
    # instance disk type verification
10959
    # TODO: Implement export support for file-based disks
10960
    for disk in self.instance.disks:
10961
      if disk.dev_type == constants.LD_FILE:
10962
        raise errors.OpPrereqError("Export not supported for instances with"
10963
                                   " file-based disks", errors.ECODE_INVAL)
10964

    
10965
  def _CleanupExports(self, feedback_fn):
10966
    """Removes exports of current instance from all other nodes.
10967

10968
    If an instance in a cluster with nodes A..D was exported to node C, its
10969
    exports will be removed from the nodes A, B and D.
10970

10971
    """
10972
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
10973

    
10974
    nodelist = self.cfg.GetNodeList()
10975
    nodelist.remove(self.dst_node.name)
10976

    
10977
    # on one-node clusters nodelist will be empty after the removal
10978
    # if we proceed the backup would be removed because OpBackupQuery
10979
    # substitutes an empty list with the full cluster node list.
10980
    iname = self.instance.name
10981
    if nodelist:
10982
      feedback_fn("Removing old exports for instance %s" % iname)
10983
      exportlist = self.rpc.call_export_list(nodelist)
10984
      for node in exportlist:
10985
        if exportlist[node].fail_msg:
10986
          continue
10987
        if iname in exportlist[node].payload:
10988
          msg = self.rpc.call_export_remove(node, iname).fail_msg
10989
          if msg:
10990
            self.LogWarning("Could not remove older export for instance %s"
10991
                            " on node %s: %s", iname, node, msg)
10992

    
10993
  def Exec(self, feedback_fn):
10994
    """Export an instance to an image in the cluster.
10995

10996
    """
10997
    assert self.op.mode in constants.EXPORT_MODES
10998

    
10999
    instance = self.instance
11000
    src_node = instance.primary_node
11001

    
11002
    if self.op.shutdown:
11003
      # shutdown the instance, but not the disks
11004
      feedback_fn("Shutting down instance %s" % instance.name)
11005
      result = self.rpc.call_instance_shutdown(src_node, instance,
11006
                                               self.op.shutdown_timeout)
11007
      # TODO: Maybe ignore failures if ignore_remove_failures is set
11008
      result.Raise("Could not shutdown instance %s on"
11009
                   " node %s" % (instance.name, src_node))
11010

    
11011
    # set the disks ID correctly since call_instance_start needs the
11012
    # correct drbd minor to create the symlinks
11013
    for disk in instance.disks:
11014
      self.cfg.SetDiskID(disk, src_node)
11015

    
11016
    activate_disks = (not instance.admin_up)
11017

    
11018
    if activate_disks:
11019
      # Activate the instance disks if we'exporting a stopped instance
11020
      feedback_fn("Activating disks for %s" % instance.name)
11021
      _StartInstanceDisks(self, instance, None)
11022

    
11023
    try:
11024
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
11025
                                                     instance)
11026

    
11027
      helper.CreateSnapshots()
11028
      try:
11029
        if (self.op.shutdown and instance.admin_up and
11030
            not self.op.remove_instance):
11031
          assert not activate_disks
11032
          feedback_fn("Starting instance %s" % instance.name)
11033
          result = self.rpc.call_instance_start(src_node, instance, None, None)
11034
          msg = result.fail_msg
11035
          if msg:
11036
            feedback_fn("Failed to start instance: %s" % msg)
11037
            _ShutdownInstanceDisks(self, instance)
11038
            raise errors.OpExecError("Could not start instance: %s" % msg)
11039

    
11040
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
11041
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
11042
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
11043
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
11044
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
11045

    
11046
          (key_name, _, _) = self.x509_key_name
11047

    
11048
          dest_ca_pem = \
11049
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
11050
                                            self.dest_x509_ca)
11051

    
11052
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
11053
                                                     key_name, dest_ca_pem,
11054
                                                     timeouts)
11055
      finally:
11056
        helper.Cleanup()
11057

    
11058
      # Check for backwards compatibility
11059
      assert len(dresults) == len(instance.disks)
11060
      assert compat.all(isinstance(i, bool) for i in dresults), \
11061
             "Not all results are boolean: %r" % dresults
11062

    
11063
    finally:
11064
      if activate_disks:
11065
        feedback_fn("Deactivating disks for %s" % instance.name)
11066
        _ShutdownInstanceDisks(self, instance)
11067

    
11068
    if not (compat.all(dresults) and fin_resu):
11069
      failures = []
11070
      if not fin_resu:
11071
        failures.append("export finalization")
11072
      if not compat.all(dresults):
11073
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
11074
                               if not dsk)
11075
        failures.append("disk export: disk(s) %s" % fdsk)
11076

    
11077
      raise errors.OpExecError("Export failed, errors in %s" %
11078
                               utils.CommaJoin(failures))
11079

    
11080
    # At this point, the export was successful, we can cleanup/finish
11081

    
11082
    # Remove instance if requested
11083
    if self.op.remove_instance:
11084
      feedback_fn("Removing instance %s" % instance.name)
11085
      _RemoveInstance(self, feedback_fn, instance,
11086
                      self.op.ignore_remove_failures)
11087

    
11088
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
11089
      self._CleanupExports(feedback_fn)
11090

    
11091
    return fin_resu, dresults
11092

    
11093

    
11094
class LUBackupRemove(NoHooksLU):
11095
  """Remove exports related to the named instance.
11096

11097
  """
11098
  REQ_BGL = False
11099

    
11100
  def ExpandNames(self):
11101
    self.needed_locks = {}
11102
    # We need all nodes to be locked in order for RemoveExport to work, but we
11103
    # don't need to lock the instance itself, as nothing will happen to it (and
11104
    # we can remove exports also for a removed instance)
11105
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
11106

    
11107
  def Exec(self, feedback_fn):
11108
    """Remove any export.
11109

11110
    """
11111
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
11112
    # If the instance was not found we'll try with the name that was passed in.
11113
    # This will only work if it was an FQDN, though.
11114
    fqdn_warn = False
11115
    if not instance_name:
11116
      fqdn_warn = True
11117
      instance_name = self.op.instance_name
11118

    
11119
    locked_nodes = self.glm.list_owned(locking.LEVEL_NODE)
11120
    exportlist = self.rpc.call_export_list(locked_nodes)
11121
    found = False
11122
    for node in exportlist:
11123
      msg = exportlist[node].fail_msg
11124
      if msg:
11125
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
11126
        continue
11127
      if instance_name in exportlist[node].payload:
11128
        found = True
11129
        result = self.rpc.call_export_remove(node, instance_name)
11130
        msg = result.fail_msg
11131
        if msg:
11132
          logging.error("Could not remove export for instance %s"
11133
                        " on node %s: %s", instance_name, node, msg)
11134

    
11135
    if fqdn_warn and not found:
11136
      feedback_fn("Export not found. If trying to remove an export belonging"
11137
                  " to a deleted instance please use its Fully Qualified"
11138
                  " Domain Name.")
11139

    
11140

    
11141
class LUGroupAdd(LogicalUnit):
11142
  """Logical unit for creating node groups.
11143

11144
  """
11145
  HPATH = "group-add"
11146
  HTYPE = constants.HTYPE_GROUP
11147
  REQ_BGL = False
11148

    
11149
  def ExpandNames(self):
11150
    # We need the new group's UUID here so that we can create and acquire the
11151
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
11152
    # that it should not check whether the UUID exists in the configuration.
11153
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
11154
    self.needed_locks = {}
11155
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
11156

    
11157
  def CheckPrereq(self):
11158
    """Check prerequisites.
11159

11160
    This checks that the given group name is not an existing node group
11161
    already.
11162

11163
    """
11164
    try:
11165
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11166
    except errors.OpPrereqError:
11167
      pass
11168
    else:
11169
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
11170
                                 " node group (UUID: %s)" %
11171
                                 (self.op.group_name, existing_uuid),
11172
                                 errors.ECODE_EXISTS)
11173

    
11174
    if self.op.ndparams:
11175
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
11176

    
11177
  def BuildHooksEnv(self):
11178
    """Build hooks env.
11179

11180
    """
11181
    return {
11182
      "GROUP_NAME": self.op.group_name,
11183
      }
11184

    
11185
  def BuildHooksNodes(self):
11186
    """Build hooks nodes.
11187

11188
    """
11189
    mn = self.cfg.GetMasterNode()
11190
    return ([mn], [mn])
11191

    
11192
  def Exec(self, feedback_fn):
11193
    """Add the node group to the cluster.
11194

11195
    """
11196
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
11197
                                  uuid=self.group_uuid,
11198
                                  alloc_policy=self.op.alloc_policy,
11199
                                  ndparams=self.op.ndparams)
11200

    
11201
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
11202
    del self.remove_locks[locking.LEVEL_NODEGROUP]
11203

    
11204

    
11205
class LUGroupAssignNodes(NoHooksLU):
11206
  """Logical unit for assigning nodes to groups.
11207

11208
  """
11209
  REQ_BGL = False
11210

    
11211
  def ExpandNames(self):
11212
    # These raise errors.OpPrereqError on their own:
11213
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11214
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
11215

    
11216
    # We want to lock all the affected nodes and groups. We have readily
11217
    # available the list of nodes, and the *destination* group. To gather the
11218
    # list of "source" groups, we need to fetch node information later on.
11219
    self.needed_locks = {
11220
      locking.LEVEL_NODEGROUP: set([self.group_uuid]),
11221
      locking.LEVEL_NODE: self.op.nodes,
11222
      }
11223

    
11224
  def DeclareLocks(self, level):
11225
    if level == locking.LEVEL_NODEGROUP:
11226
      assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
11227

    
11228
      # Try to get all affected nodes' groups without having the group or node
11229
      # lock yet. Needs verification later in the code flow.
11230
      groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
11231

    
11232
      self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
11233

    
11234
  def CheckPrereq(self):
11235
    """Check prerequisites.
11236

11237
    """
11238
    assert self.needed_locks[locking.LEVEL_NODEGROUP]
11239
    assert (frozenset(self.glm.list_owned(locking.LEVEL_NODE)) ==
11240
            frozenset(self.op.nodes))
11241

    
11242
    expected_locks = (set([self.group_uuid]) |
11243
                      self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
11244
    actual_locks = self.glm.list_owned(locking.LEVEL_NODEGROUP)
11245
    if actual_locks != expected_locks:
11246
      raise errors.OpExecError("Nodes changed groups since locks were acquired,"
11247
                               " current groups are '%s', used to be '%s'" %
11248
                               (utils.CommaJoin(expected_locks),
11249
                                utils.CommaJoin(actual_locks)))
11250

    
11251
    self.node_data = self.cfg.GetAllNodesInfo()
11252
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
11253
    instance_data = self.cfg.GetAllInstancesInfo()
11254

    
11255
    if self.group is None:
11256
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
11257
                               (self.op.group_name, self.group_uuid))
11258

    
11259
    (new_splits, previous_splits) = \
11260
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
11261
                                             for node in self.op.nodes],
11262
                                            self.node_data, instance_data)
11263

    
11264
    if new_splits:
11265
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
11266

    
11267
      if not self.op.force:
11268
        raise errors.OpExecError("The following instances get split by this"
11269
                                 " change and --force was not given: %s" %
11270
                                 fmt_new_splits)
11271
      else:
11272
        self.LogWarning("This operation will split the following instances: %s",
11273
                        fmt_new_splits)
11274

    
11275
        if previous_splits:
11276
          self.LogWarning("In addition, these already-split instances continue"
11277
                          " to be split across groups: %s",
11278
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
11279

    
11280
  def Exec(self, feedback_fn):
11281
    """Assign nodes to a new group.
11282

11283
    """
11284
    for node in self.op.nodes:
11285
      self.node_data[node].group = self.group_uuid
11286

    
11287
    # FIXME: Depends on side-effects of modifying the result of
11288
    # C{cfg.GetAllNodesInfo}
11289

    
11290
    self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
11291

    
11292
  @staticmethod
11293
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
11294
    """Check for split instances after a node assignment.
11295

11296
    This method considers a series of node assignments as an atomic operation,
11297
    and returns information about split instances after applying the set of
11298
    changes.
11299

11300
    In particular, it returns information about newly split instances, and
11301
    instances that were already split, and remain so after the change.
11302

11303
    Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
11304
    considered.
11305

11306
    @type changes: list of (node_name, new_group_uuid) pairs.
11307
    @param changes: list of node assignments to consider.
11308
    @param node_data: a dict with data for all nodes
11309
    @param instance_data: a dict with all instances to consider
11310
    @rtype: a two-tuple
11311
    @return: a list of instances that were previously okay and result split as a
11312
      consequence of this change, and a list of instances that were previously
11313
      split and this change does not fix.
11314

11315
    """
11316
    changed_nodes = dict((node, group) for node, group in changes
11317
                         if node_data[node].group != group)
11318

    
11319
    all_split_instances = set()
11320
    previously_split_instances = set()
11321

    
11322
    def InstanceNodes(instance):
11323
      return [instance.primary_node] + list(instance.secondary_nodes)
11324

    
11325
    for inst in instance_data.values():
11326
      if inst.disk_template not in constants.DTS_INT_MIRROR:
11327
        continue
11328

    
11329
      instance_nodes = InstanceNodes(inst)
11330

    
11331
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
11332
        previously_split_instances.add(inst.name)
11333

    
11334
      if len(set(changed_nodes.get(node, node_data[node].group)
11335
                 for node in instance_nodes)) > 1:
11336
        all_split_instances.add(inst.name)
11337

    
11338
    return (list(all_split_instances - previously_split_instances),
11339
            list(previously_split_instances & all_split_instances))
11340

    
11341

    
11342
class _GroupQuery(_QueryBase):
11343
  FIELDS = query.GROUP_FIELDS
11344

    
11345
  def ExpandNames(self, lu):
11346
    lu.needed_locks = {}
11347

    
11348
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
11349
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
11350

    
11351
    if not self.names:
11352
      self.wanted = [name_to_uuid[name]
11353
                     for name in utils.NiceSort(name_to_uuid.keys())]
11354
    else:
11355
      # Accept names to be either names or UUIDs.
11356
      missing = []
11357
      self.wanted = []
11358
      all_uuid = frozenset(self._all_groups.keys())
11359

    
11360
      for name in self.names:
11361
        if name in all_uuid:
11362
          self.wanted.append(name)
11363
        elif name in name_to_uuid:
11364
          self.wanted.append(name_to_uuid[name])
11365
        else:
11366
          missing.append(name)
11367

    
11368
      if missing:
11369
        raise errors.OpPrereqError("Some groups do not exist: %s" %
11370
                                   utils.CommaJoin(missing),
11371
                                   errors.ECODE_NOENT)
11372

    
11373
  def DeclareLocks(self, lu, level):
11374
    pass
11375

    
11376
  def _GetQueryData(self, lu):
11377
    """Computes the list of node groups and their attributes.
11378

11379
    """
11380
    do_nodes = query.GQ_NODE in self.requested_data
11381
    do_instances = query.GQ_INST in self.requested_data
11382

    
11383
    group_to_nodes = None
11384
    group_to_instances = None
11385

    
11386
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
11387
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
11388
    # latter GetAllInstancesInfo() is not enough, for we have to go through
11389
    # instance->node. Hence, we will need to process nodes even if we only need
11390
    # instance information.
11391
    if do_nodes or do_instances:
11392
      all_nodes = lu.cfg.GetAllNodesInfo()
11393
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
11394
      node_to_group = {}
11395

    
11396
      for node in all_nodes.values():
11397
        if node.group in group_to_nodes:
11398
          group_to_nodes[node.group].append(node.name)
11399
          node_to_group[node.name] = node.group
11400

    
11401
      if do_instances:
11402
        all_instances = lu.cfg.GetAllInstancesInfo()
11403
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
11404

    
11405
        for instance in all_instances.values():
11406
          node = instance.primary_node
11407
          if node in node_to_group:
11408
            group_to_instances[node_to_group[node]].append(instance.name)
11409

    
11410
        if not do_nodes:
11411
          # Do not pass on node information if it was not requested.
11412
          group_to_nodes = None
11413

    
11414
    return query.GroupQueryData([self._all_groups[uuid]
11415
                                 for uuid in self.wanted],
11416
                                group_to_nodes, group_to_instances)
11417

    
11418

    
11419
class LUGroupQuery(NoHooksLU):
11420
  """Logical unit for querying node groups.
11421

11422
  """
11423
  REQ_BGL = False
11424

    
11425
  def CheckArguments(self):
11426
    self.gq = _GroupQuery(qlang.MakeSimpleFilter("name", self.op.names),
11427
                          self.op.output_fields, False)
11428

    
11429
  def ExpandNames(self):
11430
    self.gq.ExpandNames(self)
11431

    
11432
  def Exec(self, feedback_fn):
11433
    return self.gq.OldStyleQuery(self)
11434

    
11435

    
11436
class LUGroupSetParams(LogicalUnit):
11437
  """Modifies the parameters of a node group.
11438

11439
  """
11440
  HPATH = "group-modify"
11441
  HTYPE = constants.HTYPE_GROUP
11442
  REQ_BGL = False
11443

    
11444
  def CheckArguments(self):
11445
    all_changes = [
11446
      self.op.ndparams,
11447
      self.op.alloc_policy,
11448
      ]
11449

    
11450
    if all_changes.count(None) == len(all_changes):
11451
      raise errors.OpPrereqError("Please pass at least one modification",
11452
                                 errors.ECODE_INVAL)
11453

    
11454
  def ExpandNames(self):
11455
    # This raises errors.OpPrereqError on its own:
11456
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11457

    
11458
    self.needed_locks = {
11459
      locking.LEVEL_NODEGROUP: [self.group_uuid],
11460
      }
11461

    
11462
  def CheckPrereq(self):
11463
    """Check prerequisites.
11464

11465
    """
11466
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
11467

    
11468
    if self.group is None:
11469
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
11470
                               (self.op.group_name, self.group_uuid))
11471

    
11472
    if self.op.ndparams:
11473
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
11474
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
11475
      self.new_ndparams = new_ndparams
11476

    
11477
  def BuildHooksEnv(self):
11478
    """Build hooks env.
11479

11480
    """
11481
    return {
11482
      "GROUP_NAME": self.op.group_name,
11483
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
11484
      }
11485

    
11486
  def BuildHooksNodes(self):
11487
    """Build hooks nodes.
11488

11489
    """
11490
    mn = self.cfg.GetMasterNode()
11491
    return ([mn], [mn])
11492

    
11493
  def Exec(self, feedback_fn):
11494
    """Modifies the node group.
11495

11496
    """
11497
    result = []
11498

    
11499
    if self.op.ndparams:
11500
      self.group.ndparams = self.new_ndparams
11501
      result.append(("ndparams", str(self.group.ndparams)))
11502

    
11503
    if self.op.alloc_policy:
11504
      self.group.alloc_policy = self.op.alloc_policy
11505

    
11506
    self.cfg.Update(self.group, feedback_fn)
11507
    return result
11508

    
11509

    
11510

    
11511
class LUGroupRemove(LogicalUnit):
11512
  HPATH = "group-remove"
11513
  HTYPE = constants.HTYPE_GROUP
11514
  REQ_BGL = False
11515

    
11516
  def ExpandNames(self):
11517
    # This will raises errors.OpPrereqError on its own:
11518
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11519
    self.needed_locks = {
11520
      locking.LEVEL_NODEGROUP: [self.group_uuid],
11521
      }
11522

    
11523
  def CheckPrereq(self):
11524
    """Check prerequisites.
11525

11526
    This checks that the given group name exists as a node group, that is
11527
    empty (i.e., contains no nodes), and that is not the last group of the
11528
    cluster.
11529

11530
    """
11531
    # Verify that the group is empty.
11532
    group_nodes = [node.name
11533
                   for node in self.cfg.GetAllNodesInfo().values()
11534
                   if node.group == self.group_uuid]
11535

    
11536
    if group_nodes:
11537
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
11538
                                 " nodes: %s" %
11539
                                 (self.op.group_name,
11540
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
11541
                                 errors.ECODE_STATE)
11542

    
11543
    # Verify the cluster would not be left group-less.
11544
    if len(self.cfg.GetNodeGroupList()) == 1:
11545
      raise errors.OpPrereqError("Group '%s' is the only group,"
11546
                                 " cannot be removed" %
11547
                                 self.op.group_name,
11548
                                 errors.ECODE_STATE)
11549

    
11550
  def BuildHooksEnv(self):
11551
    """Build hooks env.
11552

11553
    """
11554
    return {
11555
      "GROUP_NAME": self.op.group_name,
11556
      }
11557

    
11558
  def BuildHooksNodes(self):
11559
    """Build hooks nodes.
11560

11561
    """
11562
    mn = self.cfg.GetMasterNode()
11563
    return ([mn], [mn])
11564

    
11565
  def Exec(self, feedback_fn):
11566
    """Remove the node group.
11567

11568
    """
11569
    try:
11570
      self.cfg.RemoveNodeGroup(self.group_uuid)
11571
    except errors.ConfigurationError:
11572
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
11573
                               (self.op.group_name, self.group_uuid))
11574

    
11575
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
11576

    
11577

    
11578
class LUGroupRename(LogicalUnit):
11579
  HPATH = "group-rename"
11580
  HTYPE = constants.HTYPE_GROUP
11581
  REQ_BGL = False
11582

    
11583
  def ExpandNames(self):
11584
    # This raises errors.OpPrereqError on its own:
11585
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
11586

    
11587
    self.needed_locks = {
11588
      locking.LEVEL_NODEGROUP: [self.group_uuid],
11589
      }
11590

    
11591
  def CheckPrereq(self):
11592
    """Check prerequisites.
11593

11594
    Ensures requested new name is not yet used.
11595

11596
    """
11597
    try:
11598
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
11599
    except errors.OpPrereqError:
11600
      pass
11601
    else:
11602
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
11603
                                 " node group (UUID: %s)" %
11604
                                 (self.op.new_name, new_name_uuid),
11605
                                 errors.ECODE_EXISTS)
11606

    
11607
  def BuildHooksEnv(self):
11608
    """Build hooks env.
11609

11610
    """
11611
    return {
11612
      "OLD_NAME": self.op.group_name,
11613
      "NEW_NAME": self.op.new_name,
11614
      }
11615

    
11616
  def BuildHooksNodes(self):
11617
    """Build hooks nodes.
11618

11619
    """
11620
    mn = self.cfg.GetMasterNode()
11621

    
11622
    all_nodes = self.cfg.GetAllNodesInfo()
11623
    all_nodes.pop(mn, None)
11624

    
11625
    run_nodes = [mn]
11626
    run_nodes.extend(node.name for node in all_nodes.values()
11627
                     if node.group == self.group_uuid)
11628

    
11629
    return (run_nodes, run_nodes)
11630

    
11631
  def Exec(self, feedback_fn):
11632
    """Rename the node group.
11633

11634
    """
11635
    group = self.cfg.GetNodeGroup(self.group_uuid)
11636

    
11637
    if group is None:
11638
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
11639
                               (self.op.group_name, self.group_uuid))
11640

    
11641
    group.name = self.op.new_name
11642
    self.cfg.Update(group, feedback_fn)
11643

    
11644
    return self.op.new_name
11645

    
11646

    
11647
class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
11648
  """Generic tags LU.
11649

11650
  This is an abstract class which is the parent of all the other tags LUs.
11651

11652
  """
11653
  def ExpandNames(self):
11654
    self.group_uuid = None
11655
    self.needed_locks = {}
11656
    if self.op.kind == constants.TAG_NODE:
11657
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
11658
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
11659
    elif self.op.kind == constants.TAG_INSTANCE:
11660
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
11661
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
11662
    elif self.op.kind == constants.TAG_NODEGROUP:
11663
      self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)
11664

    
11665
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
11666
    # not possible to acquire the BGL based on opcode parameters)
11667

    
11668
  def CheckPrereq(self):
11669
    """Check prerequisites.
11670

11671
    """
11672
    if self.op.kind == constants.TAG_CLUSTER:
11673
      self.target = self.cfg.GetClusterInfo()
11674
    elif self.op.kind == constants.TAG_NODE:
11675
      self.target = self.cfg.GetNodeInfo(self.op.name)
11676
    elif self.op.kind == constants.TAG_INSTANCE:
11677
      self.target = self.cfg.GetInstanceInfo(self.op.name)
11678
    elif self.op.kind == constants.TAG_NODEGROUP:
11679
      self.target = self.cfg.GetNodeGroup(self.group_uuid)
11680
    else:
11681
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
11682
                                 str(self.op.kind), errors.ECODE_INVAL)
11683

    
11684

    
11685
class LUTagsGet(TagsLU):
11686
  """Returns the tags of a given object.
11687

11688
  """
11689
  REQ_BGL = False
11690

    
11691
  def ExpandNames(self):
11692
    TagsLU.ExpandNames(self)
11693

    
11694
    # Share locks as this is only a read operation
11695
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
11696

    
11697
  def Exec(self, feedback_fn):
11698
    """Returns the tag list.
11699

11700
    """
11701
    return list(self.target.GetTags())
11702

    
11703

    
11704
class LUTagsSearch(NoHooksLU):
11705
  """Searches the tags for a given pattern.
11706

11707
  """
11708
  REQ_BGL = False
11709

    
11710
  def ExpandNames(self):
11711
    self.needed_locks = {}
11712

    
11713
  def CheckPrereq(self):
11714
    """Check prerequisites.
11715

11716
    This checks the pattern passed for validity by compiling it.
11717

11718
    """
11719
    try:
11720
      self.re = re.compile(self.op.pattern)
11721
    except re.error, err:
11722
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
11723
                                 (self.op.pattern, err), errors.ECODE_INVAL)
11724

    
11725
  def Exec(self, feedback_fn):
11726
    """Returns the tag list.
11727

11728
    """
11729
    cfg = self.cfg
11730
    tgts = [("/cluster", cfg.GetClusterInfo())]
11731
    ilist = cfg.GetAllInstancesInfo().values()
11732
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
11733
    nlist = cfg.GetAllNodesInfo().values()
11734
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
11735
    tgts.extend(("/nodegroup/%s" % n.name, n)
11736
                for n in cfg.GetAllNodeGroupsInfo().values())
11737
    results = []
11738
    for path, target in tgts:
11739
      for tag in target.GetTags():
11740
        if self.re.search(tag):
11741
          results.append((path, tag))
11742
    return results
11743

    
11744

    
11745
class LUTagsSet(TagsLU):
11746
  """Sets a tag on a given object.
11747

11748
  """
11749
  REQ_BGL = False
11750

    
11751
  def CheckPrereq(self):
11752
    """Check prerequisites.
11753

11754
    This checks the type and length of the tag name and value.
11755

11756
    """
11757
    TagsLU.CheckPrereq(self)
11758
    for tag in self.op.tags:
11759
      objects.TaggableObject.ValidateTag(tag)
11760

    
11761
  def Exec(self, feedback_fn):
11762
    """Sets the tag.
11763

11764
    """
11765
    try:
11766
      for tag in self.op.tags:
11767
        self.target.AddTag(tag)
11768
    except errors.TagError, err:
11769
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
11770
    self.cfg.Update(self.target, feedback_fn)
11771

    
11772

    
11773
class LUTagsDel(TagsLU):
11774
  """Delete a list of tags from a given object.
11775

11776
  """
11777
  REQ_BGL = False
11778

    
11779
  def CheckPrereq(self):
11780
    """Check prerequisites.
11781

11782
    This checks that we have the given tag.
11783

11784
    """
11785
    TagsLU.CheckPrereq(self)
11786
    for tag in self.op.tags:
11787
      objects.TaggableObject.ValidateTag(tag)
11788
    del_tags = frozenset(self.op.tags)
11789
    cur_tags = self.target.GetTags()
11790

    
11791
    diff_tags = del_tags - cur_tags
11792
    if diff_tags:
11793
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
11794
      raise errors.OpPrereqError("Tag(s) %s not found" %
11795
                                 (utils.CommaJoin(diff_names), ),
11796
                                 errors.ECODE_NOENT)
11797

    
11798
  def Exec(self, feedback_fn):
11799
    """Remove the tag from the object.
11800

11801
    """
11802
    for tag in self.op.tags:
11803
      self.target.RemoveTag(tag)
11804
    self.cfg.Update(self.target, feedback_fn)
11805

    
11806

    
11807
class LUTestDelay(NoHooksLU):
11808
  """Sleep for a specified amount of time.
11809

11810
  This LU sleeps on the master and/or nodes for a specified amount of
11811
  time.
11812

11813
  """
11814
  REQ_BGL = False
11815

    
11816
  def ExpandNames(self):
11817
    """Expand names and set required locks.
11818

11819
    This expands the node list, if any.
11820

11821
    """
11822
    self.needed_locks = {}
11823
    if self.op.on_nodes:
11824
      # _GetWantedNodes can be used here, but is not always appropriate to use
11825
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
11826
      # more information.
11827
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
11828
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
11829

    
11830
  def _TestDelay(self):
11831
    """Do the actual sleep.
11832

11833
    """
11834
    if self.op.on_master:
11835
      if not utils.TestDelay(self.op.duration):
11836
        raise errors.OpExecError("Error during master delay test")
11837
    if self.op.on_nodes:
11838
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
11839
      for node, node_result in result.items():
11840
        node_result.Raise("Failure during rpc call to node %s" % node)
11841

    
11842
  def Exec(self, feedback_fn):
11843
    """Execute the test delay opcode, with the wanted repetitions.
11844

11845
    """
11846
    if self.op.repeat == 0:
11847
      self._TestDelay()
11848
    else:
11849
      top_value = self.op.repeat - 1
11850
      for i in range(self.op.repeat):
11851
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
11852
        self._TestDelay()
11853

    
11854

    
11855
class LUTestJqueue(NoHooksLU):
11856
  """Utility LU to test some aspects of the job queue.
11857

11858
  """
11859
  REQ_BGL = False
11860

    
11861
  # Must be lower than default timeout for WaitForJobChange to see whether it
11862
  # notices changed jobs
11863
  _CLIENT_CONNECT_TIMEOUT = 20.0
11864
  _CLIENT_CONFIRM_TIMEOUT = 60.0
11865

    
11866
  @classmethod
11867
  def _NotifyUsingSocket(cls, cb, errcls):
11868
    """Opens a Unix socket and waits for another program to connect.
11869

11870
    @type cb: callable
11871
    @param cb: Callback to send socket name to client
11872
    @type errcls: class
11873
    @param errcls: Exception class to use for errors
11874

11875
    """
11876
    # Using a temporary directory as there's no easy way to create temporary
11877
    # sockets without writing a custom loop around tempfile.mktemp and
11878
    # socket.bind
11879
    tmpdir = tempfile.mkdtemp()
11880
    try:
11881
      tmpsock = utils.PathJoin(tmpdir, "sock")
11882

    
11883
      logging.debug("Creating temporary socket at %s", tmpsock)
11884
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
11885
      try:
11886
        sock.bind(tmpsock)
11887
        sock.listen(1)
11888

    
11889
        # Send details to client
11890
        cb(tmpsock)
11891

    
11892
        # Wait for client to connect before continuing
11893
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
11894
        try:
11895
          (conn, _) = sock.accept()
11896
        except socket.error, err:
11897
          raise errcls("Client didn't connect in time (%s)" % err)
11898
      finally:
11899
        sock.close()
11900
    finally:
11901
      # Remove as soon as client is connected
11902
      shutil.rmtree(tmpdir)
11903

    
11904
    # Wait for client to close
11905
    try:
11906
      try:
11907
        # pylint: disable-msg=E1101
11908
        # Instance of '_socketobject' has no ... member
11909
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
11910
        conn.recv(1)
11911
      except socket.error, err:
11912
        raise errcls("Client failed to confirm notification (%s)" % err)
11913
    finally:
11914
      conn.close()
11915

    
11916
  def _SendNotification(self, test, arg, sockname):
11917
    """Sends a notification to the client.
11918

11919
    @type test: string
11920
    @param test: Test name
11921
    @param arg: Test argument (depends on test)
11922
    @type sockname: string
11923
    @param sockname: Socket path
11924

11925
    """
11926
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
11927

    
11928
  def _Notify(self, prereq, test, arg):
11929
    """Notifies the client of a test.
11930

11931
    @type prereq: bool
11932
    @param prereq: Whether this is a prereq-phase test
11933
    @type test: string
11934
    @param test: Test name
11935
    @param arg: Test argument (depends on test)
11936

11937
    """
11938
    if prereq:
11939
      errcls = errors.OpPrereqError
11940
    else:
11941
      errcls = errors.OpExecError
11942

    
11943
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
11944
                                                  test, arg),
11945
                                   errcls)
11946

    
11947
  def CheckArguments(self):
11948
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
11949
    self.expandnames_calls = 0
11950

    
11951
  def ExpandNames(self):
11952
    checkargs_calls = getattr(self, "checkargs_calls", 0)
11953
    if checkargs_calls < 1:
11954
      raise errors.ProgrammerError("CheckArguments was not called")
11955

    
11956
    self.expandnames_calls += 1
11957

    
11958
    if self.op.notify_waitlock:
11959
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
11960

    
11961
    self.LogInfo("Expanding names")
11962

    
11963
    # Get lock on master node (just to get a lock, not for a particular reason)
11964
    self.needed_locks = {
11965
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
11966
      }
11967

    
11968
  def Exec(self, feedback_fn):
11969
    if self.expandnames_calls < 1:
11970
      raise errors.ProgrammerError("ExpandNames was not called")
11971

    
11972
    if self.op.notify_exec:
11973
      self._Notify(False, constants.JQT_EXEC, None)
11974

    
11975
    self.LogInfo("Executing")
11976

    
11977
    if self.op.log_messages:
11978
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
11979
      for idx, msg in enumerate(self.op.log_messages):
11980
        self.LogInfo("Sending log message %s", idx + 1)
11981
        feedback_fn(constants.JQT_MSGPREFIX + msg)
11982
        # Report how many test messages have been sent
11983
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
11984

    
11985
    if self.op.fail:
11986
      raise errors.OpExecError("Opcode failure was requested")
11987

    
11988
    return True
11989

    
11990

    
11991
class IAllocator(object):
11992
  """IAllocator framework.
11993

11994
  An IAllocator instance has three sets of attributes:
11995
    - cfg that is needed to query the cluster
11996
    - input data (all members of the _KEYS class attribute are required)
11997
    - four buffer attributes (in|out_data|text), that represent the
11998
      input (to the external script) in text and data structure format,
11999
      and the output from it, again in two formats
12000
    - the result variables from the script (success, info, nodes) for
12001
      easy usage
12002

12003
  """
12004
  # pylint: disable-msg=R0902
12005
  # lots of instance attributes
12006

    
12007
  def __init__(self, cfg, rpc, mode, **kwargs):
12008
    self.cfg = cfg
12009
    self.rpc = rpc
12010
    # init buffer variables
12011
    self.in_text = self.out_text = self.in_data = self.out_data = None
12012
    # init all input fields so that pylint is happy
12013
    self.mode = mode
12014
    self.mem_size = self.disks = self.disk_template = None
12015
    self.os = self.tags = self.nics = self.vcpus = None
12016
    self.hypervisor = None
12017
    self.relocate_from = None
12018
    self.name = None
12019
    self.evac_nodes = None
12020
    self.instances = None
12021
    self.reloc_mode = None
12022
    self.target_groups = None
12023
    # computed fields
12024
    self.required_nodes = None
12025
    # init result fields
12026
    self.success = self.info = self.result = None
12027

    
12028
    try:
12029
      (fn, keyset, self._result_check) = self._MODE_DATA[self.mode]
12030
    except KeyError:
12031
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
12032
                                   " IAllocator" % self.mode)
12033

    
12034
    for key in kwargs:
12035
      if key not in keyset:
12036
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
12037
                                     " IAllocator" % key)
12038
      setattr(self, key, kwargs[key])
12039

    
12040
    for key in keyset:
12041
      if key not in kwargs:
12042
        raise errors.ProgrammerError("Missing input parameter '%s' to"
12043
                                     " IAllocator" % key)
12044
    self._BuildInputData(compat.partial(fn, self))
12045

    
12046
  def _ComputeClusterData(self):
12047
    """Compute the generic allocator input data.
12048

12049
    This is the data that is independent of the actual operation.
12050

12051
    """
12052
    cfg = self.cfg
12053
    cluster_info = cfg.GetClusterInfo()
12054
    # cluster data
12055
    data = {
12056
      "version": constants.IALLOCATOR_VERSION,
12057
      "cluster_name": cfg.GetClusterName(),
12058
      "cluster_tags": list(cluster_info.GetTags()),
12059
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
12060
      # we don't have job IDs
12061
      }
12062
    ninfo = cfg.GetAllNodesInfo()
12063
    iinfo = cfg.GetAllInstancesInfo().values()
12064
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
12065

    
12066
    # node data
12067
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
12068

    
12069
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
12070
      hypervisor_name = self.hypervisor
12071
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
12072
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
12073
    elif self.mode in (constants.IALLOCATOR_MODE_MEVAC,
12074
                       constants.IALLOCATOR_MODE_MRELOC):
12075
      hypervisor_name = cluster_info.enabled_hypervisors[0]
12076

    
12077
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
12078
                                        hypervisor_name)
12079
    node_iinfo = \
12080
      self.rpc.call_all_instances_info(node_list,
12081
                                       cluster_info.enabled_hypervisors)
12082

    
12083
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
12084

    
12085
    config_ndata = self._ComputeBasicNodeData(ninfo)
12086
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
12087
                                                 i_list, config_ndata)
12088
    assert len(data["nodes"]) == len(ninfo), \
12089
        "Incomplete node data computed"
12090

    
12091
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
12092

    
12093
    self.in_data = data
12094

    
12095
  @staticmethod
12096
  def _ComputeNodeGroupData(cfg):
12097
    """Compute node groups data.
12098

12099
    """
12100
    ng = dict((guuid, {
12101
      "name": gdata.name,
12102
      "alloc_policy": gdata.alloc_policy,
12103
      })
12104
      for guuid, gdata in cfg.GetAllNodeGroupsInfo().items())
12105

    
12106
    return ng
12107

    
12108
  @staticmethod
12109
  def _ComputeBasicNodeData(node_cfg):
12110
    """Compute global node data.
12111

12112
    @rtype: dict
12113
    @returns: a dict of name: (node dict, node config)
12114

12115
    """
12116
    # fill in static (config-based) values
12117
    node_results = dict((ninfo.name, {
12118
      "tags": list(ninfo.GetTags()),
12119
      "primary_ip": ninfo.primary_ip,
12120
      "secondary_ip": ninfo.secondary_ip,
12121
      "offline": ninfo.offline,
12122
      "drained": ninfo.drained,
12123
      "master_candidate": ninfo.master_candidate,
12124
      "group": ninfo.group,
12125
      "master_capable": ninfo.master_capable,
12126
      "vm_capable": ninfo.vm_capable,
12127
      })
12128
      for ninfo in node_cfg.values())
12129

    
12130
    return node_results
12131

    
12132
  @staticmethod
12133
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
12134
                              node_results):
12135
    """Compute global node data.
12136

12137
    @param node_results: the basic node structures as filled from the config
12138

12139
    """
12140
    # make a copy of the current dict
12141
    node_results = dict(node_results)
12142
    for nname, nresult in node_data.items():
12143
      assert nname in node_results, "Missing basic data for node %s" % nname
12144
      ninfo = node_cfg[nname]
12145

    
12146
      if not (ninfo.offline or ninfo.drained):
12147
        nresult.Raise("Can't get data for node %s" % nname)
12148
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
12149
                                nname)
12150
        remote_info = nresult.payload
12151

    
12152
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
12153
                     'vg_size', 'vg_free', 'cpu_total']:
12154
          if attr not in remote_info:
12155
            raise errors.OpExecError("Node '%s' didn't return attribute"
12156
                                     " '%s'" % (nname, attr))
12157
          if not isinstance(remote_info[attr], int):
12158
            raise errors.OpExecError("Node '%s' returned invalid value"
12159
                                     " for '%s': %s" %
12160
                                     (nname, attr, remote_info[attr]))
12161
        # compute memory used by primary instances
12162
        i_p_mem = i_p_up_mem = 0
12163
        for iinfo, beinfo in i_list:
12164
          if iinfo.primary_node == nname:
12165
            i_p_mem += beinfo[constants.BE_MEMORY]
12166
            if iinfo.name not in node_iinfo[nname].payload:
12167
              i_used_mem = 0
12168
            else:
12169
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
12170
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
12171
            remote_info['memory_free'] -= max(0, i_mem_diff)
12172

    
12173
            if iinfo.admin_up:
12174
              i_p_up_mem += beinfo[constants.BE_MEMORY]
12175

    
12176
        # compute memory used by instances
12177
        pnr_dyn = {
12178
          "total_memory": remote_info['memory_total'],
12179
          "reserved_memory": remote_info['memory_dom0'],
12180
          "free_memory": remote_info['memory_free'],
12181
          "total_disk": remote_info['vg_size'],
12182
          "free_disk": remote_info['vg_free'],
12183
          "total_cpus": remote_info['cpu_total'],
12184
          "i_pri_memory": i_p_mem,
12185
          "i_pri_up_memory": i_p_up_mem,
12186
          }
12187
        pnr_dyn.update(node_results[nname])
12188
        node_results[nname] = pnr_dyn
12189

    
12190
    return node_results
12191

    
12192
  @staticmethod
12193
  def _ComputeInstanceData(cluster_info, i_list):
12194
    """Compute global instance data.
12195

12196
    """
12197
    instance_data = {}
12198
    for iinfo, beinfo in i_list:
12199
      nic_data = []
12200
      for nic in iinfo.nics:
12201
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
12202
        nic_dict = {
12203
          "mac": nic.mac,
12204
          "ip": nic.ip,
12205
          "mode": filled_params[constants.NIC_MODE],
12206
          "link": filled_params[constants.NIC_LINK],
12207
          }
12208
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
12209
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
12210
        nic_data.append(nic_dict)
12211
      pir = {
12212
        "tags": list(iinfo.GetTags()),
12213
        "admin_up": iinfo.admin_up,
12214
        "vcpus": beinfo[constants.BE_VCPUS],
12215
        "memory": beinfo[constants.BE_MEMORY],
12216
        "os": iinfo.os,
12217
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
12218
        "nics": nic_data,
12219
        "disks": [{constants.IDISK_SIZE: dsk.size,
12220
                   constants.IDISK_MODE: dsk.mode}
12221
                  for dsk in iinfo.disks],
12222
        "disk_template": iinfo.disk_template,
12223
        "hypervisor": iinfo.hypervisor,
12224
        }
12225
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
12226
                                                 pir["disks"])
12227
      instance_data[iinfo.name] = pir
12228

    
12229
    return instance_data
12230

    
12231
  def _AddNewInstance(self):
12232
    """Add new instance data to allocator structure.
12233

12234
    This in combination with _AllocatorGetClusterData will create the
12235
    correct structure needed as input for the allocator.
12236

12237
    The checks for the completeness of the opcode must have already been
12238
    done.
12239

12240
    """
12241
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
12242

    
12243
    if self.disk_template in constants.DTS_INT_MIRROR:
12244
      self.required_nodes = 2
12245
    else:
12246
      self.required_nodes = 1
12247

    
12248
    request = {
12249
      "name": self.name,
12250
      "disk_template": self.disk_template,
12251
      "tags": self.tags,
12252
      "os": self.os,
12253
      "vcpus": self.vcpus,
12254
      "memory": self.mem_size,
12255
      "disks": self.disks,
12256
      "disk_space_total": disk_space,
12257
      "nics": self.nics,
12258
      "required_nodes": self.required_nodes,
12259
      }
12260

    
12261
    return request
12262

    
12263
  def _AddRelocateInstance(self):
12264
    """Add relocate instance data to allocator structure.
12265

12266
    This in combination with _IAllocatorGetClusterData will create the
12267
    correct structure needed as input for the allocator.
12268

12269
    The checks for the completeness of the opcode must have already been
12270
    done.
12271

12272
    """
12273
    instance = self.cfg.GetInstanceInfo(self.name)
12274
    if instance is None:
12275
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
12276
                                   " IAllocator" % self.name)
12277

    
12278
    if instance.disk_template not in constants.DTS_MIRRORED:
12279
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
12280
                                 errors.ECODE_INVAL)
12281

    
12282
    if instance.disk_template in constants.DTS_INT_MIRROR and \
12283
        len(instance.secondary_nodes) != 1:
12284
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
12285
                                 errors.ECODE_STATE)
12286

    
12287
    self.required_nodes = 1
12288
    disk_sizes = [{constants.IDISK_SIZE: disk.size} for disk in instance.disks]
12289
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
12290

    
12291
    request = {
12292
      "name": self.name,
12293
      "disk_space_total": disk_space,
12294
      "required_nodes": self.required_nodes,
12295
      "relocate_from": self.relocate_from,
12296
      }
12297
    return request
12298

    
12299
  def _AddEvacuateNodes(self):
12300
    """Add evacuate nodes data to allocator structure.
12301

12302
    """
12303
    request = {
12304
      "evac_nodes": self.evac_nodes
12305
      }
12306
    return request
12307

    
12308
  def _AddMultiRelocate(self):
12309
    """Get data for multi-relocate requests.
12310

12311
    """
12312
    return {
12313
      "instances": self.instances,
12314
      "reloc_mode": self.reloc_mode,
12315
      "target_groups": self.target_groups,
12316
      }
12317

    
12318
  def _BuildInputData(self, fn):
12319
    """Build input data structures.
12320

12321
    """
12322
    self._ComputeClusterData()
12323

    
12324
    request = fn()
12325
    request["type"] = self.mode
12326
    self.in_data["request"] = request
12327

    
12328
    self.in_text = serializer.Dump(self.in_data)
12329

    
12330
  _MODE_DATA = {
12331
    constants.IALLOCATOR_MODE_ALLOC:
12332
      (_AddNewInstance,
12333
       ["name", "mem_size", "disks", "disk_template", "os", "tags", "nics",
12334
        "vcpus", "hypervisor"], ht.TList),
12335
    constants.IALLOCATOR_MODE_RELOC:
12336
      (_AddRelocateInstance, ["name", "relocate_from"], ht.TList),
12337
    constants.IALLOCATOR_MODE_MEVAC:
12338
      (_AddEvacuateNodes, ["evac_nodes"],
12339
       ht.TListOf(ht.TAnd(ht.TIsLength(2),
12340
                          ht.TListOf(ht.TString)))),
12341
    constants.IALLOCATOR_MODE_MRELOC:
12342
      (_AddMultiRelocate, ["instances", "reloc_mode", "target_groups"],
12343
       ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
12344
         # pylint: disable-msg=E1101
12345
         # Class '...' has no 'OP_ID' member
12346
         "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
12347
                              opcodes.OpInstanceMigrate.OP_ID,
12348
                              opcodes.OpInstanceReplaceDisks.OP_ID])
12349
         })))),
12350
    }
12351

    
12352
  def Run(self, name, validate=True, call_fn=None):
12353
    """Run an instance allocator and return the results.
12354

12355
    """
12356
    if call_fn is None:
12357
      call_fn = self.rpc.call_iallocator_runner
12358

    
12359
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
12360
    result.Raise("Failure while running the iallocator script")
12361

    
12362
    self.out_text = result.payload
12363
    if validate:
12364
      self._ValidateResult()
12365

    
12366
  def _ValidateResult(self):
12367
    """Process the allocator results.
12368

12369
    This will process and if successful save the result in
12370
    self.out_data and the other parameters.
12371

12372
    """
12373
    try:
12374
      rdict = serializer.Load(self.out_text)
12375
    except Exception, err:
12376
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
12377

    
12378
    if not isinstance(rdict, dict):
12379
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
12380

    
12381
    # TODO: remove backwards compatiblity in later versions
12382
    if "nodes" in rdict and "result" not in rdict:
12383
      rdict["result"] = rdict["nodes"]
12384
      del rdict["nodes"]
12385

    
12386
    for key in "success", "info", "result":
12387
      if key not in rdict:
12388
        raise errors.OpExecError("Can't parse iallocator results:"
12389
                                 " missing key '%s'" % key)
12390
      setattr(self, key, rdict[key])
12391

    
12392
    if not self._result_check(self.result):
12393
      raise errors.OpExecError("Iallocator returned invalid result,"
12394
                               " expected %s, got %s" %
12395
                               (self._result_check, self.result),
12396
                               errors.ECODE_INVAL)
12397

    
12398
    if self.mode in (constants.IALLOCATOR_MODE_RELOC,
12399
                     constants.IALLOCATOR_MODE_MEVAC):
12400
      node2group = dict((name, ndata["group"])
12401
                        for (name, ndata) in self.in_data["nodes"].items())
12402

    
12403
      fn = compat.partial(self._NodesToGroups, node2group,
12404
                          self.in_data["nodegroups"])
12405

    
12406
      if self.mode == constants.IALLOCATOR_MODE_RELOC:
12407
        assert self.relocate_from is not None
12408
        assert self.required_nodes == 1
12409

    
12410
        request_groups = fn(self.relocate_from)
12411
        result_groups = fn(rdict["result"])
12412

    
12413
        if result_groups != request_groups:
12414
          raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
12415
                                   " differ from original groups (%s)" %
12416
                                   (utils.CommaJoin(result_groups),
12417
                                    utils.CommaJoin(request_groups)))
12418
      elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
12419
        request_groups = fn(self.evac_nodes)
12420
        for (instance_name, secnode) in self.result:
12421
          result_groups = fn([secnode])
12422
          if result_groups != request_groups:
12423
            raise errors.OpExecError("Iallocator returned new secondary node"
12424
                                     " '%s' (group '%s') for instance '%s'"
12425
                                     " which is not in original group '%s'" %
12426
                                     (secnode, utils.CommaJoin(result_groups),
12427
                                      instance_name,
12428
                                      utils.CommaJoin(request_groups)))
12429
      else:
12430
        raise errors.ProgrammerError("Unhandled mode '%s'" % self.mode)
12431

    
12432
    self.out_data = rdict
12433

    
12434
  @staticmethod
12435
  def _NodesToGroups(node2group, groups, nodes):
12436
    """Returns a list of unique group names for a list of nodes.
12437

12438
    @type node2group: dict
12439
    @param node2group: Map from node name to group UUID
12440
    @type groups: dict
12441
    @param groups: Group information
12442
    @type nodes: list
12443
    @param nodes: Node names
12444

12445
    """
12446
    result = set()
12447

    
12448
    for node in nodes:
12449
      try:
12450
        group_uuid = node2group[node]
12451
      except KeyError:
12452
        # Ignore unknown node
12453
        pass
12454
      else:
12455
        try:
12456
          group = groups[group_uuid]
12457
        except KeyError:
12458
          # Can't find group, let's use UUID
12459
          group_name = group_uuid
12460
        else:
12461
          group_name = group["name"]
12462

    
12463
        result.add(group_name)
12464

    
12465
    return sorted(result)
12466

    
12467

    
12468
class LUTestAllocator(NoHooksLU):
12469
  """Run allocator tests.
12470

12471
  This LU runs the allocator tests
12472

12473
  """
12474
  def CheckPrereq(self):
12475
    """Check prerequisites.
12476

12477
    This checks the opcode parameters depending on the director and mode test.
12478

12479
    """
12480
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
12481
      for attr in ["mem_size", "disks", "disk_template",
12482
                   "os", "tags", "nics", "vcpus"]:
12483
        if not hasattr(self.op, attr):
12484
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
12485
                                     attr, errors.ECODE_INVAL)
12486
      iname = self.cfg.ExpandInstanceName(self.op.name)
12487
      if iname is not None:
12488
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
12489
                                   iname, errors.ECODE_EXISTS)
12490
      if not isinstance(self.op.nics, list):
12491
        raise errors.OpPrereqError("Invalid parameter 'nics'",
12492
                                   errors.ECODE_INVAL)
12493
      if not isinstance(self.op.disks, list):
12494
        raise errors.OpPrereqError("Invalid parameter 'disks'",
12495
                                   errors.ECODE_INVAL)
12496
      for row in self.op.disks:
12497
        if (not isinstance(row, dict) or
12498
            "size" not in row or
12499
            not isinstance(row["size"], int) or
12500
            "mode" not in row or
12501
            row["mode"] not in ['r', 'w']):
12502
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
12503
                                     " parameter", errors.ECODE_INVAL)
12504
      if self.op.hypervisor is None:
12505
        self.op.hypervisor = self.cfg.GetHypervisorType()
12506
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
12507
      fname = _ExpandInstanceName(self.cfg, self.op.name)
12508
      self.op.name = fname
12509
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
12510
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
12511
      if not hasattr(self.op, "evac_nodes"):
12512
        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
12513
                                   " opcode input", errors.ECODE_INVAL)
12514
    elif self.op.mode == constants.IALLOCATOR_MODE_MRELOC:
12515
      if self.op.instances:
12516
        self.op.instances = _GetWantedInstances(self, self.op.instances)
12517
      else:
12518
        raise errors.OpPrereqError("Missing instances to relocate",
12519
                                   errors.ECODE_INVAL)
12520
    else:
12521
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
12522
                                 self.op.mode, errors.ECODE_INVAL)
12523

    
12524
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
12525
      if self.op.allocator is None:
12526
        raise errors.OpPrereqError("Missing allocator name",
12527
                                   errors.ECODE_INVAL)
12528
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
12529
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
12530
                                 self.op.direction, errors.ECODE_INVAL)
12531

    
12532
  def Exec(self, feedback_fn):
12533
    """Run the allocator test.
12534

12535
    """
12536
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
12537
      ial = IAllocator(self.cfg, self.rpc,
12538
                       mode=self.op.mode,
12539
                       name=self.op.name,
12540
                       mem_size=self.op.mem_size,
12541
                       disks=self.op.disks,
12542
                       disk_template=self.op.disk_template,
12543
                       os=self.op.os,
12544
                       tags=self.op.tags,
12545
                       nics=self.op.nics,
12546
                       vcpus=self.op.vcpus,
12547
                       hypervisor=self.op.hypervisor,
12548
                       )
12549
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
12550
      ial = IAllocator(self.cfg, self.rpc,
12551
                       mode=self.op.mode,
12552
                       name=self.op.name,
12553
                       relocate_from=list(self.relocate_from),
12554
                       )
12555
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
12556
      ial = IAllocator(self.cfg, self.rpc,
12557
                       mode=self.op.mode,
12558
                       evac_nodes=self.op.evac_nodes)
12559
    elif self.op.mode == constants.IALLOCATOR_MODE_MRELOC:
12560
      ial = IAllocator(self.cfg, self.rpc,
12561
                       mode=self.op.mode,
12562
                       instances=self.op.instances,
12563
                       reloc_mode=self.op.reloc_mode,
12564
                       target_groups=self.op.target_groups)
12565
    else:
12566
      raise errors.ProgrammerError("Uncatched mode %s in"
12567
                                   " LUTestAllocator.Exec", self.op.mode)
12568

    
12569
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
12570
      result = ial.in_text
12571
    else:
12572
      ial.Run(self.op.allocator, validate=False)
12573
      result = ial.out_text
12574
    return result
12575

    
12576

    
12577
#: Query type implementations
12578
_QUERY_IMPL = {
12579
  constants.QR_INSTANCE: _InstanceQuery,
12580
  constants.QR_NODE: _NodeQuery,
12581
  constants.QR_GROUP: _GroupQuery,
12582
  constants.QR_OS: _OsQuery,
12583
  }
12584

    
12585
assert set(_QUERY_IMPL.keys()) == constants.QR_VIA_OP
12586

    
12587

    
12588
def _GetQueryImplementation(name):
12589
  """Returns the implemtnation for a query type.
12590

12591
  @param name: Query type, must be one of L{constants.QR_VIA_OP}
12592

12593
  """
12594
  try:
12595
    return _QUERY_IMPL[name]
12596
  except KeyError:
12597
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
12598
                               errors.ECODE_INVAL)