Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 9b64e486

History | View | Annotate | Download (405.7 kB)

1
#
2
#
3

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

    
21

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

    
24
# pylint: disable-msg=W0201,C0302
25

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

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

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

    
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

    
61
import ganeti.masterd.instance # pylint: disable-msg=W0611
62

    
63

    
64
def _SupportsOob(cfg, node):
65
  """Tells if node supports OOB.
66

67
  @type cfg: L{config.ConfigWriter}
68
  @param cfg: The cluster configuration
69
  @type node: L{objects.Node}
70
  @param node: The node
71
  @return: The OOB script if supported or an empty string otherwise
72

73
  """
74
  return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
75

    
76

    
77
# End types
78
class LogicalUnit(object):
79
  """Logical Unit base class.
80

81
  Subclasses must follow these rules:
82
    - implement ExpandNames
83
    - implement CheckPrereq (except when tasklets are used)
84
    - implement Exec (except when tasklets are used)
85
    - implement BuildHooksEnv
86
    - redefine HPATH and HTYPE
87
    - optionally redefine their run requirements:
88
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
89

90
  Note that all commands require root permissions.
91

92
  @ivar dry_run_result: the value (if any) that will be returned to the caller
93
      in dry-run mode (signalled by opcode dry_run parameter)
94

95
  """
96
  HPATH = None
97
  HTYPE = None
98
  REQ_BGL = True
99

    
100
  def __init__(self, processor, op, context, rpc):
101
    """Constructor for LogicalUnit.
102

103
    This needs to be overridden in derived classes in order to check op
104
    validity.
105

106
    """
107
    self.proc = processor
108
    self.op = op
109
    self.cfg = context.cfg
110
    self.context = context
111
    self.rpc = rpc
112
    # Dicts used to declare locking needs to mcpu
113
    self.needed_locks = None
114
    self.acquired_locks = {}
115
    self.share_locks = dict.fromkeys(locking.LEVELS, 0)
116
    self.add_locks = {}
117
    self.remove_locks = {}
118
    # Used to force good behavior when calling helper functions
119
    self.recalculate_locks = {}
120
    self.__ssh = None
121
    # logging
122
    self.Log = processor.Log # pylint: disable-msg=C0103
123
    self.LogWarning = processor.LogWarning # pylint: disable-msg=C0103
124
    self.LogInfo = processor.LogInfo # pylint: disable-msg=C0103
125
    self.LogStep = processor.LogStep # pylint: disable-msg=C0103
126
    # support for dry-run
127
    self.dry_run_result = None
128
    # support for generic debug attribute
129
    if (not hasattr(self.op, "debug_level") or
130
        not isinstance(self.op.debug_level, int)):
131
      self.op.debug_level = 0
132

    
133
    # Tasklets
134
    self.tasklets = None
135

    
136
    # Validate opcode parameters and set defaults
137
    self.op.Validate(True)
138

    
139
    self.CheckArguments()
140

    
141
  def __GetSSH(self):
142
    """Returns the SshRunner object
143

144
    """
145
    if not self.__ssh:
146
      self.__ssh = ssh.SshRunner(self.cfg.GetClusterName())
147
    return self.__ssh
148

    
149
  ssh = property(fget=__GetSSH)
150

    
151
  def CheckArguments(self):
152
    """Check syntactic validity for the opcode arguments.
153

154
    This method is for doing a simple syntactic check and ensure
155
    validity of opcode parameters, without any cluster-related
156
    checks. While the same can be accomplished in ExpandNames and/or
157
    CheckPrereq, doing these separate is better because:
158

159
      - ExpandNames is left as as purely a lock-related function
160
      - CheckPrereq is run after we have acquired locks (and possible
161
        waited for them)
162

163
    The function is allowed to change the self.op attribute so that
164
    later methods can no longer worry about missing parameters.
165

166
    """
167
    pass
168

    
169
  def ExpandNames(self):
170
    """Expand names for this LU.
171

172
    This method is called before starting to execute the opcode, and it should
173
    update all the parameters of the opcode to their canonical form (e.g. a
174
    short node name must be fully expanded after this method has successfully
175
    completed). This way locking, hooks, logging, etc. can work correctly.
176

177
    LUs which implement this method must also populate the self.needed_locks
178
    member, as a dict with lock levels as keys, and a list of needed lock names
179
    as values. Rules:
180

181
      - use an empty dict if you don't need any lock
182
      - if you don't need any lock at a particular level omit that level
183
      - don't put anything for the BGL level
184
      - if you want all locks at a level use locking.ALL_SET as a value
185

186
    If you need to share locks (rather than acquire them exclusively) at one
187
    level you can modify self.share_locks, setting a true value (usually 1) for
188
    that level. By default locks are not shared.
189

190
    This function can also define a list of tasklets, which then will be
191
    executed in order instead of the usual LU-level CheckPrereq and Exec
192
    functions, if those are not defined by the LU.
193

194
    Examples::
195

196
      # Acquire all nodes and one instance
197
      self.needed_locks = {
198
        locking.LEVEL_NODE: locking.ALL_SET,
199
        locking.LEVEL_INSTANCE: ['instance1.example.com'],
200
      }
201
      # Acquire just two nodes
202
      self.needed_locks = {
203
        locking.LEVEL_NODE: ['node1.example.com', 'node2.example.com'],
204
      }
205
      # Acquire no locks
206
      self.needed_locks = {} # No, you can't leave it to the default value None
207

208
    """
209
    # The implementation of this method is mandatory only if the new LU is
210
    # concurrent, so that old LUs don't need to be changed all at the same
211
    # time.
212
    if self.REQ_BGL:
213
      self.needed_locks = {} # Exclusive LUs don't need locks.
214
    else:
215
      raise NotImplementedError
216

    
217
  def DeclareLocks(self, level):
218
    """Declare LU locking needs for a level
219

220
    While most LUs can just declare their locking needs at ExpandNames time,
221
    sometimes there's the need to calculate some locks after having acquired
222
    the ones before. This function is called just before acquiring locks at a
223
    particular level, but after acquiring the ones at lower levels, and permits
224
    such calculations. It can be used to modify self.needed_locks, and by
225
    default it does nothing.
226

227
    This function is only called if you have something already set in
228
    self.needed_locks for the level.
229

230
    @param level: Locking level which is going to be locked
231
    @type level: member of ganeti.locking.LEVELS
232

233
    """
234

    
235
  def CheckPrereq(self):
236
    """Check prerequisites for this LU.
237

238
    This method should check that the prerequisites for the execution
239
    of this LU are fulfilled. It can do internode communication, but
240
    it should be idempotent - no cluster or system changes are
241
    allowed.
242

243
    The method should raise errors.OpPrereqError in case something is
244
    not fulfilled. Its return value is ignored.
245

246
    This method should also update all the parameters of the opcode to
247
    their canonical form if it hasn't been done by ExpandNames before.
248

249
    """
250
    if self.tasklets is not None:
251
      for (idx, tl) in enumerate(self.tasklets):
252
        logging.debug("Checking prerequisites for tasklet %s/%s",
253
                      idx + 1, len(self.tasklets))
254
        tl.CheckPrereq()
255
    else:
256
      pass
257

    
258
  def Exec(self, feedback_fn):
259
    """Execute the LU.
260

261
    This method should implement the actual work. It should raise
262
    errors.OpExecError for failures that are somewhat dealt with in
263
    code, or expected.
264

265
    """
266
    if self.tasklets is not None:
267
      for (idx, tl) in enumerate(self.tasklets):
268
        logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
269
        tl.Exec(feedback_fn)
270
    else:
271
      raise NotImplementedError
272

    
273
  def BuildHooksEnv(self):
274
    """Build hooks environment for this LU.
275

276
    This method should return a three-node tuple consisting of: a dict
277
    containing the environment that will be used for running the
278
    specific hook for this LU, a list of node names on which the hook
279
    should run before the execution, and a list of node names on which
280
    the hook should run after the execution.
281

282
    The keys of the dict must not have 'GANETI_' prefixed as this will
283
    be handled in the hooks runner. Also note additional keys will be
284
    added by the hooks runner. If the LU doesn't define any
285
    environment, an empty dict (and not None) should be returned.
286

287
    No nodes should be returned as an empty list (and not None).
288

289
    Note that if the HPATH for a LU class is None, this function will
290
    not be called.
291

292
    """
293
    raise NotImplementedError
294

    
295
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
296
    """Notify the LU about the results of its hooks.
297

298
    This method is called every time a hooks phase is executed, and notifies
299
    the Logical Unit about the hooks' result. The LU can then use it to alter
300
    its result based on the hooks.  By default the method does nothing and the
301
    previous result is passed back unchanged but any LU can define it if it
302
    wants to use the local cluster hook-scripts somehow.
303

304
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
305
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
306
    @param hook_results: the results of the multi-node hooks rpc call
307
    @param feedback_fn: function used send feedback back to the caller
308
    @param lu_result: the previous Exec result this LU had, or None
309
        in the PRE phase
310
    @return: the new Exec result, based on the previous result
311
        and hook results
312

313
    """
314
    # API must be kept, thus we ignore the unused argument and could
315
    # be a function warnings
316
    # pylint: disable-msg=W0613,R0201
317
    return lu_result
318

    
319
  def _ExpandAndLockInstance(self):
320
    """Helper function to expand and lock an instance.
321

322
    Many LUs that work on an instance take its name in self.op.instance_name
323
    and need to expand it and then declare the expanded name for locking. This
324
    function does it, and then updates self.op.instance_name to the expanded
325
    name. It also initializes needed_locks as a dict, if this hasn't been done
326
    before.
327

328
    """
329
    if self.needed_locks is None:
330
      self.needed_locks = {}
331
    else:
332
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
333
        "_ExpandAndLockInstance called with instance-level locks set"
334
    self.op.instance_name = _ExpandInstanceName(self.cfg,
335
                                                self.op.instance_name)
336
    self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
337

    
338
  def _LockInstancesNodes(self, primary_only=False):
339
    """Helper function to declare instances' nodes for locking.
340

341
    This function should be called after locking one or more instances to lock
342
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
343
    with all primary or secondary nodes for instances already locked and
344
    present in self.needed_locks[locking.LEVEL_INSTANCE].
345

346
    It should be called from DeclareLocks, and for safety only works if
347
    self.recalculate_locks[locking.LEVEL_NODE] is set.
348

349
    In the future it may grow parameters to just lock some instance's nodes, or
350
    to just lock primaries or secondary nodes, if needed.
351

352
    If should be called in DeclareLocks in a way similar to::
353

354
      if level == locking.LEVEL_NODE:
355
        self._LockInstancesNodes()
356

357
    @type primary_only: boolean
358
    @param primary_only: only lock primary nodes of locked instances
359

360
    """
361
    assert locking.LEVEL_NODE in self.recalculate_locks, \
362
      "_LockInstancesNodes helper function called with no nodes to recalculate"
363

    
364
    # TODO: check if we're really been called with the instance locks held
365

    
366
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
367
    # future we might want to have different behaviors depending on the value
368
    # of self.recalculate_locks[locking.LEVEL_NODE]
369
    wanted_nodes = []
370
    for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
371
      instance = self.context.cfg.GetInstanceInfo(instance_name)
372
      wanted_nodes.append(instance.primary_node)
373
      if not primary_only:
374
        wanted_nodes.extend(instance.secondary_nodes)
375

    
376
    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
377
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
378
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
379
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
380

    
381
    del self.recalculate_locks[locking.LEVEL_NODE]
382

    
383

    
384
class NoHooksLU(LogicalUnit): # pylint: disable-msg=W0223
385
  """Simple LU which runs no hooks.
386

387
  This LU is intended as a parent for other LogicalUnits which will
388
  run no hooks, in order to reduce duplicate code.
389

390
  """
391
  HPATH = None
392
  HTYPE = None
393

    
394
  def BuildHooksEnv(self):
395
    """Empty BuildHooksEnv for NoHooksLu.
396

397
    This just raises an error.
398

399
    """
400
    assert False, "BuildHooksEnv called for NoHooksLUs"
401

    
402

    
403
class Tasklet:
404
  """Tasklet base class.
405

406
  Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
407
  they can mix legacy code with tasklets. Locking needs to be done in the LU,
408
  tasklets know nothing about locks.
409

410
  Subclasses must follow these rules:
411
    - Implement CheckPrereq
412
    - Implement Exec
413

414
  """
415
  def __init__(self, lu):
416
    self.lu = lu
417

    
418
    # Shortcuts
419
    self.cfg = lu.cfg
420
    self.rpc = lu.rpc
421

    
422
  def CheckPrereq(self):
423
    """Check prerequisites for this tasklets.
424

425
    This method should check whether the prerequisites for the execution of
426
    this tasklet are fulfilled. It can do internode communication, but it
427
    should be idempotent - no cluster or system changes are allowed.
428

429
    The method should raise errors.OpPrereqError in case something is not
430
    fulfilled. Its return value is ignored.
431

432
    This method should also update all parameters to their canonical form if it
433
    hasn't been done before.
434

435
    """
436
    pass
437

    
438
  def Exec(self, feedback_fn):
439
    """Execute the tasklet.
440

441
    This method should implement the actual work. It should raise
442
    errors.OpExecError for failures that are somewhat dealt with in code, or
443
    expected.
444

445
    """
446
    raise NotImplementedError
447

    
448

    
449
class _QueryBase:
450
  """Base for query utility classes.
451

452
  """
453
  #: Attribute holding field definitions
454
  FIELDS = None
455

    
456
  def __init__(self, names, fields, use_locking):
457
    """Initializes this class.
458

459
    """
460
    self.names = names
461
    self.use_locking = use_locking
462

    
463
    self.query = query.Query(self.FIELDS, fields)
464
    self.requested_data = self.query.RequestedData()
465

    
466
    self.do_locking = None
467
    self.wanted = None
468

    
469
  def _GetNames(self, lu, all_names, lock_level):
470
    """Helper function to determine names asked for in the query.
471

472
    """
473
    if self.do_locking:
474
      names = lu.acquired_locks[lock_level]
475
    else:
476
      names = all_names
477

    
478
    if self.wanted == locking.ALL_SET:
479
      assert not self.names
480
      # caller didn't specify names, so ordering is not important
481
      return utils.NiceSort(names)
482

    
483
    # caller specified names and we must keep the same order
484
    assert self.names
485
    assert not self.do_locking or lu.acquired_locks[lock_level]
486

    
487
    missing = set(self.wanted).difference(names)
488
    if missing:
489
      raise errors.OpExecError("Some items were removed before retrieving"
490
                               " their data: %s" % missing)
491

    
492
    # Return expanded names
493
    return self.wanted
494

    
495
  @classmethod
496
  def FieldsQuery(cls, fields):
497
    """Returns list of available fields.
498

499
    @return: List of L{objects.QueryFieldDefinition}
500

501
    """
502
    return query.QueryFields(cls.FIELDS, fields)
503

    
504
  def ExpandNames(self, lu):
505
    """Expand names for this query.
506

507
    See L{LogicalUnit.ExpandNames}.
508

509
    """
510
    raise NotImplementedError()
511

    
512
  def DeclareLocks(self, lu, level):
513
    """Declare locks for this query.
514

515
    See L{LogicalUnit.DeclareLocks}.
516

517
    """
518
    raise NotImplementedError()
519

    
520
  def _GetQueryData(self, lu):
521
    """Collects all data for this query.
522

523
    @return: Query data object
524

525
    """
526
    raise NotImplementedError()
527

    
528
  def NewStyleQuery(self, lu):
529
    """Collect data and execute query.
530

531
    """
532
    return query.GetQueryResponse(self.query, self._GetQueryData(lu))
533

    
534
  def OldStyleQuery(self, lu):
535
    """Collect data and execute query.
536

537
    """
538
    return self.query.OldStyleQuery(self._GetQueryData(lu))
539

    
540

    
541
def _GetWantedNodes(lu, nodes):
542
  """Returns list of checked and expanded node names.
543

544
  @type lu: L{LogicalUnit}
545
  @param lu: the logical unit on whose behalf we execute
546
  @type nodes: list
547
  @param nodes: list of node names or None for all nodes
548
  @rtype: list
549
  @return: the list of nodes, sorted
550
  @raise errors.ProgrammerError: if the nodes parameter is wrong type
551

552
  """
553
  if nodes:
554
    return [_ExpandNodeName(lu.cfg, name) for name in nodes]
555

    
556
  return utils.NiceSort(lu.cfg.GetNodeList())
557

    
558

    
559
def _GetWantedInstances(lu, instances):
560
  """Returns list of checked and expanded instance names.
561

562
  @type lu: L{LogicalUnit}
563
  @param lu: the logical unit on whose behalf we execute
564
  @type instances: list
565
  @param instances: list of instance names or None for all instances
566
  @rtype: list
567
  @return: the list of instances, sorted
568
  @raise errors.OpPrereqError: if the instances parameter is wrong type
569
  @raise errors.OpPrereqError: if any of the passed instances is not found
570

571
  """
572
  if instances:
573
    wanted = [_ExpandInstanceName(lu.cfg, name) for name in instances]
574
  else:
575
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
576
  return wanted
577

    
578

    
579
def _GetUpdatedParams(old_params, update_dict,
580
                      use_default=True, use_none=False):
581
  """Return the new version of a parameter dictionary.
582

583
  @type old_params: dict
584
  @param old_params: old parameters
585
  @type update_dict: dict
586
  @param update_dict: dict containing new parameter values, or
587
      constants.VALUE_DEFAULT to reset the parameter to its default
588
      value
589
  @param use_default: boolean
590
  @type use_default: whether to recognise L{constants.VALUE_DEFAULT}
591
      values as 'to be deleted' values
592
  @param use_none: boolean
593
  @type use_none: whether to recognise C{None} values as 'to be
594
      deleted' values
595
  @rtype: dict
596
  @return: the new parameter dictionary
597

598
  """
599
  params_copy = copy.deepcopy(old_params)
600
  for key, val in update_dict.iteritems():
601
    if ((use_default and val == constants.VALUE_DEFAULT) or
602
        (use_none and val is None)):
603
      try:
604
        del params_copy[key]
605
      except KeyError:
606
        pass
607
    else:
608
      params_copy[key] = val
609
  return params_copy
610

    
611

    
612
def _CheckOutputFields(static, dynamic, selected):
613
  """Checks whether all selected fields are valid.
614

615
  @type static: L{utils.FieldSet}
616
  @param static: static fields set
617
  @type dynamic: L{utils.FieldSet}
618
  @param dynamic: dynamic fields set
619

620
  """
621
  f = utils.FieldSet()
622
  f.Extend(static)
623
  f.Extend(dynamic)
624

    
625
  delta = f.NonMatching(selected)
626
  if delta:
627
    raise errors.OpPrereqError("Unknown output fields selected: %s"
628
                               % ",".join(delta), errors.ECODE_INVAL)
629

    
630

    
631
def _CheckGlobalHvParams(params):
632
  """Validates that given hypervisor params are not global ones.
633

634
  This will ensure that instances don't get customised versions of
635
  global params.
636

637
  """
638
  used_globals = constants.HVC_GLOBALS.intersection(params)
639
  if used_globals:
640
    msg = ("The following hypervisor parameters are global and cannot"
641
           " be customized at instance level, please modify them at"
642
           " cluster level: %s" % utils.CommaJoin(used_globals))
643
    raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
644

    
645

    
646
def _CheckNodeOnline(lu, node, msg=None):
647
  """Ensure that a given node is online.
648

649
  @param lu: the LU on behalf of which we make the check
650
  @param node: the node to check
651
  @param msg: if passed, should be a message to replace the default one
652
  @raise errors.OpPrereqError: if the node is offline
653

654
  """
655
  if msg is None:
656
    msg = "Can't use offline node"
657
  if lu.cfg.GetNodeInfo(node).offline:
658
    raise errors.OpPrereqError("%s: %s" % (msg, node), errors.ECODE_STATE)
659

    
660

    
661
def _CheckNodeNotDrained(lu, node):
662
  """Ensure that a given node is not drained.
663

664
  @param lu: the LU on behalf of which we make the check
665
  @param node: the node to check
666
  @raise errors.OpPrereqError: if the node is drained
667

668
  """
669
  if lu.cfg.GetNodeInfo(node).drained:
670
    raise errors.OpPrereqError("Can't use drained node %s" % node,
671
                               errors.ECODE_STATE)
672

    
673

    
674
def _CheckNodeVmCapable(lu, node):
675
  """Ensure that a given node is vm capable.
676

677
  @param lu: the LU on behalf of which we make the check
678
  @param node: the node to check
679
  @raise errors.OpPrereqError: if the node is not vm capable
680

681
  """
682
  if not lu.cfg.GetNodeInfo(node).vm_capable:
683
    raise errors.OpPrereqError("Can't use non-vm_capable node %s" % node,
684
                               errors.ECODE_STATE)
685

    
686

    
687
def _CheckNodeHasOS(lu, node, os_name, force_variant):
688
  """Ensure that a node supports a given OS.
689

690
  @param lu: the LU on behalf of which we make the check
691
  @param node: the node to check
692
  @param os_name: the OS to query about
693
  @param force_variant: whether to ignore variant errors
694
  @raise errors.OpPrereqError: if the node is not supporting the OS
695

696
  """
697
  result = lu.rpc.call_os_get(node, os_name)
698
  result.Raise("OS '%s' not in supported OS list for node %s" %
699
               (os_name, node),
700
               prereq=True, ecode=errors.ECODE_INVAL)
701
  if not force_variant:
702
    _CheckOSVariant(result.payload, os_name)
703

    
704

    
705
def _CheckNodeHasSecondaryIP(lu, node, secondary_ip, prereq):
706
  """Ensure that a node has the given secondary ip.
707

708
  @type lu: L{LogicalUnit}
709
  @param lu: the LU on behalf of which we make the check
710
  @type node: string
711
  @param node: the node to check
712
  @type secondary_ip: string
713
  @param secondary_ip: the ip to check
714
  @type prereq: boolean
715
  @param prereq: whether to throw a prerequisite or an execute error
716
  @raise errors.OpPrereqError: if the node doesn't have the ip, and prereq=True
717
  @raise errors.OpExecError: if the node doesn't have the ip, and prereq=False
718

719
  """
720
  result = lu.rpc.call_node_has_ip_address(node, secondary_ip)
721
  result.Raise("Failure checking secondary ip on node %s" % node,
722
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
723
  if not result.payload:
724
    msg = ("Node claims it doesn't have the secondary ip you gave (%s),"
725
           " please fix and re-run this command" % secondary_ip)
726
    if prereq:
727
      raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
728
    else:
729
      raise errors.OpExecError(msg)
730

    
731

    
732
def _GetClusterDomainSecret():
733
  """Reads the cluster domain secret.
734

735
  """
736
  return utils.ReadOneLineFile(constants.CLUSTER_DOMAIN_SECRET_FILE,
737
                               strict=True)
738

    
739

    
740
def _CheckInstanceDown(lu, instance, reason):
741
  """Ensure that an instance is not running."""
742
  if instance.admin_up:
743
    raise errors.OpPrereqError("Instance %s is marked to be up, %s" %
744
                               (instance.name, reason), errors.ECODE_STATE)
745

    
746
  pnode = instance.primary_node
747
  ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
748
  ins_l.Raise("Can't contact node %s for instance information" % pnode,
749
              prereq=True, ecode=errors.ECODE_ENVIRON)
750

    
751
  if instance.name in ins_l.payload:
752
    raise errors.OpPrereqError("Instance %s is running, %s" %
753
                               (instance.name, reason), errors.ECODE_STATE)
754

    
755

    
756
def _ExpandItemName(fn, name, kind):
757
  """Expand an item name.
758

759
  @param fn: the function to use for expansion
760
  @param name: requested item name
761
  @param kind: text description ('Node' or 'Instance')
762
  @return: the resolved (full) name
763
  @raise errors.OpPrereqError: if the item is not found
764

765
  """
766
  full_name = fn(name)
767
  if full_name is None:
768
    raise errors.OpPrereqError("%s '%s' not known" % (kind, name),
769
                               errors.ECODE_NOENT)
770
  return full_name
771

    
772

    
773
def _ExpandNodeName(cfg, name):
774
  """Wrapper over L{_ExpandItemName} for nodes."""
775
  return _ExpandItemName(cfg.ExpandNodeName, name, "Node")
776

    
777

    
778
def _ExpandInstanceName(cfg, name):
779
  """Wrapper over L{_ExpandItemName} for instance."""
780
  return _ExpandItemName(cfg.ExpandInstanceName, name, "Instance")
781

    
782

    
783
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
784
                          memory, vcpus, nics, disk_template, disks,
785
                          bep, hvp, hypervisor_name):
786
  """Builds instance related env variables for hooks
787

788
  This builds the hook environment from individual variables.
789

790
  @type name: string
791
  @param name: the name of the instance
792
  @type primary_node: string
793
  @param primary_node: the name of the instance's primary node
794
  @type secondary_nodes: list
795
  @param secondary_nodes: list of secondary nodes as strings
796
  @type os_type: string
797
  @param os_type: the name of the instance's OS
798
  @type status: boolean
799
  @param status: the should_run status of the instance
800
  @type memory: string
801
  @param memory: the memory size of the instance
802
  @type vcpus: string
803
  @param vcpus: the count of VCPUs the instance has
804
  @type nics: list
805
  @param nics: list of tuples (ip, mac, mode, link) representing
806
      the NICs the instance has
807
  @type disk_template: string
808
  @param disk_template: the disk template of the instance
809
  @type disks: list
810
  @param disks: the list of (size, mode) pairs
811
  @type bep: dict
812
  @param bep: the backend parameters for the instance
813
  @type hvp: dict
814
  @param hvp: the hypervisor parameters for the instance
815
  @type hypervisor_name: string
816
  @param hypervisor_name: the hypervisor for the instance
817
  @rtype: dict
818
  @return: the hook environment for this instance
819

820
  """
821
  if status:
822
    str_status = "up"
823
  else:
824
    str_status = "down"
825
  env = {
826
    "OP_TARGET": name,
827
    "INSTANCE_NAME": name,
828
    "INSTANCE_PRIMARY": primary_node,
829
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
830
    "INSTANCE_OS_TYPE": os_type,
831
    "INSTANCE_STATUS": str_status,
832
    "INSTANCE_MEMORY": memory,
833
    "INSTANCE_VCPUS": vcpus,
834
    "INSTANCE_DISK_TEMPLATE": disk_template,
835
    "INSTANCE_HYPERVISOR": hypervisor_name,
836
  }
837

    
838
  if nics:
839
    nic_count = len(nics)
840
    for idx, (ip, mac, mode, link) in enumerate(nics):
841
      if ip is None:
842
        ip = ""
843
      env["INSTANCE_NIC%d_IP" % idx] = ip
844
      env["INSTANCE_NIC%d_MAC" % idx] = mac
845
      env["INSTANCE_NIC%d_MODE" % idx] = mode
846
      env["INSTANCE_NIC%d_LINK" % idx] = link
847
      if mode == constants.NIC_MODE_BRIDGED:
848
        env["INSTANCE_NIC%d_BRIDGE" % idx] = link
849
  else:
850
    nic_count = 0
851

    
852
  env["INSTANCE_NIC_COUNT"] = nic_count
853

    
854
  if disks:
855
    disk_count = len(disks)
856
    for idx, (size, mode) in enumerate(disks):
857
      env["INSTANCE_DISK%d_SIZE" % idx] = size
858
      env["INSTANCE_DISK%d_MODE" % idx] = mode
859
  else:
860
    disk_count = 0
861

    
862
  env["INSTANCE_DISK_COUNT"] = disk_count
863

    
864
  for source, kind in [(bep, "BE"), (hvp, "HV")]:
865
    for key, value in source.items():
866
      env["INSTANCE_%s_%s" % (kind, key)] = value
867

    
868
  return env
869

    
870

    
871
def _NICListToTuple(lu, nics):
872
  """Build a list of nic information tuples.
873

874
  This list is suitable to be passed to _BuildInstanceHookEnv or as a return
875
  value in LUInstanceQueryData.
876

877
  @type lu:  L{LogicalUnit}
878
  @param lu: the logical unit on whose behalf we execute
879
  @type nics: list of L{objects.NIC}
880
  @param nics: list of nics to convert to hooks tuples
881

882
  """
883
  hooks_nics = []
884
  cluster = lu.cfg.GetClusterInfo()
885
  for nic in nics:
886
    ip = nic.ip
887
    mac = nic.mac
888
    filled_params = cluster.SimpleFillNIC(nic.nicparams)
889
    mode = filled_params[constants.NIC_MODE]
890
    link = filled_params[constants.NIC_LINK]
891
    hooks_nics.append((ip, mac, mode, link))
892
  return hooks_nics
893

    
894

    
895
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
896
  """Builds instance related env variables for hooks from an object.
897

898
  @type lu: L{LogicalUnit}
899
  @param lu: the logical unit on whose behalf we execute
900
  @type instance: L{objects.Instance}
901
  @param instance: the instance for which we should build the
902
      environment
903
  @type override: dict
904
  @param override: dictionary with key/values that will override
905
      our values
906
  @rtype: dict
907
  @return: the hook environment dictionary
908

909
  """
910
  cluster = lu.cfg.GetClusterInfo()
911
  bep = cluster.FillBE(instance)
912
  hvp = cluster.FillHV(instance)
913
  args = {
914
    'name': instance.name,
915
    'primary_node': instance.primary_node,
916
    'secondary_nodes': instance.secondary_nodes,
917
    'os_type': instance.os,
918
    'status': instance.admin_up,
919
    'memory': bep[constants.BE_MEMORY],
920
    'vcpus': bep[constants.BE_VCPUS],
921
    'nics': _NICListToTuple(lu, instance.nics),
922
    'disk_template': instance.disk_template,
923
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
924
    'bep': bep,
925
    'hvp': hvp,
926
    'hypervisor_name': instance.hypervisor,
927
  }
928
  if override:
929
    args.update(override)
930
  return _BuildInstanceHookEnv(**args) # pylint: disable-msg=W0142
931

    
932

    
933
def _AdjustCandidatePool(lu, exceptions):
934
  """Adjust the candidate pool after node operations.
935

936
  """
937
  mod_list = lu.cfg.MaintainCandidatePool(exceptions)
938
  if mod_list:
939
    lu.LogInfo("Promoted nodes to master candidate role: %s",
940
               utils.CommaJoin(node.name for node in mod_list))
941
    for name in mod_list:
942
      lu.context.ReaddNode(name)
943
  mc_now, mc_max, _ = lu.cfg.GetMasterCandidateStats(exceptions)
944
  if mc_now > mc_max:
945
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
946
               (mc_now, mc_max))
947

    
948

    
949
def _DecideSelfPromotion(lu, exceptions=None):
950
  """Decide whether I should promote myself as a master candidate.
951

952
  """
953
  cp_size = lu.cfg.GetClusterInfo().candidate_pool_size
954
  mc_now, mc_should, _ = lu.cfg.GetMasterCandidateStats(exceptions)
955
  # the new node will increase mc_max with one, so:
956
  mc_should = min(mc_should + 1, cp_size)
957
  return mc_now < mc_should
958

    
959

    
960
def _CheckNicsBridgesExist(lu, target_nics, target_node):
961
  """Check that the brigdes needed by a list of nics exist.
962

963
  """
964
  cluster = lu.cfg.GetClusterInfo()
965
  paramslist = [cluster.SimpleFillNIC(nic.nicparams) for nic in target_nics]
966
  brlist = [params[constants.NIC_LINK] for params in paramslist
967
            if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
968
  if brlist:
969
    result = lu.rpc.call_bridges_exist(target_node, brlist)
970
    result.Raise("Error checking bridges on destination node '%s'" %
971
                 target_node, prereq=True, ecode=errors.ECODE_ENVIRON)
972

    
973

    
974
def _CheckInstanceBridgesExist(lu, instance, node=None):
975
  """Check that the brigdes needed by an instance exist.
976

977
  """
978
  if node is None:
979
    node = instance.primary_node
980
  _CheckNicsBridgesExist(lu, instance.nics, node)
981

    
982

    
983
def _CheckOSVariant(os_obj, name):
984
  """Check whether an OS name conforms to the os variants specification.
985

986
  @type os_obj: L{objects.OS}
987
  @param os_obj: OS object to check
988
  @type name: string
989
  @param name: OS name passed by the user, to check for validity
990

991
  """
992
  if not os_obj.supported_variants:
993
    return
994
  variant = objects.OS.GetVariant(name)
995
  if not variant:
996
    raise errors.OpPrereqError("OS name must include a variant",
997
                               errors.ECODE_INVAL)
998

    
999
  if variant not in os_obj.supported_variants:
1000
    raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)
1001

    
1002

    
1003
def _GetNodeInstancesInner(cfg, fn):
1004
  return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]
1005

    
1006

    
1007
def _GetNodeInstances(cfg, node_name):
1008
  """Returns a list of all primary and secondary instances on a node.
1009

1010
  """
1011

    
1012
  return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)
1013

    
1014

    
1015
def _GetNodePrimaryInstances(cfg, node_name):
1016
  """Returns primary instances on a node.
1017

1018
  """
1019
  return _GetNodeInstancesInner(cfg,
1020
                                lambda inst: node_name == inst.primary_node)
1021

    
1022

    
1023
def _GetNodeSecondaryInstances(cfg, node_name):
1024
  """Returns secondary instances on a node.
1025

1026
  """
1027
  return _GetNodeInstancesInner(cfg,
1028
                                lambda inst: node_name in inst.secondary_nodes)
1029

    
1030

    
1031
def _GetStorageTypeArgs(cfg, storage_type):
1032
  """Returns the arguments for a storage type.
1033

1034
  """
1035
  # Special case for file storage
1036
  if storage_type == constants.ST_FILE:
1037
    # storage.FileStorage wants a list of storage directories
1038
    return [[cfg.GetFileStorageDir()]]
1039

    
1040
  return []
1041

    
1042

    
1043
def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
1044
  faulty = []
1045

    
1046
  for dev in instance.disks:
1047
    cfg.SetDiskID(dev, node_name)
1048

    
1049
  result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
1050
  result.Raise("Failed to get disk status from node %s" % node_name,
1051
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
1052

    
1053
  for idx, bdev_status in enumerate(result.payload):
1054
    if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
1055
      faulty.append(idx)
1056

    
1057
  return faulty
1058

    
1059

    
1060
def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
1061
  """Check the sanity of iallocator and node arguments and use the
1062
  cluster-wide iallocator if appropriate.
1063

1064
  Check that at most one of (iallocator, node) is specified. If none is
1065
  specified, then the LU's opcode's iallocator slot is filled with the
1066
  cluster-wide default iallocator.
1067

1068
  @type iallocator_slot: string
1069
  @param iallocator_slot: the name of the opcode iallocator slot
1070
  @type node_slot: string
1071
  @param node_slot: the name of the opcode target node slot
1072

1073
  """
1074
  node = getattr(lu.op, node_slot, None)
1075
  iallocator = getattr(lu.op, iallocator_slot, None)
1076

    
1077
  if node is not None and iallocator is not None:
1078
    raise errors.OpPrereqError("Do not specify both, iallocator and node.",
1079
                               errors.ECODE_INVAL)
1080
  elif node is None and iallocator is None:
1081
    default_iallocator = lu.cfg.GetDefaultIAllocator()
1082
    if default_iallocator:
1083
      setattr(lu.op, iallocator_slot, default_iallocator)
1084
    else:
1085
      raise errors.OpPrereqError("No iallocator or node given and no"
1086
                                 " cluster-wide default iallocator found."
1087
                                 " Please specify either an iallocator or a"
1088
                                 " node, or set a cluster-wide default"
1089
                                 " iallocator.")
1090

    
1091

    
1092
class LUClusterPostInit(LogicalUnit):
1093
  """Logical unit for running hooks after cluster initialization.
1094

1095
  """
1096
  HPATH = "cluster-init"
1097
  HTYPE = constants.HTYPE_CLUSTER
1098

    
1099
  def BuildHooksEnv(self):
1100
    """Build hooks env.
1101

1102
    """
1103
    env = {"OP_TARGET": self.cfg.GetClusterName()}
1104
    mn = self.cfg.GetMasterNode()
1105
    return env, [], [mn]
1106

    
1107
  def Exec(self, feedback_fn):
1108
    """Nothing to do.
1109

1110
    """
1111
    return True
1112

    
1113

    
1114
class LUClusterDestroy(LogicalUnit):
1115
  """Logical unit for destroying the cluster.
1116

1117
  """
1118
  HPATH = "cluster-destroy"
1119
  HTYPE = constants.HTYPE_CLUSTER
1120

    
1121
  def BuildHooksEnv(self):
1122
    """Build hooks env.
1123

1124
    """
1125
    env = {"OP_TARGET": self.cfg.GetClusterName()}
1126
    return env, [], []
1127

    
1128
  def CheckPrereq(self):
1129
    """Check prerequisites.
1130

1131
    This checks whether the cluster is empty.
1132

1133
    Any errors are signaled by raising errors.OpPrereqError.
1134

1135
    """
1136
    master = self.cfg.GetMasterNode()
1137

    
1138
    nodelist = self.cfg.GetNodeList()
1139
    if len(nodelist) != 1 or nodelist[0] != master:
1140
      raise errors.OpPrereqError("There are still %d node(s) in"
1141
                                 " this cluster." % (len(nodelist) - 1),
1142
                                 errors.ECODE_INVAL)
1143
    instancelist = self.cfg.GetInstanceList()
1144
    if instancelist:
1145
      raise errors.OpPrereqError("There are still %d instance(s) in"
1146
                                 " this cluster." % len(instancelist),
1147
                                 errors.ECODE_INVAL)
1148

    
1149
  def Exec(self, feedback_fn):
1150
    """Destroys the cluster.
1151

1152
    """
1153
    master = self.cfg.GetMasterNode()
1154

    
1155
    # Run post hooks on master node before it's removed
1156
    hm = self.proc.hmclass(self.rpc.call_hooks_runner, self)
1157
    try:
1158
      hm.RunPhase(constants.HOOKS_PHASE_POST, [master])
1159
    except:
1160
      # pylint: disable-msg=W0702
1161
      self.LogWarning("Errors occurred running hooks on %s" % master)
1162

    
1163
    result = self.rpc.call_node_stop_master(master, False)
1164
    result.Raise("Could not disable the master role")
1165

    
1166
    return master
1167

    
1168

    
1169
def _VerifyCertificate(filename):
1170
  """Verifies a certificate for LUClusterVerify.
1171

1172
  @type filename: string
1173
  @param filename: Path to PEM file
1174

1175
  """
1176
  try:
1177
    cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1178
                                           utils.ReadFile(filename))
1179
  except Exception, err: # pylint: disable-msg=W0703
1180
    return (LUClusterVerify.ETYPE_ERROR,
1181
            "Failed to load X509 certificate %s: %s" % (filename, err))
1182

    
1183
  (errcode, msg) = \
1184
    utils.VerifyX509Certificate(cert, constants.SSL_CERT_EXPIRATION_WARN,
1185
                                constants.SSL_CERT_EXPIRATION_ERROR)
1186

    
1187
  if msg:
1188
    fnamemsg = "While verifying %s: %s" % (filename, msg)
1189
  else:
1190
    fnamemsg = None
1191

    
1192
  if errcode is None:
1193
    return (None, fnamemsg)
1194
  elif errcode == utils.CERT_WARNING:
1195
    return (LUClusterVerify.ETYPE_WARNING, fnamemsg)
1196
  elif errcode == utils.CERT_ERROR:
1197
    return (LUClusterVerify.ETYPE_ERROR, fnamemsg)
1198

    
1199
  raise errors.ProgrammerError("Unhandled certificate error code %r" % errcode)
1200

    
1201

    
1202
class LUClusterVerify(LogicalUnit):
1203
  """Verifies the cluster status.
1204

1205
  """
1206
  HPATH = "cluster-verify"
1207
  HTYPE = constants.HTYPE_CLUSTER
1208
  REQ_BGL = False
1209

    
1210
  TCLUSTER = "cluster"
1211
  TNODE = "node"
1212
  TINSTANCE = "instance"
1213

    
1214
  ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
1215
  ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
1216
  EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
1217
  EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
1218
  EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
1219
  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
1220
  EINSTANCEFAULTYDISK = (TINSTANCE, "EINSTANCEFAULTYDISK")
1221
  EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
1222
  EINSTANCESPLITGROUPS = (TINSTANCE, "EINSTANCESPLITGROUPS")
1223
  ENODEDRBD = (TNODE, "ENODEDRBD")
1224
  ENODEDRBDHELPER = (TNODE, "ENODEDRBDHELPER")
1225
  ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
1226
  ENODEHOOKS = (TNODE, "ENODEHOOKS")
1227
  ENODEHV = (TNODE, "ENODEHV")
1228
  ENODELVM = (TNODE, "ENODELVM")
1229
  ENODEN1 = (TNODE, "ENODEN1")
1230
  ENODENET = (TNODE, "ENODENET")
1231
  ENODEOS = (TNODE, "ENODEOS")
1232
  ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
1233
  ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
1234
  ENODERPC = (TNODE, "ENODERPC")
1235
  ENODESSH = (TNODE, "ENODESSH")
1236
  ENODEVERSION = (TNODE, "ENODEVERSION")
1237
  ENODESETUP = (TNODE, "ENODESETUP")
1238
  ENODETIME = (TNODE, "ENODETIME")
1239
  ENODEOOBPATH = (TNODE, "ENODEOOBPATH")
1240

    
1241
  ETYPE_FIELD = "code"
1242
  ETYPE_ERROR = "ERROR"
1243
  ETYPE_WARNING = "WARNING"
1244

    
1245
  _HOOKS_INDENT_RE = re.compile("^", re.M)
1246

    
1247
  class NodeImage(object):
1248
    """A class representing the logical and physical status of a node.
1249

1250
    @type name: string
1251
    @ivar name: the node name to which this object refers
1252
    @ivar volumes: a structure as returned from
1253
        L{ganeti.backend.GetVolumeList} (runtime)
1254
    @ivar instances: a list of running instances (runtime)
1255
    @ivar pinst: list of configured primary instances (config)
1256
    @ivar sinst: list of configured secondary instances (config)
1257
    @ivar sbp: diction of {secondary-node: list of instances} of all peers
1258
        of this node (config)
1259
    @ivar mfree: free memory, as reported by hypervisor (runtime)
1260
    @ivar dfree: free disk, as reported by the node (runtime)
1261
    @ivar offline: the offline status (config)
1262
    @type rpc_fail: boolean
1263
    @ivar rpc_fail: whether the RPC verify call was successfull (overall,
1264
        not whether the individual keys were correct) (runtime)
1265
    @type lvm_fail: boolean
1266
    @ivar lvm_fail: whether the RPC call didn't return valid LVM data
1267
    @type hyp_fail: boolean
1268
    @ivar hyp_fail: whether the RPC call didn't return the instance list
1269
    @type ghost: boolean
1270
    @ivar ghost: whether this is a known node or not (config)
1271
    @type os_fail: boolean
1272
    @ivar os_fail: whether the RPC call didn't return valid OS data
1273
    @type oslist: list
1274
    @ivar oslist: list of OSes as diagnosed by DiagnoseOS
1275
    @type vm_capable: boolean
1276
    @ivar vm_capable: whether the node can host instances
1277

1278
    """
1279
    def __init__(self, offline=False, name=None, vm_capable=True):
1280
      self.name = name
1281
      self.volumes = {}
1282
      self.instances = []
1283
      self.pinst = []
1284
      self.sinst = []
1285
      self.sbp = {}
1286
      self.mfree = 0
1287
      self.dfree = 0
1288
      self.offline = offline
1289
      self.vm_capable = vm_capable
1290
      self.rpc_fail = False
1291
      self.lvm_fail = False
1292
      self.hyp_fail = False
1293
      self.ghost = False
1294
      self.os_fail = False
1295
      self.oslist = {}
1296

    
1297
  def ExpandNames(self):
1298
    self.needed_locks = {
1299
      locking.LEVEL_NODE: locking.ALL_SET,
1300
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1301
    }
1302
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
1303

    
1304
  def _Error(self, ecode, item, msg, *args, **kwargs):
1305
    """Format an error message.
1306

1307
    Based on the opcode's error_codes parameter, either format a
1308
    parseable error code, or a simpler error string.
1309

1310
    This must be called only from Exec and functions called from Exec.
1311

1312
    """
1313
    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
1314
    itype, etxt = ecode
1315
    # first complete the msg
1316
    if args:
1317
      msg = msg % args
1318
    # then format the whole message
1319
    if self.op.error_codes:
1320
      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
1321
    else:
1322
      if item:
1323
        item = " " + item
1324
      else:
1325
        item = ""
1326
      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
1327
    # and finally report it via the feedback_fn
1328
    self._feedback_fn("  - %s" % msg)
1329

    
1330
  def _ErrorIf(self, cond, *args, **kwargs):
1331
    """Log an error message if the passed condition is True.
1332

1333
    """
1334
    cond = bool(cond) or self.op.debug_simulate_errors
1335
    if cond:
1336
      self._Error(*args, **kwargs)
1337
    # do not mark the operation as failed for WARN cases only
1338
    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
1339
      self.bad = self.bad or cond
1340

    
1341
  def _VerifyNode(self, ninfo, nresult):
1342
    """Perform some basic validation on data returned from a node.
1343

1344
      - check the result data structure is well formed and has all the
1345
        mandatory fields
1346
      - check ganeti version
1347

1348
    @type ninfo: L{objects.Node}
1349
    @param ninfo: the node to check
1350
    @param nresult: the results from the node
1351
    @rtype: boolean
1352
    @return: whether overall this call was successful (and we can expect
1353
         reasonable values in the respose)
1354

1355
    """
1356
    node = ninfo.name
1357
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1358

    
1359
    # main result, nresult should be a non-empty dict
1360
    test = not nresult or not isinstance(nresult, dict)
1361
    _ErrorIf(test, self.ENODERPC, node,
1362
                  "unable to verify node: no data returned")
1363
    if test:
1364
      return False
1365

    
1366
    # compares ganeti version
1367
    local_version = constants.PROTOCOL_VERSION
1368
    remote_version = nresult.get("version", None)
1369
    test = not (remote_version and
1370
                isinstance(remote_version, (list, tuple)) and
1371
                len(remote_version) == 2)
1372
    _ErrorIf(test, self.ENODERPC, node,
1373
             "connection to node returned invalid data")
1374
    if test:
1375
      return False
1376

    
1377
    test = local_version != remote_version[0]
1378
    _ErrorIf(test, self.ENODEVERSION, node,
1379
             "incompatible protocol versions: master %s,"
1380
             " node %s", local_version, remote_version[0])
1381
    if test:
1382
      return False
1383

    
1384
    # node seems compatible, we can actually try to look into its results
1385

    
1386
    # full package version
1387
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1388
                  self.ENODEVERSION, node,
1389
                  "software version mismatch: master %s, node %s",
1390
                  constants.RELEASE_VERSION, remote_version[1],
1391
                  code=self.ETYPE_WARNING)
1392

    
1393
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1394
    if ninfo.vm_capable and isinstance(hyp_result, dict):
1395
      for hv_name, hv_result in hyp_result.iteritems():
1396
        test = hv_result is not None
1397
        _ErrorIf(test, self.ENODEHV, node,
1398
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1399

    
1400
    hvp_result = nresult.get(constants.NV_HVPARAMS, None)
1401
    if ninfo.vm_capable and isinstance(hvp_result, list):
1402
      for item, hv_name, hv_result in hvp_result:
1403
        _ErrorIf(True, self.ENODEHV, node,
1404
                 "hypervisor %s parameter verify failure (source %s): %s",
1405
                 hv_name, item, hv_result)
1406

    
1407
    test = nresult.get(constants.NV_NODESETUP,
1408
                           ["Missing NODESETUP results"])
1409
    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
1410
             "; ".join(test))
1411

    
1412
    return True
1413

    
1414
  def _VerifyNodeTime(self, ninfo, nresult,
1415
                      nvinfo_starttime, nvinfo_endtime):
1416
    """Check the node time.
1417

1418
    @type ninfo: L{objects.Node}
1419
    @param ninfo: the node to check
1420
    @param nresult: the remote results for the node
1421
    @param nvinfo_starttime: the start time of the RPC call
1422
    @param nvinfo_endtime: the end time of the RPC call
1423

1424
    """
1425
    node = ninfo.name
1426
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1427

    
1428
    ntime = nresult.get(constants.NV_TIME, None)
1429
    try:
1430
      ntime_merged = utils.MergeTime(ntime)
1431
    except (ValueError, TypeError):
1432
      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
1433
      return
1434

    
1435
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1436
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1437
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1438
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1439
    else:
1440
      ntime_diff = None
1441

    
1442
    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
1443
             "Node time diverges by at least %s from master node time",
1444
             ntime_diff)
1445

    
1446
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1447
    """Check the node LVM results.
1448

1449
    @type ninfo: L{objects.Node}
1450
    @param ninfo: the node to check
1451
    @param nresult: the remote results for the node
1452
    @param vg_name: the configured VG name
1453

1454
    """
1455
    if vg_name is None:
1456
      return
1457

    
1458
    node = ninfo.name
1459
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1460

    
1461
    # checks vg existence and size > 20G
1462
    vglist = nresult.get(constants.NV_VGLIST, None)
1463
    test = not vglist
1464
    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
1465
    if not test:
1466
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
1467
                                            constants.MIN_VG_SIZE)
1468
      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
1469

    
1470
    # check pv names
1471
    pvlist = nresult.get(constants.NV_PVLIST, None)
1472
    test = pvlist is None
1473
    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
1474
    if not test:
1475
      # check that ':' is not present in PV names, since it's a
1476
      # special character for lvcreate (denotes the range of PEs to
1477
      # use on the PV)
1478
      for _, pvname, owner_vg in pvlist:
1479
        test = ":" in pvname
1480
        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
1481
                 " '%s' of VG '%s'", pvname, owner_vg)
1482

    
1483
  def _VerifyNodeBridges(self, ninfo, nresult, bridges):
1484
    """Check the node bridges.
1485

1486
    @type ninfo: L{objects.Node}
1487
    @param ninfo: the node to check
1488
    @param nresult: the remote results for the node
1489
    @param bridges: the expected list of bridges
1490

1491
    """
1492
    if not bridges:
1493
      return
1494

    
1495
    node = ninfo.name
1496
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1497

    
1498
    missing = nresult.get(constants.NV_BRIDGES, None)
1499
    test = not isinstance(missing, list)
1500
    _ErrorIf(test, self.ENODENET, node,
1501
             "did not return valid bridge information")
1502
    if not test:
1503
      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
1504
               utils.CommaJoin(sorted(missing)))
1505

    
1506
  def _VerifyNodeNetwork(self, ninfo, nresult):
1507
    """Check the node network connectivity results.
1508

1509
    @type ninfo: L{objects.Node}
1510
    @param ninfo: the node to check
1511
    @param nresult: the remote results for the node
1512

1513
    """
1514
    node = ninfo.name
1515
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1516

    
1517
    test = constants.NV_NODELIST not in nresult
1518
    _ErrorIf(test, self.ENODESSH, node,
1519
             "node hasn't returned node ssh connectivity data")
1520
    if not test:
1521
      if nresult[constants.NV_NODELIST]:
1522
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1523
          _ErrorIf(True, self.ENODESSH, node,
1524
                   "ssh communication with node '%s': %s", a_node, a_msg)
1525

    
1526
    test = constants.NV_NODENETTEST not in nresult
1527
    _ErrorIf(test, self.ENODENET, node,
1528
             "node hasn't returned node tcp connectivity data")
1529
    if not test:
1530
      if nresult[constants.NV_NODENETTEST]:
1531
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1532
        for anode in nlist:
1533
          _ErrorIf(True, self.ENODENET, node,
1534
                   "tcp communication with node '%s': %s",
1535
                   anode, nresult[constants.NV_NODENETTEST][anode])
1536

    
1537
    test = constants.NV_MASTERIP not in nresult
1538
    _ErrorIf(test, self.ENODENET, node,
1539
             "node hasn't returned node master IP reachability data")
1540
    if not test:
1541
      if not nresult[constants.NV_MASTERIP]:
1542
        if node == self.master_node:
1543
          msg = "the master node cannot reach the master IP (not configured?)"
1544
        else:
1545
          msg = "cannot reach the master IP"
1546
        _ErrorIf(True, self.ENODENET, node, msg)
1547

    
1548
  def _VerifyInstance(self, instance, instanceconfig, node_image,
1549
                      diskstatus):
1550
    """Verify an instance.
1551

1552
    This function checks to see if the required block devices are
1553
    available on the instance's node.
1554

1555
    """
1556
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1557
    node_current = instanceconfig.primary_node
1558

    
1559
    node_vol_should = {}
1560
    instanceconfig.MapLVsByNode(node_vol_should)
1561

    
1562
    for node in node_vol_should:
1563
      n_img = node_image[node]
1564
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1565
        # ignore missing volumes on offline or broken nodes
1566
        continue
1567
      for volume in node_vol_should[node]:
1568
        test = volume not in n_img.volumes
1569
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
1570
                 "volume %s missing on node %s", volume, node)
1571

    
1572
    if instanceconfig.admin_up:
1573
      pri_img = node_image[node_current]
1574
      test = instance not in pri_img.instances and not pri_img.offline
1575
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
1576
               "instance not running on its primary node %s",
1577
               node_current)
1578

    
1579
    for node, n_img in node_image.items():
1580
      if node != node_current:
1581
        test = instance in n_img.instances
1582
        _ErrorIf(test, self.EINSTANCEWRONGNODE, instance,
1583
                 "instance should not run on node %s", node)
1584

    
1585
    diskdata = [(nname, success, status, idx)
1586
                for (nname, disks) in diskstatus.items()
1587
                for idx, (success, status) in enumerate(disks)]
1588

    
1589
    for nname, success, bdev_status, idx in diskdata:
1590
      # the 'ghost node' construction in Exec() ensures that we have a
1591
      # node here
1592
      snode = node_image[nname]
1593
      bad_snode = snode.ghost or snode.offline
1594
      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
1595
               self.EINSTANCEFAULTYDISK, instance,
1596
               "couldn't retrieve status for disk/%s on %s: %s",
1597
               idx, nname, bdev_status)
1598
      _ErrorIf((instanceconfig.admin_up and success and
1599
                bdev_status.ldisk_status == constants.LDS_FAULTY),
1600
               self.EINSTANCEFAULTYDISK, instance,
1601
               "disk/%s on %s is faulty", idx, nname)
1602

    
1603
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
1604
    """Verify if there are any unknown volumes in the cluster.
1605

1606
    The .os, .swap and backup volumes are ignored. All other volumes are
1607
    reported as unknown.
1608

1609
    @type reserved: L{ganeti.utils.FieldSet}
1610
    @param reserved: a FieldSet of reserved volume names
1611

1612
    """
1613
    for node, n_img in node_image.items():
1614
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1615
        # skip non-healthy nodes
1616
        continue
1617
      for volume in n_img.volumes:
1618
        test = ((node not in node_vol_should or
1619
                volume not in node_vol_should[node]) and
1620
                not reserved.Matches(volume))
1621
        self._ErrorIf(test, self.ENODEORPHANLV, node,
1622
                      "volume %s is unknown", volume)
1623

    
1624
  def _VerifyOrphanInstances(self, instancelist, node_image):
1625
    """Verify the list of running instances.
1626

1627
    This checks what instances are running but unknown to the cluster.
1628

1629
    """
1630
    for node, n_img in node_image.items():
1631
      for o_inst in n_img.instances:
1632
        test = o_inst not in instancelist
1633
        self._ErrorIf(test, self.ENODEORPHANINSTANCE, node,
1634
                      "instance %s on node %s should not exist", o_inst, node)
1635

    
1636
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
1637
    """Verify N+1 Memory Resilience.
1638

1639
    Check that if one single node dies we can still start all the
1640
    instances it was primary for.
1641

1642
    """
1643
    for node, n_img in node_image.items():
1644
      # This code checks that every node which is now listed as
1645
      # secondary has enough memory to host all instances it is
1646
      # supposed to should a single other node in the cluster fail.
1647
      # FIXME: not ready for failover to an arbitrary node
1648
      # FIXME: does not support file-backed instances
1649
      # WARNING: we currently take into account down instances as well
1650
      # as up ones, considering that even if they're down someone
1651
      # might want to start them even in the event of a node failure.
1652
      if n_img.offline:
1653
        # we're skipping offline nodes from the N+1 warning, since
1654
        # most likely we don't have good memory infromation from them;
1655
        # we already list instances living on such nodes, and that's
1656
        # enough warning
1657
        continue
1658
      for prinode, instances in n_img.sbp.items():
1659
        needed_mem = 0
1660
        for instance in instances:
1661
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
1662
          if bep[constants.BE_AUTO_BALANCE]:
1663
            needed_mem += bep[constants.BE_MEMORY]
1664
        test = n_img.mfree < needed_mem
1665
        self._ErrorIf(test, self.ENODEN1, node,
1666
                      "not enough memory to accomodate instance failovers"
1667
                      " should node %s fail (%dMiB needed, %dMiB available)",
1668
                      prinode, needed_mem, n_img.mfree)
1669

    
1670
  def _VerifyNodeFiles(self, ninfo, nresult, file_list, local_cksum,
1671
                       master_files):
1672
    """Verifies and computes the node required file checksums.
1673

1674
    @type ninfo: L{objects.Node}
1675
    @param ninfo: the node to check
1676
    @param nresult: the remote results for the node
1677
    @param file_list: required list of files
1678
    @param local_cksum: dictionary of local files and their checksums
1679
    @param master_files: list of files that only masters should have
1680

1681
    """
1682
    node = ninfo.name
1683
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1684

    
1685
    remote_cksum = nresult.get(constants.NV_FILELIST, None)
1686
    test = not isinstance(remote_cksum, dict)
1687
    _ErrorIf(test, self.ENODEFILECHECK, node,
1688
             "node hasn't returned file checksum data")
1689
    if test:
1690
      return
1691

    
1692
    for file_name in file_list:
1693
      node_is_mc = ninfo.master_candidate
1694
      must_have = (file_name not in master_files) or node_is_mc
1695
      # missing
1696
      test1 = file_name not in remote_cksum
1697
      # invalid checksum
1698
      test2 = not test1 and remote_cksum[file_name] != local_cksum[file_name]
1699
      # existing and good
1700
      test3 = not test1 and remote_cksum[file_name] == local_cksum[file_name]
1701
      _ErrorIf(test1 and must_have, self.ENODEFILECHECK, node,
1702
               "file '%s' missing", file_name)
1703
      _ErrorIf(test2 and must_have, self.ENODEFILECHECK, node,
1704
               "file '%s' has wrong checksum", file_name)
1705
      # not candidate and this is not a must-have file
1706
      _ErrorIf(test2 and not must_have, self.ENODEFILECHECK, node,
1707
               "file '%s' should not exist on non master"
1708
               " candidates (and the file is outdated)", file_name)
1709
      # all good, except non-master/non-must have combination
1710
      _ErrorIf(test3 and not must_have, self.ENODEFILECHECK, node,
1711
               "file '%s' should not exist"
1712
               " on non master candidates", file_name)
1713

    
1714
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
1715
                      drbd_map):
1716
    """Verifies and the node DRBD status.
1717

1718
    @type ninfo: L{objects.Node}
1719
    @param ninfo: the node to check
1720
    @param nresult: the remote results for the node
1721
    @param instanceinfo: the dict of instances
1722
    @param drbd_helper: the configured DRBD usermode helper
1723
    @param drbd_map: the DRBD map as returned by
1724
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
1725

1726
    """
1727
    node = ninfo.name
1728
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1729

    
1730
    if drbd_helper:
1731
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
1732
      test = (helper_result == None)
1733
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
1734
               "no drbd usermode helper returned")
1735
      if helper_result:
1736
        status, payload = helper_result
1737
        test = not status
1738
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
1739
                 "drbd usermode helper check unsuccessful: %s", payload)
1740
        test = status and (payload != drbd_helper)
1741
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
1742
                 "wrong drbd usermode helper: %s", payload)
1743

    
1744
    # compute the DRBD minors
1745
    node_drbd = {}
1746
    for minor, instance in drbd_map[node].items():
1747
      test = instance not in instanceinfo
1748
      _ErrorIf(test, self.ECLUSTERCFG, None,
1749
               "ghost instance '%s' in temporary DRBD map", instance)
1750
        # ghost instance should not be running, but otherwise we
1751
        # don't give double warnings (both ghost instance and
1752
        # unallocated minor in use)
1753
      if test:
1754
        node_drbd[minor] = (instance, False)
1755
      else:
1756
        instance = instanceinfo[instance]
1757
        node_drbd[minor] = (instance.name, instance.admin_up)
1758

    
1759
    # and now check them
1760
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
1761
    test = not isinstance(used_minors, (tuple, list))
1762
    _ErrorIf(test, self.ENODEDRBD, node,
1763
             "cannot parse drbd status file: %s", str(used_minors))
1764
    if test:
1765
      # we cannot check drbd status
1766
      return
1767

    
1768
    for minor, (iname, must_exist) in node_drbd.items():
1769
      test = minor not in used_minors and must_exist
1770
      _ErrorIf(test, self.ENODEDRBD, node,
1771
               "drbd minor %d of instance %s is not active", minor, iname)
1772
    for minor in used_minors:
1773
      test = minor not in node_drbd
1774
      _ErrorIf(test, self.ENODEDRBD, node,
1775
               "unallocated drbd minor %d is in use", minor)
1776

    
1777
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
1778
    """Builds the node OS structures.
1779

1780
    @type ninfo: L{objects.Node}
1781
    @param ninfo: the node to check
1782
    @param nresult: the remote results for the node
1783
    @param nimg: the node image object
1784

1785
    """
1786
    node = ninfo.name
1787
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1788

    
1789
    remote_os = nresult.get(constants.NV_OSLIST, None)
1790
    test = (not isinstance(remote_os, list) or
1791
            not compat.all(isinstance(v, list) and len(v) == 7
1792
                           for v in remote_os))
1793

    
1794
    _ErrorIf(test, self.ENODEOS, node,
1795
             "node hasn't returned valid OS data")
1796

    
1797
    nimg.os_fail = test
1798

    
1799
    if test:
1800
      return
1801

    
1802
    os_dict = {}
1803

    
1804
    for (name, os_path, status, diagnose,
1805
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
1806

    
1807
      if name not in os_dict:
1808
        os_dict[name] = []
1809

    
1810
      # parameters is a list of lists instead of list of tuples due to
1811
      # JSON lacking a real tuple type, fix it:
1812
      parameters = [tuple(v) for v in parameters]
1813
      os_dict[name].append((os_path, status, diagnose,
1814
                            set(variants), set(parameters), set(api_ver)))
1815

    
1816
    nimg.oslist = os_dict
1817

    
1818
  def _VerifyNodeOS(self, ninfo, nimg, base):
1819
    """Verifies the node OS list.
1820

1821
    @type ninfo: L{objects.Node}
1822
    @param ninfo: the node to check
1823
    @param nimg: the node image object
1824
    @param base: the 'template' node we match against (e.g. from the master)
1825

1826
    """
1827
    node = ninfo.name
1828
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1829

    
1830
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
1831

    
1832
    beautify_params = lambda l: ["%s: %s" % (k, v) for (k, v) in l]
1833
    for os_name, os_data in nimg.oslist.items():
1834
      assert os_data, "Empty OS status for OS %s?!" % os_name
1835
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
1836
      _ErrorIf(not f_status, self.ENODEOS, node,
1837
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
1838
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
1839
               "OS '%s' has multiple entries (first one shadows the rest): %s",
1840
               os_name, utils.CommaJoin([v[0] for v in os_data]))
1841
      # this will catched in backend too
1842
      _ErrorIf(compat.any(v >= constants.OS_API_V15 for v in f_api)
1843
               and not f_var, self.ENODEOS, node,
1844
               "OS %s with API at least %d does not declare any variant",
1845
               os_name, constants.OS_API_V15)
1846
      # comparisons with the 'base' image
1847
      test = os_name not in base.oslist
1848
      _ErrorIf(test, self.ENODEOS, node,
1849
               "Extra OS %s not present on reference node (%s)",
1850
               os_name, base.name)
1851
      if test:
1852
        continue
1853
      assert base.oslist[os_name], "Base node has empty OS status?"
1854
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
1855
      if not b_status:
1856
        # base OS is invalid, skipping
1857
        continue
1858
      for kind, a, b in [("API version", f_api, b_api),
1859
                         ("variants list", f_var, b_var),
1860
                         ("parameters", beautify_params(f_param),
1861
                          beautify_params(b_param))]:
1862
        _ErrorIf(a != b, self.ENODEOS, node,
1863
                 "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
1864
                 kind, os_name, base.name,
1865
                 utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
1866

    
1867
    # check any missing OSes
1868
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
1869
    _ErrorIf(missing, self.ENODEOS, node,
1870
             "OSes present on reference node %s but missing on this node: %s",
1871
             base.name, utils.CommaJoin(missing))
1872

    
1873
  def _VerifyOob(self, ninfo, nresult):
1874
    """Verifies out of band functionality of a node.
1875

1876
    @type ninfo: L{objects.Node}
1877
    @param ninfo: the node to check
1878
    @param nresult: the remote results for the node
1879

1880
    """
1881
    node = ninfo.name
1882
    # We just have to verify the paths on master and/or master candidates
1883
    # as the oob helper is invoked on the master
1884
    if ((ninfo.master_candidate or ninfo.master_capable) and
1885
        constants.NV_OOB_PATHS in nresult):
1886
      for path_result in nresult[constants.NV_OOB_PATHS]:
1887
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
1888

    
1889
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
1890
    """Verifies and updates the node volume data.
1891

1892
    This function will update a L{NodeImage}'s internal structures
1893
    with data from the remote call.
1894

1895
    @type ninfo: L{objects.Node}
1896
    @param ninfo: the node to check
1897
    @param nresult: the remote results for the node
1898
    @param nimg: the node image object
1899
    @param vg_name: the configured VG name
1900

1901
    """
1902
    node = ninfo.name
1903
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1904

    
1905
    nimg.lvm_fail = True
1906
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
1907
    if vg_name is None:
1908
      pass
1909
    elif isinstance(lvdata, basestring):
1910
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
1911
               utils.SafeEncode(lvdata))
1912
    elif not isinstance(lvdata, dict):
1913
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
1914
    else:
1915
      nimg.volumes = lvdata
1916
      nimg.lvm_fail = False
1917

    
1918
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
1919
    """Verifies and updates the node instance list.
1920

1921
    If the listing was successful, then updates this node's instance
1922
    list. Otherwise, it marks the RPC call as failed for the instance
1923
    list key.
1924

1925
    @type ninfo: L{objects.Node}
1926
    @param ninfo: the node to check
1927
    @param nresult: the remote results for the node
1928
    @param nimg: the node image object
1929

1930
    """
1931
    idata = nresult.get(constants.NV_INSTANCELIST, None)
1932
    test = not isinstance(idata, list)
1933
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
1934
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
1935
    if test:
1936
      nimg.hyp_fail = True
1937
    else:
1938
      nimg.instances = idata
1939

    
1940
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
1941
    """Verifies and computes a node information map
1942

1943
    @type ninfo: L{objects.Node}
1944
    @param ninfo: the node to check
1945
    @param nresult: the remote results for the node
1946
    @param nimg: the node image object
1947
    @param vg_name: the configured VG name
1948

1949
    """
1950
    node = ninfo.name
1951
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1952

    
1953
    # try to read free memory (from the hypervisor)
1954
    hv_info = nresult.get(constants.NV_HVINFO, None)
1955
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
1956
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
1957
    if not test:
1958
      try:
1959
        nimg.mfree = int(hv_info["memory_free"])
1960
      except (ValueError, TypeError):
1961
        _ErrorIf(True, self.ENODERPC, node,
1962
                 "node returned invalid nodeinfo, check hypervisor")
1963

    
1964
    # FIXME: devise a free space model for file based instances as well
1965
    if vg_name is not None:
1966
      test = (constants.NV_VGLIST not in nresult or
1967
              vg_name not in nresult[constants.NV_VGLIST])
1968
      _ErrorIf(test, self.ENODELVM, node,
1969
               "node didn't return data for the volume group '%s'"
1970
               " - it is either missing or broken", vg_name)
1971
      if not test:
1972
        try:
1973
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
1974
        except (ValueError, TypeError):
1975
          _ErrorIf(True, self.ENODERPC, node,
1976
                   "node returned invalid LVM info, check LVM status")
1977

    
1978
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
1979
    """Gets per-disk status information for all instances.
1980

1981
    @type nodelist: list of strings
1982
    @param nodelist: Node names
1983
    @type node_image: dict of (name, L{objects.Node})
1984
    @param node_image: Node objects
1985
    @type instanceinfo: dict of (name, L{objects.Instance})
1986
    @param instanceinfo: Instance objects
1987
    @rtype: {instance: {node: [(succes, payload)]}}
1988
    @return: a dictionary of per-instance dictionaries with nodes as
1989
        keys and disk information as values; the disk information is a
1990
        list of tuples (success, payload)
1991

1992
    """
1993
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1994

    
1995
    node_disks = {}
1996
    node_disks_devonly = {}
1997
    diskless_instances = set()
1998
    diskless = constants.DT_DISKLESS
1999

    
2000
    for nname in nodelist:
2001
      node_instances = list(itertools.chain(node_image[nname].pinst,
2002
                                            node_image[nname].sinst))
2003
      diskless_instances.update(inst for inst in node_instances
2004
                                if instanceinfo[inst].disk_template == diskless)
2005
      disks = [(inst, disk)
2006
               for inst in node_instances
2007
               for disk in instanceinfo[inst].disks]
2008

    
2009
      if not disks:
2010
        # No need to collect data
2011
        continue
2012

    
2013
      node_disks[nname] = disks
2014

    
2015
      # Creating copies as SetDiskID below will modify the objects and that can
2016
      # lead to incorrect data returned from nodes
2017
      devonly = [dev.Copy() for (_, dev) in disks]
2018

    
2019
      for dev in devonly:
2020
        self.cfg.SetDiskID(dev, nname)
2021

    
2022
      node_disks_devonly[nname] = devonly
2023

    
2024
    assert len(node_disks) == len(node_disks_devonly)
2025

    
2026
    # Collect data from all nodes with disks
2027
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2028
                                                          node_disks_devonly)
2029

    
2030
    assert len(result) == len(node_disks)
2031

    
2032
    instdisk = {}
2033

    
2034
    for (nname, nres) in result.items():
2035
      disks = node_disks[nname]
2036

    
2037
      if nres.offline:
2038
        # No data from this node
2039
        data = len(disks) * [(False, "node offline")]
2040
      else:
2041
        msg = nres.fail_msg
2042
        _ErrorIf(msg, self.ENODERPC, nname,
2043
                 "while getting disk information: %s", msg)
2044
        if msg:
2045
          # No data from this node
2046
          data = len(disks) * [(False, msg)]
2047
        else:
2048
          data = []
2049
          for idx, i in enumerate(nres.payload):
2050
            if isinstance(i, (tuple, list)) and len(i) == 2:
2051
              data.append(i)
2052
            else:
2053
              logging.warning("Invalid result from node %s, entry %d: %s",
2054
                              nname, idx, i)
2055
              data.append((False, "Invalid result from the remote node"))
2056

    
2057
      for ((inst, _), status) in zip(disks, data):
2058
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2059

    
2060
    # Add empty entries for diskless instances.
2061
    for inst in diskless_instances:
2062
      assert inst not in instdisk
2063
      instdisk[inst] = {}
2064

    
2065
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2066
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2067
                      compat.all(isinstance(s, (tuple, list)) and
2068
                                 len(s) == 2 for s in statuses)
2069
                      for inst, nnames in instdisk.items()
2070
                      for nname, statuses in nnames.items())
2071
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2072

    
2073
    return instdisk
2074

    
2075
  def _VerifyHVP(self, hvp_data):
2076
    """Verifies locally the syntax of the hypervisor parameters.
2077

2078
    """
2079
    for item, hv_name, hv_params in hvp_data:
2080
      msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
2081
             (item, hv_name))
2082
      try:
2083
        hv_class = hypervisor.GetHypervisor(hv_name)
2084
        utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
2085
        hv_class.CheckParameterSyntax(hv_params)
2086
      except errors.GenericError, err:
2087
        self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
2088

    
2089

    
2090
  def BuildHooksEnv(self):
2091
    """Build hooks env.
2092

2093
    Cluster-Verify hooks just ran in the post phase and their failure makes
2094
    the output be logged in the verify output and the verification to fail.
2095

2096
    """
2097
    all_nodes = self.cfg.GetNodeList()
2098
    env = {
2099
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2100
      }
2101
    for node in self.cfg.GetAllNodesInfo().values():
2102
      env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())
2103

    
2104
    return env, [], all_nodes
2105

    
2106
  def Exec(self, feedback_fn):
2107
    """Verify integrity of cluster, performing various test on nodes.
2108

2109
    """
2110
    # This method has too many local variables. pylint: disable-msg=R0914
2111
    self.bad = False
2112
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2113
    verbose = self.op.verbose
2114
    self._feedback_fn = feedback_fn
2115
    feedback_fn("* Verifying global settings")
2116
    for msg in self.cfg.VerifyConfig():
2117
      _ErrorIf(True, self.ECLUSTERCFG, None, msg)
2118

    
2119
    # Check the cluster certificates
2120
    for cert_filename in constants.ALL_CERT_FILES:
2121
      (errcode, msg) = _VerifyCertificate(cert_filename)
2122
      _ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
2123

    
2124
    vg_name = self.cfg.GetVGName()
2125
    drbd_helper = self.cfg.GetDRBDHelper()
2126
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
2127
    cluster = self.cfg.GetClusterInfo()
2128
    nodeinfo_byname = self.cfg.GetAllNodesInfo()
2129
    nodelist = utils.NiceSort(nodeinfo_byname.keys())
2130
    nodeinfo = [nodeinfo_byname[nname] for nname in nodelist]
2131
    instanceinfo = self.cfg.GetAllInstancesInfo()
2132
    instancelist = utils.NiceSort(instanceinfo.keys())
2133
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2134
    i_non_redundant = [] # Non redundant instances
2135
    i_non_a_balanced = [] # Non auto-balanced instances
2136
    n_offline = 0 # Count of offline nodes
2137
    n_drained = 0 # Count of nodes being drained
2138
    node_vol_should = {}
2139

    
2140
    # FIXME: verify OS list
2141
    # do local checksums
2142
    master_files = [constants.CLUSTER_CONF_FILE]
2143
    master_node = self.master_node = self.cfg.GetMasterNode()
2144
    master_ip = self.cfg.GetMasterIP()
2145

    
2146
    file_names = ssconf.SimpleStore().GetFileList()
2147
    file_names.extend(constants.ALL_CERT_FILES)
2148
    file_names.extend(master_files)
2149
    if cluster.modify_etc_hosts:
2150
      file_names.append(constants.ETC_HOSTS)
2151

    
2152
    local_checksums = utils.FingerprintFiles(file_names)
2153

    
2154
    # Compute the set of hypervisor parameters
2155
    hvp_data = []
2156
    for hv_name in hypervisors:
2157
      hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
2158
    for os_name, os_hvp in cluster.os_hvp.items():
2159
      for hv_name, hv_params in os_hvp.items():
2160
        if not hv_params:
2161
          continue
2162
        full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
2163
        hvp_data.append(("os %s" % os_name, hv_name, full_params))
2164
    # TODO: collapse identical parameter values in a single one
2165
    for instance in instanceinfo.values():
2166
      if not instance.hvparams:
2167
        continue
2168
      hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
2169
                       cluster.FillHV(instance)))
2170
    # and verify them locally
2171
    self._VerifyHVP(hvp_data)
2172

    
2173
    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
2174
    node_verify_param = {
2175
      constants.NV_FILELIST: file_names,
2176
      constants.NV_NODELIST: [node.name for node in nodeinfo
2177
                              if not node.offline],
2178
      constants.NV_HYPERVISOR: hypervisors,
2179
      constants.NV_HVPARAMS: hvp_data,
2180
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
2181
                                  node.secondary_ip) for node in nodeinfo
2182
                                 if not node.offline],
2183
      constants.NV_INSTANCELIST: hypervisors,
2184
      constants.NV_VERSION: None,
2185
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2186
      constants.NV_NODESETUP: None,
2187
      constants.NV_TIME: None,
2188
      constants.NV_MASTERIP: (master_node, master_ip),
2189
      constants.NV_OSLIST: None,
2190
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2191
      }
2192

    
2193
    if vg_name is not None:
2194
      node_verify_param[constants.NV_VGLIST] = None
2195
      node_verify_param[constants.NV_LVLIST] = vg_name
2196
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2197
      node_verify_param[constants.NV_DRBDLIST] = None
2198

    
2199
    if drbd_helper:
2200
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2201

    
2202
    # bridge checks
2203
    # FIXME: this needs to be changed per node-group, not cluster-wide
2204
    bridges = set()
2205
    default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
2206
    if default_nicpp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2207
      bridges.add(default_nicpp[constants.NIC_LINK])
2208
    for instance in instanceinfo.values():
2209
      for nic in instance.nics:
2210
        full_nic = cluster.SimpleFillNIC(nic.nicparams)
2211
        if full_nic[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
2212
          bridges.add(full_nic[constants.NIC_LINK])
2213

    
2214
    if bridges:
2215
      node_verify_param[constants.NV_BRIDGES] = list(bridges)
2216

    
2217
    # Build our expected cluster state
2218
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2219
                                                 name=node.name,
2220
                                                 vm_capable=node.vm_capable))
2221
                      for node in nodeinfo)
2222

    
2223
    # Gather OOB paths
2224
    oob_paths = []
2225
    for node in nodeinfo:
2226
      path = _SupportsOob(self.cfg, node)
2227
      if path and path not in oob_paths:
2228
        oob_paths.append(path)
2229

    
2230
    if oob_paths:
2231
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2232

    
2233
    for instance in instancelist:
2234
      inst_config = instanceinfo[instance]
2235

    
2236
      for nname in inst_config.all_nodes:
2237
        if nname not in node_image:
2238
          # ghost node
2239
          gnode = self.NodeImage(name=nname)
2240
          gnode.ghost = True
2241
          node_image[nname] = gnode
2242

    
2243
      inst_config.MapLVsByNode(node_vol_should)
2244

    
2245
      pnode = inst_config.primary_node
2246
      node_image[pnode].pinst.append(instance)
2247

    
2248
      for snode in inst_config.secondary_nodes:
2249
        nimg = node_image[snode]
2250
        nimg.sinst.append(instance)
2251
        if pnode not in nimg.sbp:
2252
          nimg.sbp[pnode] = []
2253
        nimg.sbp[pnode].append(instance)
2254

    
2255
    # At this point, we have the in-memory data structures complete,
2256
    # except for the runtime information, which we'll gather next
2257

    
2258
    # Due to the way our RPC system works, exact response times cannot be
2259
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2260
    # time before and after executing the request, we can at least have a time
2261
    # window.
2262
    nvinfo_starttime = time.time()
2263
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
2264
                                           self.cfg.GetClusterName())
2265
    nvinfo_endtime = time.time()
2266

    
2267
    all_drbd_map = self.cfg.ComputeDRBDMap()
2268

    
2269
    feedback_fn("* Gathering disk information (%s nodes)" % len(nodelist))
2270
    instdisk = self._CollectDiskInfo(nodelist, node_image, instanceinfo)
2271

    
2272
    feedback_fn("* Verifying node status")
2273

    
2274
    refos_img = None
2275

    
2276
    for node_i in nodeinfo:
2277
      node = node_i.name
2278
      nimg = node_image[node]
2279

    
2280
      if node_i.offline:
2281
        if verbose:
2282
          feedback_fn("* Skipping offline node %s" % (node,))
2283
        n_offline += 1
2284
        continue
2285

    
2286
      if node == master_node:
2287
        ntype = "master"
2288
      elif node_i.master_candidate:
2289
        ntype = "master candidate"
2290
      elif node_i.drained:
2291
        ntype = "drained"
2292
        n_drained += 1
2293
      else:
2294
        ntype = "regular"
2295
      if verbose:
2296
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2297

    
2298
      msg = all_nvinfo[node].fail_msg
2299
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2300
      if msg:
2301
        nimg.rpc_fail = True
2302
        continue
2303

    
2304
      nresult = all_nvinfo[node].payload
2305

    
2306
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2307
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2308
      self._VerifyNodeNetwork(node_i, nresult)
2309
      self._VerifyNodeFiles(node_i, nresult, file_names, local_checksums,
2310
                            master_files)
2311

    
2312
      self._VerifyOob(node_i, nresult)
2313

    
2314
      if nimg.vm_capable:
2315
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2316
        self._VerifyNodeDrbd(node_i, nresult, instanceinfo, drbd_helper,
2317
                             all_drbd_map)
2318

    
2319
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2320
        self._UpdateNodeInstances(node_i, nresult, nimg)
2321
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2322
        self._UpdateNodeOS(node_i, nresult, nimg)
2323
        if not nimg.os_fail:
2324
          if refos_img is None:
2325
            refos_img = nimg
2326
          self._VerifyNodeOS(node_i, nimg, refos_img)
2327
        self._VerifyNodeBridges(node_i, nresult, bridges)
2328

    
2329
    feedback_fn("* Verifying instance status")
2330
    for instance in instancelist:
2331
      if verbose:
2332
        feedback_fn("* Verifying instance %s" % instance)
2333
      inst_config = instanceinfo[instance]
2334
      self._VerifyInstance(instance, inst_config, node_image,
2335
                           instdisk[instance])
2336
      inst_nodes_offline = []
2337

    
2338
      pnode = inst_config.primary_node
2339
      pnode_img = node_image[pnode]
2340
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2341
               self.ENODERPC, pnode, "instance %s, connection to"
2342
               " primary node failed", instance)
2343

    
2344
      _ErrorIf(pnode_img.offline, self.EINSTANCEBADNODE, instance,
2345
               "instance lives on offline node %s", inst_config.primary_node)
2346

    
2347
      # If the instance is non-redundant we cannot survive losing its primary
2348
      # node, so we are not N+1 compliant. On the other hand we have no disk
2349
      # templates with more than one secondary so that situation is not well
2350
      # supported either.
2351
      # FIXME: does not support file-backed instances
2352
      if not inst_config.secondary_nodes:
2353
        i_non_redundant.append(instance)
2354

    
2355
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2356
               instance, "instance has multiple secondary nodes: %s",
2357
               utils.CommaJoin(inst_config.secondary_nodes),
2358
               code=self.ETYPE_WARNING)
2359

    
2360
      if inst_config.disk_template in constants.DTS_NET_MIRROR:
2361
        pnode = inst_config.primary_node
2362
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2363
        instance_groups = {}
2364

    
2365
        for node in instance_nodes:
2366
          instance_groups.setdefault(nodeinfo_byname[node].group,
2367
                                     []).append(node)
2368

    
2369
        pretty_list = [
2370
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2371
          # Sort so that we always list the primary node first.
2372
          for group, nodes in sorted(instance_groups.items(),
2373
                                     key=lambda (_, nodes): pnode in nodes,
2374
                                     reverse=True)]
2375

    
2376
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2377
                      instance, "instance has primary and secondary nodes in"
2378
                      " different groups: %s", utils.CommaJoin(pretty_list),
2379
                      code=self.ETYPE_WARNING)
2380

    
2381
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2382
        i_non_a_balanced.append(instance)
2383

    
2384
      for snode in inst_config.secondary_nodes:
2385
        s_img = node_image[snode]
2386
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2387
                 "instance %s, connection to secondary node failed", instance)
2388

    
2389
        if s_img.offline:
2390
          inst_nodes_offline.append(snode)
2391

    
2392
      # warn that the instance lives on offline nodes
2393
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2394
               "instance has offline secondary node(s) %s",
2395
               utils.CommaJoin(inst_nodes_offline))
2396
      # ... or ghost/non-vm_capable nodes
2397
      for node in inst_config.all_nodes:
2398
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2399
                 "instance lives on ghost node %s", node)
2400
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2401
                 instance, "instance lives on non-vm_capable node %s", node)
2402

    
2403
    feedback_fn("* Verifying orphan volumes")
2404
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2405
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2406

    
2407
    feedback_fn("* Verifying orphan instances")
2408
    self._VerifyOrphanInstances(instancelist, node_image)
2409

    
2410
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2411
      feedback_fn("* Verifying N+1 Memory redundancy")
2412
      self._VerifyNPlusOneMemory(node_image, instanceinfo)
2413

    
2414
    feedback_fn("* Other Notes")
2415
    if i_non_redundant:
2416
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2417
                  % len(i_non_redundant))
2418

    
2419
    if i_non_a_balanced:
2420
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2421
                  % len(i_non_a_balanced))
2422

    
2423
    if n_offline:
2424
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2425

    
2426
    if n_drained:
2427
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2428

    
2429
    return not self.bad
2430

    
2431
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2432
    """Analyze the post-hooks' result
2433

2434
    This method analyses the hook result, handles it, and sends some
2435
    nicely-formatted feedback back to the user.
2436

2437
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
2438
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
2439
    @param hooks_results: the results of the multi-node hooks rpc call
2440
    @param feedback_fn: function used send feedback back to the caller
2441
    @param lu_result: previous Exec result
2442
    @return: the new Exec result, based on the previous result
2443
        and hook results
2444

2445
    """
2446
    # We only really run POST phase hooks, and are only interested in
2447
    # their results
2448
    if phase == constants.HOOKS_PHASE_POST:
2449
      # Used to change hooks' output to proper indentation
2450
      feedback_fn("* Hooks Results")
2451
      assert hooks_results, "invalid result from hooks"
2452

    
2453
      for node_name in hooks_results:
2454
        res = hooks_results[node_name]
2455
        msg = res.fail_msg
2456
        test = msg and not res.offline
2457
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
2458
                      "Communication failure in hooks execution: %s", msg)
2459
        if res.offline or msg:
2460
          # No need to investigate payload if node is offline or gave an error.
2461
          # override manually lu_result here as _ErrorIf only
2462
          # overrides self.bad
2463
          lu_result = 1
2464
          continue
2465
        for script, hkr, output in res.payload:
2466
          test = hkr == constants.HKR_FAIL
2467
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
2468
                        "Script %s failed, output:", script)
2469
          if test:
2470
            output = self._HOOKS_INDENT_RE.sub('      ', output)
2471
            feedback_fn("%s" % output)
2472
            lu_result = 0
2473

    
2474
      return lu_result
2475

    
2476

    
2477
class LUClusterVerifyDisks(NoHooksLU):
2478
  """Verifies the cluster disks status.
2479

2480
  """
2481
  REQ_BGL = False
2482

    
2483
  def ExpandNames(self):
2484
    self.needed_locks = {
2485
      locking.LEVEL_NODE: locking.ALL_SET,
2486
      locking.LEVEL_INSTANCE: locking.ALL_SET,
2487
    }
2488
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
2489

    
2490
  def Exec(self, feedback_fn):
2491
    """Verify integrity of cluster disks.
2492

2493
    @rtype: tuple of three items
2494
    @return: a tuple of (dict of node-to-node_error, list of instances
2495
        which need activate-disks, dict of instance: (node, volume) for
2496
        missing volumes
2497

2498
    """
2499
    result = res_nodes, res_instances, res_missing = {}, [], {}
2500

    
2501
    nodes = utils.NiceSort(self.cfg.GetVmCapableNodeList())
2502
    instances = self.cfg.GetAllInstancesInfo().values()
2503

    
2504
    nv_dict = {}
2505
    for inst in instances:
2506
      inst_lvs = {}
2507
      if not inst.admin_up:
2508
        continue
2509
      inst.MapLVsByNode(inst_lvs)
2510
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
2511
      for node, vol_list in inst_lvs.iteritems():
2512
        for vol in vol_list:
2513
          nv_dict[(node, vol)] = inst
2514

    
2515
    if not nv_dict:
2516
      return result
2517

    
2518
    node_lvs = self.rpc.call_lv_list(nodes, [])
2519
    for node, node_res in node_lvs.items():
2520
      if node_res.offline:
2521
        continue
2522
      msg = node_res.fail_msg
2523
      if msg:
2524
        logging.warning("Error enumerating LVs on node %s: %s", node, msg)
2525
        res_nodes[node] = msg
2526
        continue
2527

    
2528
      lvs = node_res.payload
2529
      for lv_name, (_, _, lv_online) in lvs.items():
2530
        inst = nv_dict.pop((node, lv_name), None)
2531
        if (not lv_online and inst is not None
2532
            and inst.name not in res_instances):
2533
          res_instances.append(inst.name)
2534

    
2535
    # any leftover items in nv_dict are missing LVs, let's arrange the
2536
    # data better
2537
    for key, inst in nv_dict.iteritems():
2538
      if inst.name not in res_missing:
2539
        res_missing[inst.name] = []
2540
      res_missing[inst.name].append(key)
2541

    
2542
    return result
2543

    
2544

    
2545
class LUClusterRepairDiskSizes(NoHooksLU):
2546
  """Verifies the cluster disks sizes.
2547

2548
  """
2549
  REQ_BGL = False
2550

    
2551
  def ExpandNames(self):
2552
    if self.op.instances:
2553
      self.wanted_names = []
2554
      for name in self.op.instances:
2555
        full_name = _ExpandInstanceName(self.cfg, name)
2556
        self.wanted_names.append(full_name)
2557
      self.needed_locks = {
2558
        locking.LEVEL_NODE: [],
2559
        locking.LEVEL_INSTANCE: self.wanted_names,
2560
        }
2561
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2562
    else:
2563
      self.wanted_names = None
2564
      self.needed_locks = {
2565
        locking.LEVEL_NODE: locking.ALL_SET,
2566
        locking.LEVEL_INSTANCE: locking.ALL_SET,
2567
        }
2568
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
2569

    
2570
  def DeclareLocks(self, level):
2571
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
2572
      self._LockInstancesNodes(primary_only=True)
2573

    
2574
  def CheckPrereq(self):
2575
    """Check prerequisites.
2576

2577
    This only checks the optional instance list against the existing names.
2578

2579
    """
2580
    if self.wanted_names is None:
2581
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
2582

    
2583
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
2584
                             in self.wanted_names]
2585

    
2586
  def _EnsureChildSizes(self, disk):
2587
    """Ensure children of the disk have the needed disk size.
2588

2589
    This is valid mainly for DRBD8 and fixes an issue where the
2590
    children have smaller disk size.
2591

2592
    @param disk: an L{ganeti.objects.Disk} object
2593

2594
    """
2595
    if disk.dev_type == constants.LD_DRBD8:
2596
      assert disk.children, "Empty children for DRBD8?"
2597
      fchild = disk.children[0]
2598
      mismatch = fchild.size < disk.size
2599
      if mismatch:
2600
        self.LogInfo("Child disk has size %d, parent %d, fixing",
2601
                     fchild.size, disk.size)
2602
        fchild.size = disk.size
2603

    
2604
      # and we recurse on this child only, not on the metadev
2605
      return self._EnsureChildSizes(fchild) or mismatch
2606
    else:
2607
      return False
2608

    
2609
  def Exec(self, feedback_fn):
2610
    """Verify the size of cluster disks.
2611

2612
    """
2613
    # TODO: check child disks too
2614
    # TODO: check differences in size between primary/secondary nodes
2615
    per_node_disks = {}
2616
    for instance in self.wanted_instances:
2617
      pnode = instance.primary_node
2618
      if pnode not in per_node_disks:
2619
        per_node_disks[pnode] = []
2620
      for idx, disk in enumerate(instance.disks):
2621
        per_node_disks[pnode].append((instance, idx, disk))
2622

    
2623
    changed = []
2624
    for node, dskl in per_node_disks.items():
2625
      newl = [v[2].Copy() for v in dskl]
2626
      for dsk in newl:
2627
        self.cfg.SetDiskID(dsk, node)
2628
      result = self.rpc.call_blockdev_getsize(node, newl)
2629
      if result.fail_msg:
2630
        self.LogWarning("Failure in blockdev_getsize call to node"
2631
                        " %s, ignoring", node)
2632
        continue
2633
      if len(result.payload) != len(dskl):
2634
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
2635
                        " result.payload=%s", node, len(dskl), result.payload)
2636
        self.LogWarning("Invalid result from node %s, ignoring node results",
2637
                        node)
2638
        continue
2639
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
2640
        if size is None:
2641
          self.LogWarning("Disk %d of instance %s did not return size"
2642
                          " information, ignoring", idx, instance.name)
2643
          continue
2644
        if not isinstance(size, (int, long)):
2645
          self.LogWarning("Disk %d of instance %s did not return valid"
2646
                          " size information, ignoring", idx, instance.name)
2647
          continue
2648
        size = size >> 20
2649
        if size != disk.size:
2650
          self.LogInfo("Disk %d of instance %s has mismatched size,"
2651
                       " correcting: recorded %d, actual %d", idx,
2652
                       instance.name, disk.size, size)
2653
          disk.size = size
2654
          self.cfg.Update(instance, feedback_fn)
2655
          changed.append((instance.name, idx, size))
2656
        if self._EnsureChildSizes(disk):
2657
          self.cfg.Update(instance, feedback_fn)
2658
          changed.append((instance.name, idx, disk.size))
2659
    return changed
2660

    
2661

    
2662
class LUClusterRename(LogicalUnit):
2663
  """Rename the cluster.
2664

2665
  """
2666
  HPATH = "cluster-rename"
2667
  HTYPE = constants.HTYPE_CLUSTER
2668

    
2669
  def BuildHooksEnv(self):
2670
    """Build hooks env.
2671

2672
    """
2673
    env = {
2674
      "OP_TARGET": self.cfg.GetClusterName(),
2675
      "NEW_NAME": self.op.name,
2676
      }
2677
    mn = self.cfg.GetMasterNode()
2678
    all_nodes = self.cfg.GetNodeList()
2679
    return env, [mn], all_nodes
2680

    
2681
  def CheckPrereq(self):
2682
    """Verify that the passed name is a valid one.
2683

2684
    """
2685
    hostname = netutils.GetHostname(name=self.op.name,
2686
                                    family=self.cfg.GetPrimaryIPFamily())
2687

    
2688
    new_name = hostname.name
2689
    self.ip = new_ip = hostname.ip
2690
    old_name = self.cfg.GetClusterName()
2691
    old_ip = self.cfg.GetMasterIP()
2692
    if new_name == old_name and new_ip == old_ip:
2693
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
2694
                                 " cluster has changed",
2695
                                 errors.ECODE_INVAL)
2696
    if new_ip != old_ip:
2697
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
2698
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
2699
                                   " reachable on the network" %
2700
                                   new_ip, errors.ECODE_NOTUNIQUE)
2701

    
2702
    self.op.name = new_name
2703

    
2704
  def Exec(self, feedback_fn):
2705
    """Rename the cluster.
2706

2707
    """
2708
    clustername = self.op.name
2709
    ip = self.ip
2710

    
2711
    # shutdown the master IP
2712
    master = self.cfg.GetMasterNode()
2713
    result = self.rpc.call_node_stop_master(master, False)
2714
    result.Raise("Could not disable the master role")
2715

    
2716
    try:
2717
      cluster = self.cfg.GetClusterInfo()
2718
      cluster.cluster_name = clustername
2719
      cluster.master_ip = ip
2720
      self.cfg.Update(cluster, feedback_fn)
2721

    
2722
      # update the known hosts file
2723
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
2724
      node_list = self.cfg.GetOnlineNodeList()
2725
      try:
2726
        node_list.remove(master)
2727
      except ValueError:
2728
        pass
2729
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
2730
    finally:
2731
      result = self.rpc.call_node_start_master(master, False, False)
2732
      msg = result.fail_msg
2733
      if msg:
2734
        self.LogWarning("Could not re-enable the master role on"
2735
                        " the master, please restart manually: %s", msg)
2736

    
2737
    return clustername
2738

    
2739

    
2740
class LUClusterSetParams(LogicalUnit):
2741
  """Change the parameters of the cluster.
2742

2743
  """
2744
  HPATH = "cluster-modify"
2745
  HTYPE = constants.HTYPE_CLUSTER
2746
  REQ_BGL = False
2747

    
2748
  def CheckArguments(self):
2749
    """Check parameters
2750

2751
    """
2752
    if self.op.uid_pool:
2753
      uidpool.CheckUidPool(self.op.uid_pool)
2754

    
2755
    if self.op.add_uids:
2756
      uidpool.CheckUidPool(self.op.add_uids)
2757

    
2758
    if self.op.remove_uids:
2759
      uidpool.CheckUidPool(self.op.remove_uids)
2760

    
2761
  def ExpandNames(self):
2762
    # FIXME: in the future maybe other cluster params won't require checking on
2763
    # all nodes to be modified.
2764
    self.needed_locks = {
2765
      locking.LEVEL_NODE: locking.ALL_SET,
2766
    }
2767
    self.share_locks[locking.LEVEL_NODE] = 1
2768

    
2769
  def BuildHooksEnv(self):
2770
    """Build hooks env.
2771

2772
    """
2773
    env = {
2774
      "OP_TARGET": self.cfg.GetClusterName(),
2775
      "NEW_VG_NAME": self.op.vg_name,
2776
      }
2777
    mn = self.cfg.GetMasterNode()
2778
    return env, [mn], [mn]
2779

    
2780
  def CheckPrereq(self):
2781
    """Check prerequisites.
2782

2783
    This checks whether the given params don't conflict and
2784
    if the given volume group is valid.
2785

2786
    """
2787
    if self.op.vg_name is not None and not self.op.vg_name:
2788
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
2789
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
2790
                                   " instances exist", errors.ECODE_INVAL)
2791

    
2792
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
2793
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
2794
        raise errors.OpPrereqError("Cannot disable drbd helper while"
2795
                                   " drbd-based instances exist",
2796
                                   errors.ECODE_INVAL)
2797

    
2798
    node_list = self.acquired_locks[locking.LEVEL_NODE]
2799

    
2800
    # if vg_name not None, checks given volume group on all nodes
2801
    if self.op.vg_name:
2802
      vglist = self.rpc.call_vg_list(node_list)
2803
      for node in node_list:
2804
        msg = vglist[node].fail_msg
2805
        if msg:
2806
          # ignoring down node
2807
          self.LogWarning("Error while gathering data on node %s"
2808
                          " (ignoring node): %s", node, msg)
2809
          continue
2810
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
2811
                                              self.op.vg_name,
2812
                                              constants.MIN_VG_SIZE)
2813
        if vgstatus:
2814
          raise errors.OpPrereqError("Error on node '%s': %s" %
2815
                                     (node, vgstatus), errors.ECODE_ENVIRON)
2816

    
2817
    if self.op.drbd_helper:
2818
      # checks given drbd helper on all nodes
2819
      helpers = self.rpc.call_drbd_helper(node_list)
2820
      for node in node_list:
2821
        ninfo = self.cfg.GetNodeInfo(node)
2822
        if ninfo.offline:
2823
          self.LogInfo("Not checking drbd helper on offline node %s", node)
2824
          continue
2825
        msg = helpers[node].fail_msg
2826
        if msg:
2827
          raise errors.OpPrereqError("Error checking drbd helper on node"
2828
                                     " '%s': %s" % (node, msg),
2829
                                     errors.ECODE_ENVIRON)
2830
        node_helper = helpers[node].payload
2831
        if node_helper != self.op.drbd_helper:
2832
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
2833
                                     (node, node_helper), errors.ECODE_ENVIRON)
2834

    
2835
    self.cluster = cluster = self.cfg.GetClusterInfo()
2836
    # validate params changes
2837
    if self.op.beparams:
2838
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
2839
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
2840

    
2841
    if self.op.ndparams:
2842
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
2843
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
2844

    
2845
      # TODO: we need a more general way to handle resetting
2846
      # cluster-level parameters to default values
2847
      if self.new_ndparams["oob_program"] == "":
2848
        self.new_ndparams["oob_program"] = \
2849
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
2850

    
2851
    if self.op.nicparams:
2852
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
2853
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
2854
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
2855
      nic_errors = []
2856

    
2857
      # check all instances for consistency
2858
      for instance in self.cfg.GetAllInstancesInfo().values():
2859
        for nic_idx, nic in enumerate(instance.nics):
2860
          params_copy = copy.deepcopy(nic.nicparams)
2861
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
2862

    
2863
          # check parameter syntax
2864
          try:
2865
            objects.NIC.CheckParameterSyntax(params_filled)
2866
          except errors.ConfigurationError, err:
2867
            nic_errors.append("Instance %s, nic/%d: %s" %
2868
                              (instance.name, nic_idx, err))
2869

    
2870
          # if we're moving instances to routed, check that they have an ip
2871
          target_mode = params_filled[constants.NIC_MODE]
2872
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
2873
            nic_errors.append("Instance %s, nic/%d: routed NIC with no ip"
2874
                              " address" % (instance.name, nic_idx))
2875
      if nic_errors:
2876
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
2877
                                   "\n".join(nic_errors))
2878

    
2879
    # hypervisor list/parameters
2880
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
2881
    if self.op.hvparams:
2882
      for hv_name, hv_dict in self.op.hvparams.items():
2883
        if hv_name not in self.new_hvparams:
2884
          self.new_hvparams[hv_name] = hv_dict
2885
        else:
2886
          self.new_hvparams[hv_name].update(hv_dict)
2887

    
2888
    # os hypervisor parameters
2889
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
2890
    if self.op.os_hvp:
2891
      for os_name, hvs in self.op.os_hvp.items():
2892
        if os_name not in self.new_os_hvp:
2893
          self.new_os_hvp[os_name] = hvs
2894
        else:
2895
          for hv_name, hv_dict in hvs.items():
2896
            if hv_name not in self.new_os_hvp[os_name]:
2897
              self.new_os_hvp[os_name][hv_name] = hv_dict
2898
            else:
2899
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
2900

    
2901
    # os parameters
2902
    self.new_osp = objects.FillDict(cluster.osparams, {})
2903
    if self.op.osparams:
2904
      for os_name, osp in self.op.osparams.items():
2905
        if os_name not in self.new_osp:
2906
          self.new_osp[os_name] = {}
2907

    
2908
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
2909
                                                  use_none=True)
2910

    
2911
        if not self.new_osp[os_name]:
2912
          # we removed all parameters
2913
          del self.new_osp[os_name]
2914
        else:
2915
          # check the parameter validity (remote check)
2916
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
2917
                         os_name, self.new_osp[os_name])
2918

    
2919
    # changes to the hypervisor list
2920
    if self.op.enabled_hypervisors is not None:
2921
      self.hv_list = self.op.enabled_hypervisors
2922
      for hv in self.hv_list:
2923
        # if the hypervisor doesn't already exist in the cluster
2924
        # hvparams, we initialize it to empty, and then (in both
2925
        # cases) we make sure to fill the defaults, as we might not
2926
        # have a complete defaults list if the hypervisor wasn't
2927
        # enabled before
2928
        if hv not in new_hvp:
2929
          new_hvp[hv] = {}
2930
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
2931
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
2932
    else:
2933
      self.hv_list = cluster.enabled_hypervisors
2934

    
2935
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
2936
      # either the enabled list has changed, or the parameters have, validate
2937
      for hv_name, hv_params in self.new_hvparams.items():
2938
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
2939
            (self.op.enabled_hypervisors and
2940
             hv_name in self.op.enabled_hypervisors)):
2941
          # either this is a new hypervisor, or its parameters have changed
2942
          hv_class = hypervisor.GetHypervisor(hv_name)
2943
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
2944
          hv_class.CheckParameterSyntax(hv_params)
2945
          _CheckHVParams(self, node_list, hv_name, hv_params)
2946

    
2947
    if self.op.os_hvp:
2948
      # no need to check any newly-enabled hypervisors, since the
2949
      # defaults have already been checked in the above code-block
2950
      for os_name, os_hvp in self.new_os_hvp.items():
2951
        for hv_name, hv_params in os_hvp.items():
2952
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
2953
          # we need to fill in the new os_hvp on top of the actual hv_p
2954
          cluster_defaults = self.new_hvparams.get(hv_name, {})
2955
          new_osp = objects.FillDict(cluster_defaults, hv_params)
2956
          hv_class = hypervisor.GetHypervisor(hv_name)
2957
          hv_class.CheckParameterSyntax(new_osp)
2958
          _CheckHVParams(self, node_list, hv_name, new_osp)
2959

    
2960
    if self.op.default_iallocator:
2961
      alloc_script = utils.FindFile(self.op.default_iallocator,
2962
                                    constants.IALLOCATOR_SEARCH_PATH,
2963
                                    os.path.isfile)
2964
      if alloc_script is None:
2965
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
2966
                                   " specified" % self.op.default_iallocator,
2967
                                   errors.ECODE_INVAL)
2968

    
2969
  def Exec(self, feedback_fn):
2970
    """Change the parameters of the cluster.
2971

2972
    """
2973
    if self.op.vg_name is not None:
2974
      new_volume = self.op.vg_name
2975
      if not new_volume:
2976
        new_volume = None
2977
      if new_volume != self.cfg.GetVGName():
2978
        self.cfg.SetVGName(new_volume)
2979
      else:
2980
        feedback_fn("Cluster LVM configuration already in desired"
2981
                    " state, not changing")
2982
    if self.op.drbd_helper is not None:
2983
      new_helper = self.op.drbd_helper
2984
      if not new_helper:
2985
        new_helper = None
2986
      if new_helper != self.cfg.GetDRBDHelper():
2987
        self.cfg.SetDRBDHelper(new_helper)
2988
      else:
2989
        feedback_fn("Cluster DRBD helper already in desired state,"
2990
                    " not changing")
2991
    if self.op.hvparams:
2992
      self.cluster.hvparams = self.new_hvparams
2993
    if self.op.os_hvp:
2994
      self.cluster.os_hvp = self.new_os_hvp
2995
    if self.op.enabled_hypervisors is not None:
2996
      self.cluster.hvparams = self.new_hvparams
2997
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
2998
    if self.op.beparams:
2999
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
3000
    if self.op.nicparams:
3001
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
3002
    if self.op.osparams:
3003
      self.cluster.osparams = self.new_osp
3004
    if self.op.ndparams:
3005
      self.cluster.ndparams = self.new_ndparams
3006

    
3007
    if self.op.candidate_pool_size is not None:
3008
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
3009
      # we need to update the pool size here, otherwise the save will fail
3010
      _AdjustCandidatePool(self, [])
3011

    
3012
    if self.op.maintain_node_health is not None:
3013
      self.cluster.maintain_node_health = self.op.maintain_node_health
3014

    
3015
    if self.op.prealloc_wipe_disks is not None:
3016
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
3017

    
3018
    if self.op.add_uids is not None:
3019
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
3020

    
3021
    if self.op.remove_uids is not None:
3022
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
3023

    
3024
    if self.op.uid_pool is not None:
3025
      self.cluster.uid_pool = self.op.uid_pool
3026

    
3027
    if self.op.default_iallocator is not None:
3028
      self.cluster.default_iallocator = self.op.default_iallocator
3029

    
3030
    if self.op.reserved_lvs is not None:
3031
      self.cluster.reserved_lvs = self.op.reserved_lvs
3032

    
3033
    def helper_os(aname, mods, desc):
3034
      desc += " OS list"
3035
      lst = getattr(self.cluster, aname)
3036
      for key, val in mods:
3037
        if key == constants.DDM_ADD:
3038
          if val in lst:
3039
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
3040
          else:
3041
            lst.append(val)
3042
        elif key == constants.DDM_REMOVE:
3043
          if val in lst:
3044
            lst.remove(val)
3045
          else:
3046
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3047
        else:
3048
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3049

    
3050
    if self.op.hidden_os:
3051
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3052

    
3053
    if self.op.blacklisted_os:
3054
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3055

    
3056
    if self.op.master_netdev:
3057
      master = self.cfg.GetMasterNode()
3058
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3059
                  self.cluster.master_netdev)
3060
      result = self.rpc.call_node_stop_master(master, False)
3061
      result.Raise("Could not disable the master ip")
3062
      feedback_fn("Changing master_netdev from %s to %s" %
3063
                  (self.cluster.master_netdev, self.op.master_netdev))
3064
      self.cluster.master_netdev = self.op.master_netdev
3065

    
3066
    self.cfg.Update(self.cluster, feedback_fn)
3067

    
3068
    if self.op.master_netdev:
3069
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3070
                  self.op.master_netdev)
3071
      result = self.rpc.call_node_start_master(master, False, False)
3072
      if result.fail_msg:
3073
        self.LogWarning("Could not re-enable the master ip on"
3074
                        " the master, please restart manually: %s",
3075
                        result.fail_msg)
3076

    
3077

    
3078
def _UploadHelper(lu, nodes, fname):
3079
  """Helper for uploading a file and showing warnings.
3080

3081
  """
3082
  if os.path.exists(fname):
3083
    result = lu.rpc.call_upload_file(nodes, fname)
3084
    for to_node, to_result in result.items():
3085
      msg = to_result.fail_msg
3086
      if msg:
3087
        msg = ("Copy of file %s to node %s failed: %s" %
3088
               (fname, to_node, msg))
3089
        lu.proc.LogWarning(msg)
3090

    
3091

    
3092
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3093
  """Distribute additional files which are part of the cluster configuration.
3094

3095
  ConfigWriter takes care of distributing the config and ssconf files, but
3096
  there are more files which should be distributed to all nodes. This function
3097
  makes sure those are copied.
3098

3099
  @param lu: calling logical unit
3100
  @param additional_nodes: list of nodes not in the config to distribute to
3101
  @type additional_vm: boolean
3102
  @param additional_vm: whether the additional nodes are vm-capable or not
3103

3104
  """
3105
  # 1. Gather target nodes
3106
  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3107
  dist_nodes = lu.cfg.GetOnlineNodeList()
3108
  nvm_nodes = lu.cfg.GetNonVmCapableNodeList()
3109
  vm_nodes = [name for name in dist_nodes if name not in nvm_nodes]
3110
  if additional_nodes is not None:
3111
    dist_nodes.extend(additional_nodes)
3112
    if additional_vm:
3113
      vm_nodes.extend(additional_nodes)
3114
  if myself.name in dist_nodes:
3115
    dist_nodes.remove(myself.name)
3116
  if myself.name in vm_nodes:
3117
    vm_nodes.remove(myself.name)
3118

    
3119
  # 2. Gather files to distribute
3120
  dist_files = set([constants.ETC_HOSTS,
3121
                    constants.SSH_KNOWN_HOSTS_FILE,
3122
                    constants.RAPI_CERT_FILE,
3123
                    constants.RAPI_USERS_FILE,
3124
                    constants.CONFD_HMAC_KEY,
3125
                    constants.CLUSTER_DOMAIN_SECRET_FILE,
3126
                   ])
3127

    
3128
  vm_files = set()
3129
  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
3130
  for hv_name in enabled_hypervisors:
3131
    hv_class = hypervisor.GetHypervisor(hv_name)
3132
    vm_files.update(hv_class.GetAncillaryFiles())
3133

    
3134
  # 3. Perform the files upload
3135
  for fname in dist_files:
3136
    _UploadHelper(lu, dist_nodes, fname)
3137
  for fname in vm_files:
3138
    _UploadHelper(lu, vm_nodes, fname)
3139

    
3140

    
3141
class LUClusterRedistConf(NoHooksLU):
3142
  """Force the redistribution of cluster configuration.
3143

3144
  This is a very simple LU.
3145

3146
  """
3147
  REQ_BGL = False
3148

    
3149
  def ExpandNames(self):
3150
    self.needed_locks = {
3151
      locking.LEVEL_NODE: locking.ALL_SET,
3152
    }
3153
    self.share_locks[locking.LEVEL_NODE] = 1
3154

    
3155
  def Exec(self, feedback_fn):
3156
    """Redistribute the configuration.
3157

3158
    """
3159
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3160
    _RedistributeAncillaryFiles(self)
3161

    
3162

    
3163
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3164
  """Sleep and poll for an instance's disk to sync.
3165

3166
  """
3167
  if not instance.disks or disks is not None and not disks:
3168
    return True
3169

    
3170
  disks = _ExpandCheckDisks(instance, disks)
3171

    
3172
  if not oneshot:
3173
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3174

    
3175
  node = instance.primary_node
3176

    
3177
  for dev in disks:
3178
    lu.cfg.SetDiskID(dev, node)
3179

    
3180
  # TODO: Convert to utils.Retry
3181

    
3182
  retries = 0
3183
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3184
  while True:
3185
    max_time = 0
3186
    done = True
3187
    cumul_degraded = False
3188
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3189
    msg = rstats.fail_msg
3190
    if msg:
3191
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3192
      retries += 1
3193
      if retries >= 10:
3194
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3195
                                 " aborting." % node)
3196
      time.sleep(6)
3197
      continue
3198
    rstats = rstats.payload
3199
    retries = 0
3200
    for i, mstat in enumerate(rstats):
3201
      if mstat is None:
3202
        lu.LogWarning("Can't compute data for node %s/%s",
3203
                           node, disks[i].iv_name)
3204
        continue
3205

    
3206
      cumul_degraded = (cumul_degraded or
3207
                        (mstat.is_degraded and mstat.sync_percent is None))
3208
      if mstat.sync_percent is not None:
3209
        done = False
3210
        if mstat.estimated_time is not None:
3211
          rem_time = ("%s remaining (estimated)" %
3212
                      utils.FormatSeconds(mstat.estimated_time))
3213
          max_time = mstat.estimated_time
3214
        else:
3215
          rem_time = "no time estimate"
3216
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3217
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3218

    
3219
    # if we're done but degraded, let's do a few small retries, to
3220
    # make sure we see a stable and not transient situation; therefore
3221
    # we force restart of the loop
3222
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3223
      logging.info("Degraded disks found, %d retries left", degr_retries)
3224
      degr_retries -= 1
3225
      time.sleep(1)
3226
      continue
3227

    
3228
    if done or oneshot:
3229
      break
3230

    
3231
    time.sleep(min(60, max_time))
3232

    
3233
  if done:
3234
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3235
  return not cumul_degraded
3236

    
3237

    
3238
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3239
  """Check that mirrors are not degraded.
3240

3241
  The ldisk parameter, if True, will change the test from the
3242
  is_degraded attribute (which represents overall non-ok status for
3243
  the device(s)) to the ldisk (representing the local storage status).
3244

3245
  """
3246
  lu.cfg.SetDiskID(dev, node)
3247

    
3248
  result = True
3249

    
3250
  if on_primary or dev.AssembleOnSecondary():
3251
    rstats = lu.rpc.call_blockdev_find(node, dev)
3252
    msg = rstats.fail_msg
3253
    if msg:
3254
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3255
      result = False
3256
    elif not rstats.payload:
3257
      lu.LogWarning("Can't find disk on node %s", node)
3258
      result = False
3259
    else:
3260
      if ldisk:
3261
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3262
      else:
3263
        result = result and not rstats.payload.is_degraded
3264

    
3265
  if dev.children:
3266
    for child in dev.children:
3267
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3268

    
3269
  return result
3270

    
3271

    
3272
class LUOobCommand(NoHooksLU):
3273
  """Logical unit for OOB handling.
3274

3275
  """
3276
  REG_BGL = False
3277

    
3278
  def CheckPrereq(self):
3279
    """Check prerequisites.
3280

3281
    This checks:
3282
     - the node exists in the configuration
3283
     - OOB is supported
3284

3285
    Any errors are signaled by raising errors.OpPrereqError.
3286

3287
    """
3288
    self.nodes = []
3289
    for node_name in self.op.node_names:
3290
      node = self.cfg.GetNodeInfo(node_name)
3291

    
3292
      if node is None:
3293
        raise errors.OpPrereqError("Node %s not found" % node_name,
3294
                                   errors.ECODE_NOENT)
3295
      else:
3296
        self.nodes.append(node)
3297

    
3298
      if (self.op.command == constants.OOB_POWER_OFF and not node.offline):
3299
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
3300
                                    " not marked offline") % node_name,
3301
                                   errors.ECODE_STATE)
3302

    
3303
  def ExpandNames(self):
3304
    """Gather locks we need.
3305

3306
    """
3307
    if self.op.node_names:
3308
      self.op.node_names = [_ExpandNodeName(self.cfg, name)
3309
                            for name in self.op.node_names]
3310
    else:
3311
      self.op.node_names = self.cfg.GetNodeList()
3312

    
3313
    self.needed_locks = {
3314
      locking.LEVEL_NODE: self.op.node_names,
3315
      }
3316

    
3317
  def Exec(self, feedback_fn):
3318
    """Execute OOB and return result if we expect any.
3319

3320
    """
3321
    master_node = self.cfg.GetMasterNode()
3322
    ret = []
3323

    
3324
    for node in self.nodes:
3325
      node_entry = [(constants.RS_NORMAL, node.name)]
3326
      ret.append(node_entry)
3327

    
3328
      oob_program = _SupportsOob(self.cfg, node)
3329

    
3330
      if not oob_program:
3331
        node_entry.append((constants.RS_UNAVAIL, None))
3332
        continue
3333

    
3334
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
3335
                   self.op.command, oob_program, node.name)
3336
      result = self.rpc.call_run_oob(master_node, oob_program,
3337
                                     self.op.command, node.name,
3338
                                     self.op.timeout)
3339

    
3340
      if result.fail_msg:
3341
        self.LogWarning("On node '%s' out-of-band RPC failed with: %s",
3342
                        node.name, result.fail_msg)
3343
        node_entry.append((constants.RS_NODATA, None))
3344
      else:
3345
        try:
3346
          self._CheckPayload(result)
3347
        except errors.OpExecError, err:
3348
          self.LogWarning("The payload returned by '%s' is not valid: %s",
3349
                          node.name, err)
3350
          node_entry.append((constants.RS_NODATA, None))
3351
        else:
3352
          if self.op.command == constants.OOB_HEALTH:
3353
            # For health we should log important events
3354
            for item, status in result.payload:
3355
              if status in [constants.OOB_STATUS_WARNING,
3356
                            constants.OOB_STATUS_CRITICAL]:
3357
                self.LogWarning("On node '%s' item '%s' has status '%s'",
3358
                                node.name, item, status)
3359

    
3360
          if self.op.command == constants.OOB_POWER_ON:
3361
            node.powered = True
3362
          elif self.op.command == constants.OOB_POWER_OFF:
3363
            node.powered = False
3364
          elif self.op.command == constants.OOB_POWER_STATUS:
3365
            powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
3366
            if powered != node.powered:
3367
              logging.warning(("Recorded power state (%s) of node '%s' does not"
3368
                               " match actual power state (%s)"), node.powered,
3369
                              node.name, powered)
3370

    
3371
          # For configuration changing commands we should update the node
3372
          if self.op.command in (constants.OOB_POWER_ON,
3373
                                 constants.OOB_POWER_OFF):
3374
            self.cfg.Update(node, feedback_fn)
3375

    
3376
          node_entry.append((constants.RS_NORMAL, result.payload))
3377

    
3378
    return ret
3379

    
3380
  def _CheckPayload(self, result):
3381
    """Checks if the payload is valid.
3382

3383
    @param result: RPC result
3384
    @raises errors.OpExecError: If payload is not valid
3385

3386
    """
3387
    errs = []
3388
    if self.op.command == constants.OOB_HEALTH:
3389
      if not isinstance(result.payload, list):
3390
        errs.append("command 'health' is expected to return a list but got %s" %
3391
                    type(result.payload))
3392
      else:
3393
        for item, status in result.payload:
3394
          if status not in constants.OOB_STATUSES:
3395
            errs.append("health item '%s' has invalid status '%s'" %
3396
                        (item, status))
3397

    
3398
    if self.op.command == constants.OOB_POWER_STATUS:
3399
      if not isinstance(result.payload, dict):
3400
        errs.append("power-status is expected to return a dict but got %s" %
3401
                    type(result.payload))
3402

    
3403
    if self.op.command in [
3404
        constants.OOB_POWER_ON,
3405
        constants.OOB_POWER_OFF,
3406
        constants.OOB_POWER_CYCLE,
3407
        ]:
3408
      if result.payload is not None:
3409
        errs.append("%s is expected to not return payload but got '%s'" %
3410
                    (self.op.command, result.payload))
3411

    
3412
    if errs:
3413
      raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
3414
                               utils.CommaJoin(errs))
3415

    
3416

    
3417

    
3418
class LUOsDiagnose(NoHooksLU):
3419
  """Logical unit for OS diagnose/query.
3420

3421
  """
3422
  REQ_BGL = False
3423
  _HID = "hidden"
3424
  _BLK = "blacklisted"
3425
  _VLD = "valid"
3426
  _FIELDS_STATIC = utils.FieldSet()
3427
  _FIELDS_DYNAMIC = utils.FieldSet("name", _VLD, "node_status", "variants",
3428
                                   "parameters", "api_versions", _HID, _BLK)
3429

    
3430
  def CheckArguments(self):
3431
    if self.op.names:
3432
      raise errors.OpPrereqError("Selective OS query not supported",
3433
                                 errors.ECODE_INVAL)
3434

    
3435
    _CheckOutputFields(static=self._FIELDS_STATIC,
3436
                       dynamic=self._FIELDS_DYNAMIC,
3437
                       selected=self.op.output_fields)
3438

    
3439
  def ExpandNames(self):
3440
    # Lock all nodes, in shared mode
3441
    # Temporary removal of locks, should be reverted later
3442
    # TODO: reintroduce locks when they are lighter-weight
3443
    self.needed_locks = {}
3444
    #self.share_locks[locking.LEVEL_NODE] = 1
3445
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3446

    
3447
  @staticmethod
3448
  def _DiagnoseByOS(rlist):
3449
    """Remaps a per-node return list into an a per-os per-node dictionary
3450

3451
    @param rlist: a map with node names as keys and OS objects as values
3452

3453
    @rtype: dict
3454
    @return: a dictionary with osnames as keys and as value another
3455
        map, with nodes as keys and tuples of (path, status, diagnose,
3456
        variants, parameters, api_versions) as values, eg::
3457

3458
          {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
3459
                                     (/srv/..., False, "invalid api")],
3460
                           "node2": [(/srv/..., True, "", [], [])]}
3461
          }
3462

3463
    """
3464
    all_os = {}
3465
    # we build here the list of nodes that didn't fail the RPC (at RPC
3466
    # level), so that nodes with a non-responding node daemon don't
3467
    # make all OSes invalid
3468
    good_nodes = [node_name for node_name in rlist
3469
                  if not rlist[node_name].fail_msg]
3470
    for node_name, nr in rlist.items():
3471
      if nr.fail_msg or not nr.payload:
3472
        continue
3473
      for (name, path, status, diagnose, variants,
3474
           params, api_versions) in nr.payload:
3475
        if name not in all_os:
3476
          # build a list of nodes for this os containing empty lists
3477
          # for each node in node_list
3478
          all_os[name] = {}
3479
          for nname in good_nodes:
3480
            all_os[name][nname] = []
3481
        # convert params from [name, help] to (name, help)
3482
        params = [tuple(v) for v in params]
3483
        all_os[name][node_name].append((path, status, diagnose,
3484
                                        variants, params, api_versions))
3485
    return all_os
3486

    
3487
  def Exec(self, feedback_fn):
3488
    """Compute the list of OSes.
3489

3490
    """
3491
    valid_nodes = [node.name
3492
                   for node in self.cfg.GetAllNodesInfo().values()
3493
                   if not node.offline and node.vm_capable]
3494
    node_data = self.rpc.call_os_diagnose(valid_nodes)
3495
    pol = self._DiagnoseByOS(node_data)
3496
    output = []
3497
    cluster = self.cfg.GetClusterInfo()
3498

    
3499
    for os_name in utils.NiceSort(pol.keys()):
3500
      os_data = pol[os_name]
3501
      row = []
3502
      valid = True
3503
      (variants, params, api_versions) = null_state = (set(), set(), set())
3504
      for idx, osl in enumerate(os_data.values()):
3505
        valid = bool(valid and osl and osl[0][1])
3506
        if not valid:
3507
          (variants, params, api_versions) = null_state
3508
          break
3509
        node_variants, node_params, node_api = osl[0][3:6]
3510
        if idx == 0: # first entry
3511
          variants = set(node_variants)
3512
          params = set(node_params)
3513
          api_versions = set(node_api)
3514
        else: # keep consistency
3515
          variants.intersection_update(node_variants)
3516
          params.intersection_update(node_params)
3517
          api_versions.intersection_update(node_api)
3518

    
3519
      is_hid = os_name in cluster.hidden_os
3520
      is_blk = os_name in cluster.blacklisted_os
3521
      if ((self._HID not in self.op.output_fields and is_hid) or
3522
          (self._BLK not in self.op.output_fields and is_blk) or
3523
          (self._VLD not in self.op.output_fields and not valid)):
3524
        continue
3525

    
3526
      for field in self.op.output_fields:
3527
        if field == "name":
3528
          val = os_name
3529
        elif field == self._VLD:
3530
          val = valid
3531
        elif field == "node_status":
3532
          # this is just a copy of the dict
3533
          val = {}
3534
          for node_name, nos_list in os_data.items():
3535
            val[node_name] = nos_list
3536
        elif field == "variants":
3537
          val = utils.NiceSort(list(variants))
3538
        elif field == "parameters":
3539
          val = list(params)
3540
        elif field == "api_versions":
3541
          val = list(api_versions)
3542
        elif field == self._HID:
3543
          val = is_hid
3544
        elif field == self._BLK:
3545
          val = is_blk
3546
        else:
3547
          raise errors.ParameterError(field)
3548
        row.append(val)
3549
      output.append(row)
3550

    
3551
    return output
3552

    
3553

    
3554
class LUNodeRemove(LogicalUnit):
3555
  """Logical unit for removing a node.
3556

3557
  """
3558
  HPATH = "node-remove"
3559
  HTYPE = constants.HTYPE_NODE
3560

    
3561
  def BuildHooksEnv(self):
3562
    """Build hooks env.
3563

3564
    This doesn't run on the target node in the pre phase as a failed
3565
    node would then be impossible to remove.
3566

3567
    """
3568
    env = {
3569
      "OP_TARGET": self.op.node_name,
3570
      "NODE_NAME": self.op.node_name,
3571
      }
3572
    all_nodes = self.cfg.GetNodeList()
3573
    try:
3574
      all_nodes.remove(self.op.node_name)
3575
    except ValueError:
3576
      logging.warning("Node %s which is about to be removed not found"
3577
                      " in the all nodes list", self.op.node_name)
3578
    return env, all_nodes, all_nodes
3579

    
3580
  def CheckPrereq(self):
3581
    """Check prerequisites.
3582

3583
    This checks:
3584
     - the node exists in the configuration
3585
     - it does not have primary or secondary instances
3586
     - it's not the master
3587

3588
    Any errors are signaled by raising errors.OpPrereqError.
3589

3590
    """
3591
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
3592
    node = self.cfg.GetNodeInfo(self.op.node_name)
3593
    assert node is not None
3594

    
3595
    instance_list = self.cfg.GetInstanceList()
3596

    
3597
    masternode = self.cfg.GetMasterNode()
3598
    if node.name == masternode:
3599
      raise errors.OpPrereqError("Node is the master node,"
3600
                                 " you need to failover first.",
3601
                                 errors.ECODE_INVAL)
3602

    
3603
    for instance_name in instance_list:
3604
      instance = self.cfg.GetInstanceInfo(instance_name)
3605
      if node.name in instance.all_nodes:
3606
        raise errors.OpPrereqError("Instance %s is still running on the node,"
3607
                                   " please remove first." % instance_name,
3608
                                   errors.ECODE_INVAL)
3609
    self.op.node_name = node.name
3610
    self.node = node
3611

    
3612
  def Exec(self, feedback_fn):
3613
    """Removes the node from the cluster.
3614

3615
    """
3616
    node = self.node
3617
    logging.info("Stopping the node daemon and removing configs from node %s",
3618
                 node.name)
3619

    
3620
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
3621

    
3622
    # Promote nodes to master candidate as needed
3623
    _AdjustCandidatePool(self, exceptions=[node.name])
3624
    self.context.RemoveNode(node.name)
3625

    
3626
    # Run post hooks on the node before it's removed
3627
    hm = self.proc.hmclass(self.rpc.call_hooks_runner, self)
3628
    try:
3629
      hm.RunPhase(constants.HOOKS_PHASE_POST, [node.name])
3630
    except:
3631
      # pylint: disable-msg=W0702
3632
      self.LogWarning("Errors occurred running hooks on %s" % node.name)
3633

    
3634
    result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
3635
    msg = result.fail_msg
3636
    if msg:
3637
      self.LogWarning("Errors encountered on the remote node while leaving"
3638
                      " the cluster: %s", msg)
3639

    
3640
    # Remove node from our /etc/hosts
3641
    if self.cfg.GetClusterInfo().modify_etc_hosts:
3642
      master_node = self.cfg.GetMasterNode()
3643
      result = self.rpc.call_etc_hosts_modify(master_node,
3644
                                              constants.ETC_HOSTS_REMOVE,
3645
                                              node.name, None)
3646
      result.Raise("Can't update hosts file with new host data")
3647
      _RedistributeAncillaryFiles(self)
3648

    
3649

    
3650
class _NodeQuery(_QueryBase):
3651
  FIELDS = query.NODE_FIELDS
3652

    
3653
  def ExpandNames(self, lu):
3654
    lu.needed_locks = {}
3655
    lu.share_locks[locking.LEVEL_NODE] = 1
3656

    
3657
    if self.names:
3658
      self.wanted = _GetWantedNodes(lu, self.names)
3659
    else:
3660
      self.wanted = locking.ALL_SET
3661

    
3662
    self.do_locking = (self.use_locking and
3663
                       query.NQ_LIVE in self.requested_data)
3664

    
3665
    if self.do_locking:
3666
      # if we don't request only static fields, we need to lock the nodes
3667
      lu.needed_locks[locking.LEVEL_NODE] = self.wanted
3668

    
3669
  def DeclareLocks(self, lu, level):
3670
    pass
3671

    
3672
  def _GetQueryData(self, lu):
3673
    """Computes the list of nodes and their attributes.
3674

3675
    """
3676
    all_info = lu.cfg.GetAllNodesInfo()
3677

    
3678
    nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
3679

    
3680
    # Gather data as requested
3681
    if query.NQ_LIVE in self.requested_data:
3682
      # filter out non-vm_capable nodes
3683
      toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
3684

    
3685
      node_data = lu.rpc.call_node_info(toquery_nodes, lu.cfg.GetVGName(),
3686
                                        lu.cfg.GetHypervisorType())
3687
      live_data = dict((name, nresult.payload)
3688
                       for (name, nresult) in node_data.items()
3689
                       if not nresult.fail_msg and nresult.payload)
3690
    else:
3691
      live_data = None
3692

    
3693
    if query.NQ_INST in self.requested_data:
3694
      node_to_primary = dict([(name, set()) for name in nodenames])
3695
      node_to_secondary = dict([(name, set()) for name in nodenames])
3696

    
3697
      inst_data = lu.cfg.GetAllInstancesInfo()
3698

    
3699
      for inst in inst_data.values():
3700
        if inst.primary_node in node_to_primary:
3701
          node_to_primary[inst.primary_node].add(inst.name)
3702
        for secnode in inst.secondary_nodes:
3703
          if secnode in node_to_secondary:
3704
            node_to_secondary[secnode].add(inst.name)
3705
    else:
3706
      node_to_primary = None
3707
      node_to_secondary = None
3708

    
3709
    if query.NQ_OOB in self.requested_data:
3710
      oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
3711
                         for name, node in all_info.iteritems())
3712
    else:
3713
      oob_support = None
3714

    
3715
    if query.NQ_GROUP in self.requested_data:
3716
      groups = lu.cfg.GetAllNodeGroupsInfo()
3717
    else:
3718
      groups = {}
3719

    
3720
    return query.NodeQueryData([all_info[name] for name in nodenames],
3721
                               live_data, lu.cfg.GetMasterNode(),
3722
                               node_to_primary, node_to_secondary, groups,
3723
                               oob_support, lu.cfg.GetClusterInfo())
3724

    
3725

    
3726
class LUNodeQuery(NoHooksLU):
3727
  """Logical unit for querying nodes.
3728

3729
  """
3730
  # pylint: disable-msg=W0142
3731
  REQ_BGL = False
3732

    
3733
  def CheckArguments(self):
3734
    self.nq = _NodeQuery(self.op.names, self.op.output_fields,
3735
                         self.op.use_locking)
3736

    
3737
  def ExpandNames(self):
3738
    self.nq.ExpandNames(self)
3739

    
3740
  def Exec(self, feedback_fn):
3741
    return self.nq.OldStyleQuery(self)
3742

    
3743

    
3744
class LUNodeQueryvols(NoHooksLU):
3745
  """Logical unit for getting volumes on node(s).
3746

3747
  """
3748
  REQ_BGL = False
3749
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
3750
  _FIELDS_STATIC = utils.FieldSet("node")
3751

    
3752
  def CheckArguments(self):
3753
    _CheckOutputFields(static=self._FIELDS_STATIC,
3754
                       dynamic=self._FIELDS_DYNAMIC,
3755
                       selected=self.op.output_fields)
3756

    
3757
  def ExpandNames(self):
3758
    self.needed_locks = {}
3759
    self.share_locks[locking.LEVEL_NODE] = 1
3760
    if not self.op.nodes:
3761
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3762
    else:
3763
      self.needed_locks[locking.LEVEL_NODE] = \
3764
        _GetWantedNodes(self, self.op.nodes)
3765

    
3766
  def Exec(self, feedback_fn):
3767
    """Computes the list of nodes and their attributes.
3768

3769
    """
3770
    nodenames = self.acquired_locks[locking.LEVEL_NODE]
3771
    volumes = self.rpc.call_node_volumes(nodenames)
3772

    
3773
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
3774
             in self.cfg.GetInstanceList()]
3775

    
3776
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
3777

    
3778
    output = []
3779
    for node in nodenames:
3780
      nresult = volumes[node]
3781
      if nresult.offline:
3782
        continue
3783
      msg = nresult.fail_msg
3784
      if msg:
3785
        self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
3786
        continue
3787

    
3788
      node_vols = nresult.payload[:]
3789
      node_vols.sort(key=lambda vol: vol['dev'])
3790

    
3791
      for vol in node_vols:
3792
        node_output = []
3793
        for field in self.op.output_fields:
3794
          if field == "node":
3795
            val = node
3796
          elif field == "phys":
3797
            val = vol['dev']
3798
          elif field == "vg":
3799
            val = vol['vg']
3800
          elif field == "name":
3801
            val = vol['name']
3802
          elif field == "size":
3803
            val = int(float(vol['size']))
3804
          elif field == "instance":
3805
            for inst in ilist:
3806
              if node not in lv_by_node[inst]:
3807
                continue
3808
              if vol['name'] in lv_by_node[inst][node]:
3809
                val = inst.name
3810
                break
3811
            else:
3812
              val = '-'
3813
          else:
3814
            raise errors.ParameterError(field)
3815
          node_output.append(str(val))
3816

    
3817
        output.append(node_output)
3818

    
3819
    return output
3820

    
3821

    
3822
class LUNodeQueryStorage(NoHooksLU):
3823
  """Logical unit for getting information on storage units on node(s).
3824

3825
  """
3826
  _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
3827
  REQ_BGL = False
3828

    
3829
  def CheckArguments(self):
3830
    _CheckOutputFields(static=self._FIELDS_STATIC,
3831
                       dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
3832
                       selected=self.op.output_fields)
3833

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

    
3838
    if self.op.nodes:
3839
      self.needed_locks[locking.LEVEL_NODE] = \
3840
        _GetWantedNodes(self, self.op.nodes)
3841
    else:
3842
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3843

    
3844
  def Exec(self, feedback_fn):
3845
    """Computes the list of nodes and their attributes.
3846

3847
    """
3848
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
3849

    
3850
    # Always get name to sort by
3851
    if constants.SF_NAME in self.op.output_fields:
3852
      fields = self.op.output_fields[:]
3853
    else:
3854
      fields = [constants.SF_NAME] + self.op.output_fields
3855

    
3856
    # Never ask for node or type as it's only known to the LU
3857
    for extra in [constants.SF_NODE, constants.SF_TYPE]:
3858
      while extra in fields:
3859
        fields.remove(extra)
3860

    
3861
    field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
3862
    name_idx = field_idx[constants.SF_NAME]
3863

    
3864
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
3865
    data = self.rpc.call_storage_list(self.nodes,
3866
                                      self.op.storage_type, st_args,
3867
                                      self.op.name, fields)
3868

    
3869
    result = []
3870

    
3871
    for node in utils.NiceSort(self.nodes):
3872
      nresult = data[node]
3873
      if nresult.offline:
3874
        continue
3875

    
3876
      msg = nresult.fail_msg
3877
      if msg:
3878
        self.LogWarning("Can't get storage data from node %s: %s", node, msg)
3879
        continue
3880

    
3881
      rows = dict([(row[name_idx], row) for row in nresult.payload])
3882

    
3883
      for name in utils.NiceSort(rows.keys()):
3884
        row = rows[name]
3885

    
3886
        out = []
3887

    
3888
        for field in self.op.output_fields:
3889
          if field == constants.SF_NODE:
3890
            val = node
3891
          elif field == constants.SF_TYPE:
3892
            val = self.op.storage_type
3893
          elif field in field_idx:
3894
            val = row[field_idx[field]]
3895
          else:
3896
            raise errors.ParameterError(field)
3897

    
3898
          out.append(val)
3899

    
3900
        result.append(out)
3901

    
3902
    return result
3903

    
3904

    
3905
class _InstanceQuery(_QueryBase):
3906
  FIELDS = query.INSTANCE_FIELDS
3907

    
3908
  def ExpandNames(self, lu):
3909
    lu.needed_locks = {}
3910
    lu.share_locks[locking.LEVEL_INSTANCE] = 1
3911
    lu.share_locks[locking.LEVEL_NODE] = 1
3912

    
3913
    if self.names:
3914
      self.wanted = _GetWantedInstances(lu, self.names)
3915
    else:
3916
      self.wanted = locking.ALL_SET
3917

    
3918
    self.do_locking = (self.use_locking and
3919
                       query.IQ_LIVE in self.requested_data)
3920
    if self.do_locking:
3921
      lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3922
      lu.needed_locks[locking.LEVEL_NODE] = []
3923
      lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3924

    
3925
  def DeclareLocks(self, lu, level):
3926
    if level == locking.LEVEL_NODE and self.do_locking:
3927
      lu._LockInstancesNodes() # pylint: disable-msg=W0212
3928

    
3929
  def _GetQueryData(self, lu):
3930
    """Computes the list of instances and their attributes.
3931

3932
    """
3933
    cluster = lu.cfg.GetClusterInfo()
3934
    all_info = lu.cfg.GetAllInstancesInfo()
3935

    
3936
    instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
3937

    
3938
    instance_list = [all_info[name] for name in instance_names]
3939
    nodes = frozenset(itertools.chain(*(inst.all_nodes
3940
                                        for inst in instance_list)))
3941
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3942
    bad_nodes = []
3943
    offline_nodes = []
3944
    wrongnode_inst = set()
3945

    
3946
    # Gather data as requested
3947
    if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
3948
      live_data = {}
3949
      node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
3950
      for name in nodes:
3951
        result = node_data[name]
3952
        if result.offline:
3953
          # offline nodes will be in both lists
3954
          assert result.fail_msg
3955
          offline_nodes.append(name)
3956
        if result.fail_msg:
3957
          bad_nodes.append(name)
3958
        elif result.payload:
3959
          for inst in result.payload:
3960
            if inst in all_info:
3961
              if all_info[inst].primary_node == name:
3962
                live_data.update(result.payload)
3963
              else:
3964
                wrongnode_inst.add(inst)
3965
            else:
3966
              # orphan instance; we don't list it here as we don't
3967
              # handle this case yet in the output of instance listing
3968
              logging.warning("Orphan instance '%s' found on node %s",
3969
                              inst, name)
3970
        # else no instance is alive
3971
    else:
3972
      live_data = {}
3973

    
3974
    if query.IQ_DISKUSAGE in self.requested_data:
3975
      disk_usage = dict((inst.name,
3976
                         _ComputeDiskSize(inst.disk_template,
3977
                                          [{"size": disk.size}
3978
                                           for disk in inst.disks]))
3979
                        for inst in instance_list)
3980
    else:
3981
      disk_usage = None
3982

    
3983
    if query.IQ_CONSOLE in self.requested_data:
3984
      consinfo = {}
3985
      for inst in instance_list:
3986
        if inst.name in live_data:
3987
          # Instance is running
3988
          consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
3989
        else:
3990
          consinfo[inst.name] = None
3991
      assert set(consinfo.keys()) == set(instance_names)
3992
    else:
3993
      consinfo = None
3994

    
3995
    return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
3996
                                   disk_usage, offline_nodes, bad_nodes,
3997
                                   live_data, wrongnode_inst, consinfo)
3998

    
3999

    
4000
class LUQuery(NoHooksLU):
4001
  """Query for resources/items of a certain kind.
4002

4003
  """
4004
  # pylint: disable-msg=W0142
4005
  REQ_BGL = False
4006

    
4007
  def CheckArguments(self):
4008
    qcls = _GetQueryImplementation(self.op.what)
4009
    names = qlang.ReadSimpleFilter("name", self.op.filter)
4010

    
4011
    self.impl = qcls(names, self.op.fields, False)
4012

    
4013
  def ExpandNames(self):
4014
    self.impl.ExpandNames(self)
4015

    
4016
  def DeclareLocks(self, level):
4017
    self.impl.DeclareLocks(self, level)
4018

    
4019
  def Exec(self, feedback_fn):
4020
    return self.impl.NewStyleQuery(self)
4021

    
4022

    
4023
class LUQueryFields(NoHooksLU):
4024
  """Query for resources/items of a certain kind.
4025

4026
  """
4027
  # pylint: disable-msg=W0142
4028
  REQ_BGL = False
4029

    
4030
  def CheckArguments(self):
4031
    self.qcls = _GetQueryImplementation(self.op.what)
4032

    
4033
  def ExpandNames(self):
4034
    self.needed_locks = {}
4035

    
4036
  def Exec(self, feedback_fn):
4037
    return self.qcls.FieldsQuery(self.op.fields)
4038

    
4039

    
4040
class LUNodeModifyStorage(NoHooksLU):
4041
  """Logical unit for modifying a storage volume on a node.
4042

4043
  """
4044
  REQ_BGL = False
4045

    
4046
  def CheckArguments(self):
4047
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4048

    
4049
    storage_type = self.op.storage_type
4050

    
4051
    try:
4052
      modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
4053
    except KeyError:
4054
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
4055
                                 " modified" % storage_type,
4056
                                 errors.ECODE_INVAL)
4057

    
4058
    diff = set(self.op.changes.keys()) - modifiable
4059
    if diff:
4060
      raise errors.OpPrereqError("The following fields can not be modified for"
4061
                                 " storage units of type '%s': %r" %
4062
                                 (storage_type, list(diff)),
4063
                                 errors.ECODE_INVAL)
4064

    
4065
  def ExpandNames(self):
4066
    self.needed_locks = {
4067
      locking.LEVEL_NODE: self.op.node_name,
4068
      }
4069

    
4070
  def Exec(self, feedback_fn):
4071
    """Computes the list of nodes and their attributes.
4072

4073
    """
4074
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4075
    result = self.rpc.call_storage_modify(self.op.node_name,
4076
                                          self.op.storage_type, st_args,
4077
                                          self.op.name, self.op.changes)
4078
    result.Raise("Failed to modify storage unit '%s' on %s" %
4079
                 (self.op.name, self.op.node_name))
4080

    
4081

    
4082
class LUNodeAdd(LogicalUnit):
4083