Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 9711e69d

History | View | Annotate | Download (413.6 kB)

1
#
2
#
3

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

    
21

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

    
24
# pylint: disable-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(), cfg.GetSharedFileStorageDir()]]
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 time.
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 _VerifyNodeNetwork(self, ninfo, nresult):
1484
    """Check the node time.
1485

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

1490
    """
1491
    node = ninfo.name
1492
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1493

    
1494
    test = constants.NV_NODELIST not in nresult
1495
    _ErrorIf(test, self.ENODESSH, node,
1496
             "node hasn't returned node ssh connectivity data")
1497
    if not test:
1498
      if nresult[constants.NV_NODELIST]:
1499
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1500
          _ErrorIf(True, self.ENODESSH, node,
1501
                   "ssh communication with node '%s': %s", a_node, a_msg)
1502

    
1503
    test = constants.NV_NODENETTEST not in nresult
1504
    _ErrorIf(test, self.ENODENET, node,
1505
             "node hasn't returned node tcp connectivity data")
1506
    if not test:
1507
      if nresult[constants.NV_NODENETTEST]:
1508
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1509
        for anode in nlist:
1510
          _ErrorIf(True, self.ENODENET, node,
1511
                   "tcp communication with node '%s': %s",
1512
                   anode, nresult[constants.NV_NODENETTEST][anode])
1513

    
1514
    test = constants.NV_MASTERIP not in nresult
1515
    _ErrorIf(test, self.ENODENET, node,
1516
             "node hasn't returned node master IP reachability data")
1517
    if not test:
1518
      if not nresult[constants.NV_MASTERIP]:
1519
        if node == self.master_node:
1520
          msg = "the master node cannot reach the master IP (not configured?)"
1521
        else:
1522
          msg = "cannot reach the master IP"
1523
        _ErrorIf(True, self.ENODENET, node, msg)
1524

    
1525
  def _VerifyInstance(self, instance, instanceconfig, node_image,
1526
                      diskstatus):
1527
    """Verify an instance.
1528

1529
    This function checks to see if the required block devices are
1530
    available on the instance's node.
1531

1532
    """
1533
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1534
    node_current = instanceconfig.primary_node
1535

    
1536
    node_vol_should = {}
1537
    instanceconfig.MapLVsByNode(node_vol_should)
1538

    
1539
    for node in node_vol_should:
1540
      n_img = node_image[node]
1541
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1542
        # ignore missing volumes on offline or broken nodes
1543
        continue
1544
      for volume in node_vol_should[node]:
1545
        test = volume not in n_img.volumes
1546
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
1547
                 "volume %s missing on node %s", volume, node)
1548

    
1549
    if instanceconfig.admin_up:
1550
      pri_img = node_image[node_current]
1551
      test = instance not in pri_img.instances and not pri_img.offline
1552
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
1553
               "instance not running on its primary node %s",
1554
               node_current)
1555

    
1556
    for node, n_img in node_image.items():
1557
      if node != node_current:
1558
        test = instance in n_img.instances
1559
        _ErrorIf(test, self.EINSTANCEWRONGNODE, instance,
1560
                 "instance should not run on node %s", node)
1561

    
1562
    diskdata = [(nname, success, status, idx)
1563
                for (nname, disks) in diskstatus.items()
1564
                for idx, (success, status) in enumerate(disks)]
1565

    
1566
    for nname, success, bdev_status, idx in diskdata:
1567
      # the 'ghost node' construction in Exec() ensures that we have a
1568
      # node here
1569
      snode = node_image[nname]
1570
      bad_snode = snode.ghost or snode.offline
1571
      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
1572
               self.EINSTANCEFAULTYDISK, instance,
1573
               "couldn't retrieve status for disk/%s on %s: %s",
1574
               idx, nname, bdev_status)
1575
      _ErrorIf((instanceconfig.admin_up and success and
1576
                bdev_status.ldisk_status == constants.LDS_FAULTY),
1577
               self.EINSTANCEFAULTYDISK, instance,
1578
               "disk/%s on %s is faulty", idx, nname)
1579

    
1580
  def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
1581
    """Verify if there are any unknown volumes in the cluster.
1582

1583
    The .os, .swap and backup volumes are ignored. All other volumes are
1584
    reported as unknown.
1585

1586
    @type reserved: L{ganeti.utils.FieldSet}
1587
    @param reserved: a FieldSet of reserved volume names
1588

1589
    """
1590
    for node, n_img in node_image.items():
1591
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1592
        # skip non-healthy nodes
1593
        continue
1594
      for volume in n_img.volumes:
1595
        test = ((node not in node_vol_should or
1596
                volume not in node_vol_should[node]) and
1597
                not reserved.Matches(volume))
1598
        self._ErrorIf(test, self.ENODEORPHANLV, node,
1599
                      "volume %s is unknown", volume)
1600

    
1601
  def _VerifyOrphanInstances(self, instancelist, node_image):
1602
    """Verify the list of running instances.
1603

1604
    This checks what instances are running but unknown to the cluster.
1605

1606
    """
1607
    for node, n_img in node_image.items():
1608
      for o_inst in n_img.instances:
1609
        test = o_inst not in instancelist
1610
        self._ErrorIf(test, self.ENODEORPHANINSTANCE, node,
1611
                      "instance %s on node %s should not exist", o_inst, node)
1612

    
1613
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
1614
    """Verify N+1 Memory Resilience.
1615

1616
    Check that if one single node dies we can still start all the
1617
    instances it was primary for.
1618

1619
    """
1620
    for node, n_img in node_image.items():
1621
      # This code checks that every node which is now listed as
1622
      # secondary has enough memory to host all instances it is
1623
      # supposed to should a single other node in the cluster fail.
1624
      # FIXME: not ready for failover to an arbitrary node
1625
      # FIXME: does not support file-backed instances
1626
      # WARNING: we currently take into account down instances as well
1627
      # as up ones, considering that even if they're down someone
1628
      # might want to start them even in the event of a node failure.
1629
      if n_img.offline:
1630
        # we're skipping offline nodes from the N+1 warning, since
1631
        # most likely we don't have good memory infromation from them;
1632
        # we already list instances living on such nodes, and that's
1633
        # enough warning
1634
        continue
1635
      for prinode, instances in n_img.sbp.items():
1636
        needed_mem = 0
1637
        for instance in instances:
1638
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
1639
          if bep[constants.BE_AUTO_BALANCE]:
1640
            needed_mem += bep[constants.BE_MEMORY]
1641
        test = n_img.mfree < needed_mem
1642
        self._ErrorIf(test, self.ENODEN1, node,
1643
                      "not enough memory to accomodate instance failovers"
1644
                      " should node %s fail", prinode)
1645

    
1646
  def _VerifyNodeFiles(self, ninfo, nresult, file_list, local_cksum,
1647
                       master_files):
1648
    """Verifies and computes the node required file checksums.
1649

1650
    @type ninfo: L{objects.Node}
1651
    @param ninfo: the node to check
1652
    @param nresult: the remote results for the node
1653
    @param file_list: required list of files
1654
    @param local_cksum: dictionary of local files and their checksums
1655
    @param master_files: list of files that only masters should have
1656

1657
    """
1658
    node = ninfo.name
1659
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1660

    
1661
    remote_cksum = nresult.get(constants.NV_FILELIST, None)
1662
    test = not isinstance(remote_cksum, dict)
1663
    _ErrorIf(test, self.ENODEFILECHECK, node,
1664
             "node hasn't returned file checksum data")
1665
    if test:
1666
      return
1667

    
1668
    for file_name in file_list:
1669
      node_is_mc = ninfo.master_candidate
1670
      must_have = (file_name not in master_files) or node_is_mc
1671
      # missing
1672
      test1 = file_name not in remote_cksum
1673
      # invalid checksum
1674
      test2 = not test1 and remote_cksum[file_name] != local_cksum[file_name]
1675
      # existing and good
1676
      test3 = not test1 and remote_cksum[file_name] == local_cksum[file_name]
1677
      _ErrorIf(test1 and must_have, self.ENODEFILECHECK, node,
1678
               "file '%s' missing", file_name)
1679
      _ErrorIf(test2 and must_have, self.ENODEFILECHECK, node,
1680
               "file '%s' has wrong checksum", file_name)
1681
      # not candidate and this is not a must-have file
1682
      _ErrorIf(test2 and not must_have, self.ENODEFILECHECK, node,
1683
               "file '%s' should not exist on non master"
1684
               " candidates (and the file is outdated)", file_name)
1685
      # all good, except non-master/non-must have combination
1686
      _ErrorIf(test3 and not must_have, self.ENODEFILECHECK, node,
1687
               "file '%s' should not exist"
1688
               " on non master candidates", file_name)
1689

    
1690
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_helper,
1691
                      drbd_map):
1692
    """Verifies and the node DRBD status.
1693

1694
    @type ninfo: L{objects.Node}
1695
    @param ninfo: the node to check
1696
    @param nresult: the remote results for the node
1697
    @param instanceinfo: the dict of instances
1698
    @param drbd_helper: the configured DRBD usermode helper
1699
    @param drbd_map: the DRBD map as returned by
1700
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
1701

1702
    """
1703
    node = ninfo.name
1704
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1705

    
1706
    if drbd_helper:
1707
      helper_result = nresult.get(constants.NV_DRBDHELPER, None)
1708
      test = (helper_result == None)
1709
      _ErrorIf(test, self.ENODEDRBDHELPER, node,
1710
               "no drbd usermode helper returned")
1711
      if helper_result:
1712
        status, payload = helper_result
1713
        test = not status
1714
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
1715
                 "drbd usermode helper check unsuccessful: %s", payload)
1716
        test = status and (payload != drbd_helper)
1717
        _ErrorIf(test, self.ENODEDRBDHELPER, node,
1718
                 "wrong drbd usermode helper: %s", payload)
1719

    
1720
    # compute the DRBD minors
1721
    node_drbd = {}
1722
    for minor, instance in drbd_map[node].items():
1723
      test = instance not in instanceinfo
1724
      _ErrorIf(test, self.ECLUSTERCFG, None,
1725
               "ghost instance '%s' in temporary DRBD map", instance)
1726
        # ghost instance should not be running, but otherwise we
1727
        # don't give double warnings (both ghost instance and
1728
        # unallocated minor in use)
1729
      if test:
1730
        node_drbd[minor] = (instance, False)
1731
      else:
1732
        instance = instanceinfo[instance]
1733
        node_drbd[minor] = (instance.name, instance.admin_up)
1734

    
1735
    # and now check them
1736
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
1737
    test = not isinstance(used_minors, (tuple, list))
1738
    _ErrorIf(test, self.ENODEDRBD, node,
1739
             "cannot parse drbd status file: %s", str(used_minors))
1740
    if test:
1741
      # we cannot check drbd status
1742
      return
1743

    
1744
    for minor, (iname, must_exist) in node_drbd.items():
1745
      test = minor not in used_minors and must_exist
1746
      _ErrorIf(test, self.ENODEDRBD, node,
1747
               "drbd minor %d of instance %s is not active", minor, iname)
1748
    for minor in used_minors:
1749
      test = minor not in node_drbd
1750
      _ErrorIf(test, self.ENODEDRBD, node,
1751
               "unallocated drbd minor %d is in use", minor)
1752

    
1753
  def _UpdateNodeOS(self, ninfo, nresult, nimg):
1754
    """Builds the node OS structures.
1755

1756
    @type ninfo: L{objects.Node}
1757
    @param ninfo: the node to check
1758
    @param nresult: the remote results for the node
1759
    @param nimg: the node image object
1760

1761
    """
1762
    node = ninfo.name
1763
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1764

    
1765
    remote_os = nresult.get(constants.NV_OSLIST, None)
1766
    test = (not isinstance(remote_os, list) or
1767
            not compat.all(isinstance(v, list) and len(v) == 7
1768
                           for v in remote_os))
1769

    
1770
    _ErrorIf(test, self.ENODEOS, node,
1771
             "node hasn't returned valid OS data")
1772

    
1773
    nimg.os_fail = test
1774

    
1775
    if test:
1776
      return
1777

    
1778
    os_dict = {}
1779

    
1780
    for (name, os_path, status, diagnose,
1781
         variants, parameters, api_ver) in nresult[constants.NV_OSLIST]:
1782

    
1783
      if name not in os_dict:
1784
        os_dict[name] = []
1785

    
1786
      # parameters is a list of lists instead of list of tuples due to
1787
      # JSON lacking a real tuple type, fix it:
1788
      parameters = [tuple(v) for v in parameters]
1789
      os_dict[name].append((os_path, status, diagnose,
1790
                            set(variants), set(parameters), set(api_ver)))
1791

    
1792
    nimg.oslist = os_dict
1793

    
1794
  def _VerifyNodeOS(self, ninfo, nimg, base):
1795
    """Verifies the node OS list.
1796

1797
    @type ninfo: L{objects.Node}
1798
    @param ninfo: the node to check
1799
    @param nimg: the node image object
1800
    @param base: the 'template' node we match against (e.g. from the master)
1801

1802
    """
1803
    node = ninfo.name
1804
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1805

    
1806
    assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
1807

    
1808
    for os_name, os_data in nimg.oslist.items():
1809
      assert os_data, "Empty OS status for OS %s?!" % os_name
1810
      f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
1811
      _ErrorIf(not f_status, self.ENODEOS, node,
1812
               "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
1813
      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
1814
               "OS '%s' has multiple entries (first one shadows the rest): %s",
1815
               os_name, utils.CommaJoin([v[0] for v in os_data]))
1816
      # this will catched in backend too
1817
      _ErrorIf(compat.any(v >= constants.OS_API_V15 for v in f_api)
1818
               and not f_var, self.ENODEOS, node,
1819
               "OS %s with API at least %d does not declare any variant",
1820
               os_name, constants.OS_API_V15)
1821
      # comparisons with the 'base' image
1822
      test = os_name not in base.oslist
1823
      _ErrorIf(test, self.ENODEOS, node,
1824
               "Extra OS %s not present on reference node (%s)",
1825
               os_name, base.name)
1826
      if test:
1827
        continue
1828
      assert base.oslist[os_name], "Base node has empty OS status?"
1829
      _, b_status, _, b_var, b_param, b_api = base.oslist[os_name][0]
1830
      if not b_status:
1831
        # base OS is invalid, skipping
1832
        continue
1833
      for kind, a, b in [("API version", f_api, b_api),
1834
                         ("variants list", f_var, b_var),
1835
                         ("parameters", f_param, b_param)]:
1836
        _ErrorIf(a != b, self.ENODEOS, node,
1837
                 "OS %s %s differs from reference node %s: %s vs. %s",
1838
                 kind, os_name, base.name,
1839
                 utils.CommaJoin(a), utils.CommaJoin(b))
1840

    
1841
    # check any missing OSes
1842
    missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
1843
    _ErrorIf(missing, self.ENODEOS, node,
1844
             "OSes present on reference node %s but missing on this node: %s",
1845
             base.name, utils.CommaJoin(missing))
1846

    
1847
  def _VerifyOob(self, ninfo, nresult):
1848
    """Verifies out of band functionality of a node.
1849

1850
    @type ninfo: L{objects.Node}
1851
    @param ninfo: the node to check
1852
    @param nresult: the remote results for the node
1853

1854
    """
1855
    node = ninfo.name
1856
    # We just have to verify the paths on master and/or master candidates
1857
    # as the oob helper is invoked on the master
1858
    if ((ninfo.master_candidate or ninfo.master_capable) and
1859
        constants.NV_OOB_PATHS in nresult):
1860
      for path_result in nresult[constants.NV_OOB_PATHS]:
1861
        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
1862

    
1863
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
1864
    """Verifies and updates the node volume data.
1865

1866
    This function will update a L{NodeImage}'s internal structures
1867
    with data from the remote call.
1868

1869
    @type ninfo: L{objects.Node}
1870
    @param ninfo: the node to check
1871
    @param nresult: the remote results for the node
1872
    @param nimg: the node image object
1873
    @param vg_name: the configured VG name
1874

1875
    """
1876
    node = ninfo.name
1877
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1878

    
1879
    nimg.lvm_fail = True
1880
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
1881
    if vg_name is None:
1882
      pass
1883
    elif isinstance(lvdata, basestring):
1884
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
1885
               utils.SafeEncode(lvdata))
1886
    elif not isinstance(lvdata, dict):
1887
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
1888
    else:
1889
      nimg.volumes = lvdata
1890
      nimg.lvm_fail = False
1891

    
1892
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
1893
    """Verifies and updates the node instance list.
1894

1895
    If the listing was successful, then updates this node's instance
1896
    list. Otherwise, it marks the RPC call as failed for the instance
1897
    list key.
1898

1899
    @type ninfo: L{objects.Node}
1900
    @param ninfo: the node to check
1901
    @param nresult: the remote results for the node
1902
    @param nimg: the node image object
1903

1904
    """
1905
    idata = nresult.get(constants.NV_INSTANCELIST, None)
1906
    test = not isinstance(idata, list)
1907
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
1908
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
1909
    if test:
1910
      nimg.hyp_fail = True
1911
    else:
1912
      nimg.instances = idata
1913

    
1914
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
1915
    """Verifies and computes a node information map
1916

1917
    @type ninfo: L{objects.Node}
1918
    @param ninfo: the node to check
1919
    @param nresult: the remote results for the node
1920
    @param nimg: the node image object
1921
    @param vg_name: the configured VG name
1922

1923
    """
1924
    node = ninfo.name
1925
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1926

    
1927
    # try to read free memory (from the hypervisor)
1928
    hv_info = nresult.get(constants.NV_HVINFO, None)
1929
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
1930
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
1931
    if not test:
1932
      try:
1933
        nimg.mfree = int(hv_info["memory_free"])
1934
      except (ValueError, TypeError):
1935
        _ErrorIf(True, self.ENODERPC, node,
1936
                 "node returned invalid nodeinfo, check hypervisor")
1937

    
1938
    # FIXME: devise a free space model for file based instances as well
1939
    if vg_name is not None:
1940
      test = (constants.NV_VGLIST not in nresult or
1941
              vg_name not in nresult[constants.NV_VGLIST])
1942
      _ErrorIf(test, self.ENODELVM, node,
1943
               "node didn't return data for the volume group '%s'"
1944
               " - it is either missing or broken", vg_name)
1945
      if not test:
1946
        try:
1947
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
1948
        except (ValueError, TypeError):
1949
          _ErrorIf(True, self.ENODERPC, node,
1950
                   "node returned invalid LVM info, check LVM status")
1951

    
1952
  def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
1953
    """Gets per-disk status information for all instances.
1954

1955
    @type nodelist: list of strings
1956
    @param nodelist: Node names
1957
    @type node_image: dict of (name, L{objects.Node})
1958
    @param node_image: Node objects
1959
    @type instanceinfo: dict of (name, L{objects.Instance})
1960
    @param instanceinfo: Instance objects
1961
    @rtype: {instance: {node: [(succes, payload)]}}
1962
    @return: a dictionary of per-instance dictionaries with nodes as
1963
        keys and disk information as values; the disk information is a
1964
        list of tuples (success, payload)
1965

1966
    """
1967
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1968

    
1969
    node_disks = {}
1970
    node_disks_devonly = {}
1971
    diskless_instances = set()
1972
    diskless = constants.DT_DISKLESS
1973

    
1974
    for nname in nodelist:
1975
      node_instances = list(itertools.chain(node_image[nname].pinst,
1976
                                            node_image[nname].sinst))
1977
      diskless_instances.update(inst for inst in node_instances
1978
                                if instanceinfo[inst].disk_template == diskless)
1979
      disks = [(inst, disk)
1980
               for inst in node_instances
1981
               for disk in instanceinfo[inst].disks]
1982

    
1983
      if not disks:
1984
        # No need to collect data
1985
        continue
1986

    
1987
      node_disks[nname] = disks
1988

    
1989
      # Creating copies as SetDiskID below will modify the objects and that can
1990
      # lead to incorrect data returned from nodes
1991
      devonly = [dev.Copy() for (_, dev) in disks]
1992

    
1993
      for dev in devonly:
1994
        self.cfg.SetDiskID(dev, nname)
1995

    
1996
      node_disks_devonly[nname] = devonly
1997

    
1998
    assert len(node_disks) == len(node_disks_devonly)
1999

    
2000
    # Collect data from all nodes with disks
2001
    result = self.rpc.call_blockdev_getmirrorstatus_multi(node_disks.keys(),
2002
                                                          node_disks_devonly)
2003

    
2004
    assert len(result) == len(node_disks)
2005

    
2006
    instdisk = {}
2007

    
2008
    for (nname, nres) in result.items():
2009
      disks = node_disks[nname]
2010

    
2011
      if nres.offline:
2012
        # No data from this node
2013
        data = len(disks) * [(False, "node offline")]
2014
      else:
2015
        msg = nres.fail_msg
2016
        _ErrorIf(msg, self.ENODERPC, nname,
2017
                 "while getting disk information: %s", msg)
2018
        if msg:
2019
          # No data from this node
2020
          data = len(disks) * [(False, msg)]
2021
        else:
2022
          data = []
2023
          for idx, i in enumerate(nres.payload):
2024
            if isinstance(i, (tuple, list)) and len(i) == 2:
2025
              data.append(i)
2026
            else:
2027
              logging.warning("Invalid result from node %s, entry %d: %s",
2028
                              nname, idx, i)
2029
              data.append((False, "Invalid result from the remote node"))
2030

    
2031
      for ((inst, _), status) in zip(disks, data):
2032
        instdisk.setdefault(inst, {}).setdefault(nname, []).append(status)
2033

    
2034
    # Add empty entries for diskless instances.
2035
    for inst in diskless_instances:
2036
      assert inst not in instdisk
2037
      instdisk[inst] = {}
2038

    
2039
    assert compat.all(len(statuses) == len(instanceinfo[inst].disks) and
2040
                      len(nnames) <= len(instanceinfo[inst].all_nodes) and
2041
                      compat.all(isinstance(s, (tuple, list)) and
2042
                                 len(s) == 2 for s in statuses)
2043
                      for inst, nnames in instdisk.items()
2044
                      for nname, statuses in nnames.items())
2045
    assert set(instdisk) == set(instanceinfo), "instdisk consistency failure"
2046

    
2047
    return instdisk
2048

    
2049
  def _VerifyHVP(self, hvp_data):
2050
    """Verifies locally the syntax of the hypervisor parameters.
2051

2052
    """
2053
    for item, hv_name, hv_params in hvp_data:
2054
      msg = ("hypervisor %s parameters syntax check (source %s): %%s" %
2055
             (item, hv_name))
2056
      try:
2057
        hv_class = hypervisor.GetHypervisor(hv_name)
2058
        utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
2059
        hv_class.CheckParameterSyntax(hv_params)
2060
      except errors.GenericError, err:
2061
        self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
2062

    
2063

    
2064
  def BuildHooksEnv(self):
2065
    """Build hooks env.
2066

2067
    Cluster-Verify hooks just ran in the post phase and their failure makes
2068
    the output be logged in the verify output and the verification to fail.
2069

2070
    """
2071
    all_nodes = self.cfg.GetNodeList()
2072
    env = {
2073
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
2074
      }
2075
    for node in self.cfg.GetAllNodesInfo().values():
2076
      env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())
2077

    
2078
    return env, [], all_nodes
2079

    
2080
  def Exec(self, feedback_fn):
2081
    """Verify integrity of cluster, performing various test on nodes.
2082

2083
    """
2084
    # This method has too many local variables. pylint: disable-msg=R0914
2085
    self.bad = False
2086
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
2087
    verbose = self.op.verbose
2088
    self._feedback_fn = feedback_fn
2089
    feedback_fn("* Verifying global settings")
2090
    for msg in self.cfg.VerifyConfig():
2091
      _ErrorIf(True, self.ECLUSTERCFG, None, msg)
2092

    
2093
    # Check the cluster certificates
2094
    for cert_filename in constants.ALL_CERT_FILES:
2095
      (errcode, msg) = _VerifyCertificate(cert_filename)
2096
      _ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
2097

    
2098
    vg_name = self.cfg.GetVGName()
2099
    drbd_helper = self.cfg.GetDRBDHelper()
2100
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
2101
    cluster = self.cfg.GetClusterInfo()
2102
    nodelist = utils.NiceSort(self.cfg.GetNodeList())
2103
    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
2104
    nodeinfo_byname = dict(zip(nodelist, nodeinfo))
2105
    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
2106
    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
2107
                        for iname in instancelist)
2108
    groupinfo = self.cfg.GetAllNodeGroupsInfo()
2109
    i_non_redundant = [] # Non redundant instances
2110
    i_non_a_balanced = [] # Non auto-balanced instances
2111
    n_offline = 0 # Count of offline nodes
2112
    n_drained = 0 # Count of nodes being drained
2113
    node_vol_should = {}
2114

    
2115
    # FIXME: verify OS list
2116
    # do local checksums
2117
    master_files = [constants.CLUSTER_CONF_FILE]
2118
    master_node = self.master_node = self.cfg.GetMasterNode()
2119
    master_ip = self.cfg.GetMasterIP()
2120

    
2121
    file_names = ssconf.SimpleStore().GetFileList()
2122
    file_names.extend(constants.ALL_CERT_FILES)
2123
    file_names.extend(master_files)
2124
    if cluster.modify_etc_hosts:
2125
      file_names.append(constants.ETC_HOSTS)
2126

    
2127
    local_checksums = utils.FingerprintFiles(file_names)
2128

    
2129
    # Compute the set of hypervisor parameters
2130
    hvp_data = []
2131
    for hv_name in hypervisors:
2132
      hvp_data.append(("cluster", hv_name, cluster.GetHVDefaults(hv_name)))
2133
    for os_name, os_hvp in cluster.os_hvp.items():
2134
      for hv_name, hv_params in os_hvp.items():
2135
        if not hv_params:
2136
          continue
2137
        full_params = cluster.GetHVDefaults(hv_name, os_name=os_name)
2138
        hvp_data.append(("os %s" % os_name, hv_name, full_params))
2139
    # TODO: collapse identical parameter values in a single one
2140
    for instance in instanceinfo.values():
2141
      if not instance.hvparams:
2142
        continue
2143
      hvp_data.append(("instance %s" % instance.name, instance.hypervisor,
2144
                       cluster.FillHV(instance)))
2145
    # and verify them locally
2146
    self._VerifyHVP(hvp_data)
2147

    
2148
    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
2149
    node_verify_param = {
2150
      constants.NV_FILELIST: file_names,
2151
      constants.NV_NODELIST: [node.name for node in nodeinfo
2152
                              if not node.offline],
2153
      constants.NV_HYPERVISOR: hypervisors,
2154
      constants.NV_HVPARAMS: hvp_data,
2155
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
2156
                                  node.secondary_ip) for node in nodeinfo
2157
                                 if not node.offline],
2158
      constants.NV_INSTANCELIST: hypervisors,
2159
      constants.NV_VERSION: None,
2160
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
2161
      constants.NV_NODESETUP: None,
2162
      constants.NV_TIME: None,
2163
      constants.NV_MASTERIP: (master_node, master_ip),
2164
      constants.NV_OSLIST: None,
2165
      constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
2166
      }
2167

    
2168
    if vg_name is not None:
2169
      node_verify_param[constants.NV_VGLIST] = None
2170
      node_verify_param[constants.NV_LVLIST] = vg_name
2171
      node_verify_param[constants.NV_PVLIST] = [vg_name]
2172
      node_verify_param[constants.NV_DRBDLIST] = None
2173

    
2174
    if drbd_helper:
2175
      node_verify_param[constants.NV_DRBDHELPER] = drbd_helper
2176

    
2177
    # Build our expected cluster state
2178
    node_image = dict((node.name, self.NodeImage(offline=node.offline,
2179
                                                 name=node.name,
2180
                                                 vm_capable=node.vm_capable))
2181
                      for node in nodeinfo)
2182

    
2183
    # Gather OOB paths
2184
    oob_paths = []
2185
    for node in nodeinfo:
2186
      path = _SupportsOob(self.cfg, node)
2187
      if path and path not in oob_paths:
2188
        oob_paths.append(path)
2189

    
2190
    if oob_paths:
2191
      node_verify_param[constants.NV_OOB_PATHS] = oob_paths
2192

    
2193
    for instance in instancelist:
2194
      inst_config = instanceinfo[instance]
2195

    
2196
      for nname in inst_config.all_nodes:
2197
        if nname not in node_image:
2198
          # ghost node
2199
          gnode = self.NodeImage(name=nname)
2200
          gnode.ghost = True
2201
          node_image[nname] = gnode
2202

    
2203
      inst_config.MapLVsByNode(node_vol_should)
2204

    
2205
      pnode = inst_config.primary_node
2206
      node_image[pnode].pinst.append(instance)
2207

    
2208
      for snode in inst_config.secondary_nodes:
2209
        nimg = node_image[snode]
2210
        nimg.sinst.append(instance)
2211
        if pnode not in nimg.sbp:
2212
          nimg.sbp[pnode] = []
2213
        nimg.sbp[pnode].append(instance)
2214

    
2215
    # At this point, we have the in-memory data structures complete,
2216
    # except for the runtime information, which we'll gather next
2217

    
2218
    # Due to the way our RPC system works, exact response times cannot be
2219
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
2220
    # time before and after executing the request, we can at least have a time
2221
    # window.
2222
    nvinfo_starttime = time.time()
2223
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
2224
                                           self.cfg.GetClusterName())
2225
    nvinfo_endtime = time.time()
2226

    
2227
    all_drbd_map = self.cfg.ComputeDRBDMap()
2228

    
2229
    feedback_fn("* Gathering disk information (%s nodes)" % len(nodelist))
2230
    instdisk = self._CollectDiskInfo(nodelist, node_image, instanceinfo)
2231

    
2232
    feedback_fn("* Verifying node status")
2233

    
2234
    refos_img = None
2235

    
2236
    for node_i in nodeinfo:
2237
      node = node_i.name
2238
      nimg = node_image[node]
2239

    
2240
      if node_i.offline:
2241
        if verbose:
2242
          feedback_fn("* Skipping offline node %s" % (node,))
2243
        n_offline += 1
2244
        continue
2245

    
2246
      if node == master_node:
2247
        ntype = "master"
2248
      elif node_i.master_candidate:
2249
        ntype = "master candidate"
2250
      elif node_i.drained:
2251
        ntype = "drained"
2252
        n_drained += 1
2253
      else:
2254
        ntype = "regular"
2255
      if verbose:
2256
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
2257

    
2258
      msg = all_nvinfo[node].fail_msg
2259
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
2260
      if msg:
2261
        nimg.rpc_fail = True
2262
        continue
2263

    
2264
      nresult = all_nvinfo[node].payload
2265

    
2266
      nimg.call_ok = self._VerifyNode(node_i, nresult)
2267
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
2268
      self._VerifyNodeNetwork(node_i, nresult)
2269
      self._VerifyNodeFiles(node_i, nresult, file_names, local_checksums,
2270
                            master_files)
2271

    
2272
      self._VerifyOob(node_i, nresult)
2273

    
2274
      if nimg.vm_capable:
2275
        self._VerifyNodeLVM(node_i, nresult, vg_name)
2276
        self._VerifyNodeDrbd(node_i, nresult, instanceinfo, drbd_helper,
2277
                             all_drbd_map)
2278

    
2279
        self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
2280
        self._UpdateNodeInstances(node_i, nresult, nimg)
2281
        self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
2282
        self._UpdateNodeOS(node_i, nresult, nimg)
2283
        if not nimg.os_fail:
2284
          if refos_img is None:
2285
            refos_img = nimg
2286
          self._VerifyNodeOS(node_i, nimg, refos_img)
2287

    
2288
    feedback_fn("* Verifying instance status")
2289
    for instance in instancelist:
2290
      if verbose:
2291
        feedback_fn("* Verifying instance %s" % instance)
2292
      inst_config = instanceinfo[instance]
2293
      self._VerifyInstance(instance, inst_config, node_image,
2294
                           instdisk[instance])
2295
      inst_nodes_offline = []
2296

    
2297
      pnode = inst_config.primary_node
2298
      pnode_img = node_image[pnode]
2299
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
2300
               self.ENODERPC, pnode, "instance %s, connection to"
2301
               " primary node failed", instance)
2302

    
2303
      _ErrorIf(pnode_img.offline, self.EINSTANCEBADNODE, instance,
2304
               "instance lives on offline node %s", inst_config.primary_node)
2305

    
2306
      # If the instance is non-redundant we cannot survive losing its primary
2307
      # node, so we are not N+1 compliant. On the other hand we have no disk
2308
      # templates with more than one secondary so that situation is not well
2309
      # supported either.
2310
      # FIXME: does not support file-backed instances
2311
      if not inst_config.secondary_nodes:
2312
        i_non_redundant.append(instance)
2313

    
2314
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
2315
               instance, "instance has multiple secondary nodes: %s",
2316
               utils.CommaJoin(inst_config.secondary_nodes),
2317
               code=self.ETYPE_WARNING)
2318

    
2319
      if inst_config.disk_template in constants.DTS_INT_MIRROR:
2320
        pnode = inst_config.primary_node
2321
        instance_nodes = utils.NiceSort(inst_config.all_nodes)
2322
        instance_groups = {}
2323

    
2324
        for node in instance_nodes:
2325
          instance_groups.setdefault(nodeinfo_byname[node].group,
2326
                                     []).append(node)
2327

    
2328
        pretty_list = [
2329
          "%s (group %s)" % (utils.CommaJoin(nodes), groupinfo[group].name)
2330
          # Sort so that we always list the primary node first.
2331
          for group, nodes in sorted(instance_groups.items(),
2332
                                     key=lambda (_, nodes): pnode in nodes,
2333
                                     reverse=True)]
2334

    
2335
        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
2336
                      instance, "instance has primary and secondary nodes in"
2337
                      " different groups: %s", utils.CommaJoin(pretty_list),
2338
                      code=self.ETYPE_WARNING)
2339

    
2340
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
2341
        i_non_a_balanced.append(instance)
2342

    
2343
      for snode in inst_config.secondary_nodes:
2344
        s_img = node_image[snode]
2345
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
2346
                 "instance %s, connection to secondary node failed", instance)
2347

    
2348
        if s_img.offline:
2349
          inst_nodes_offline.append(snode)
2350

    
2351
      # warn that the instance lives on offline nodes
2352
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
2353
               "instance has offline secondary node(s) %s",
2354
               utils.CommaJoin(inst_nodes_offline))
2355
      # ... or ghost/non-vm_capable nodes
2356
      for node in inst_config.all_nodes:
2357
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
2358
                 "instance lives on ghost node %s", node)
2359
        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
2360
                 instance, "instance lives on non-vm_capable node %s", node)
2361

    
2362
    feedback_fn("* Verifying orphan volumes")
2363
    reserved = utils.FieldSet(*cluster.reserved_lvs)
2364
    self._VerifyOrphanVolumes(node_vol_should, node_image, reserved)
2365

    
2366
    feedback_fn("* Verifying orphan instances")
2367
    self._VerifyOrphanInstances(instancelist, node_image)
2368

    
2369
    if constants.VERIFY_NPLUSONE_MEM not in self.op.skip_checks:
2370
      feedback_fn("* Verifying N+1 Memory redundancy")
2371
      self._VerifyNPlusOneMemory(node_image, instanceinfo)
2372

    
2373
    feedback_fn("* Other Notes")
2374
    if i_non_redundant:
2375
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
2376
                  % len(i_non_redundant))
2377

    
2378
    if i_non_a_balanced:
2379
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
2380
                  % len(i_non_a_balanced))
2381

    
2382
    if n_offline:
2383
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
2384

    
2385
    if n_drained:
2386
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
2387

    
2388
    return not self.bad
2389

    
2390
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
2391
    """Analyze the post-hooks' result
2392

2393
    This method analyses the hook result, handles it, and sends some
2394
    nicely-formatted feedback back to the user.
2395

2396
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
2397
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
2398
    @param hooks_results: the results of the multi-node hooks rpc call
2399
    @param feedback_fn: function used send feedback back to the caller
2400
    @param lu_result: previous Exec result
2401
    @return: the new Exec result, based on the previous result
2402
        and hook results
2403

2404
    """
2405
    # We only really run POST phase hooks, and are only interested in
2406
    # their results
2407
    if phase == constants.HOOKS_PHASE_POST:
2408
      # Used to change hooks' output to proper indentation
2409
      feedback_fn("* Hooks Results")
2410
      assert hooks_results, "invalid result from hooks"
2411

    
2412
      for node_name in hooks_results:
2413
        res = hooks_results[node_name]
2414
        msg = res.fail_msg
2415
        test = msg and not res.offline
2416
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
2417
                      "Communication failure in hooks execution: %s", msg)
2418
        if res.offline or msg:
2419
          # No need to investigate payload if node is offline or gave an error.
2420
          # override manually lu_result here as _ErrorIf only
2421
          # overrides self.bad
2422
          lu_result = 1
2423
          continue
2424
        for script, hkr, output in res.payload:
2425
          test = hkr == constants.HKR_FAIL
2426
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
2427
                        "Script %s failed, output:", script)
2428
          if test:
2429
            output = self._HOOKS_INDENT_RE.sub('      ', output)
2430
            feedback_fn("%s" % output)
2431
            lu_result = 0
2432

    
2433
      return lu_result
2434

    
2435

    
2436
class LUClusterVerifyDisks(NoHooksLU):
2437
  """Verifies the cluster disks status.
2438

2439
  """
2440
  REQ_BGL = False
2441

    
2442
  def ExpandNames(self):
2443
    self.needed_locks = {
2444
      locking.LEVEL_NODE: locking.ALL_SET,
2445
      locking.LEVEL_INSTANCE: locking.ALL_SET,
2446
    }
2447
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
2448

    
2449
  def Exec(self, feedback_fn):
2450
    """Verify integrity of cluster disks.
2451

2452
    @rtype: tuple of three items
2453
    @return: a tuple of (dict of node-to-node_error, list of instances
2454
        which need activate-disks, dict of instance: (node, volume) for
2455
        missing volumes
2456

2457
    """
2458
    result = res_nodes, res_instances, res_missing = {}, [], {}
2459

    
2460
    nodes = utils.NiceSort(self.cfg.GetVmCapableNodeList())
2461
    instances = self.cfg.GetAllInstancesInfo().values()
2462

    
2463
    nv_dict = {}
2464
    for inst in instances:
2465
      inst_lvs = {}
2466
      if not inst.admin_up:
2467
        continue
2468
      inst.MapLVsByNode(inst_lvs)
2469
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
2470
      for node, vol_list in inst_lvs.iteritems():
2471
        for vol in vol_list:
2472
          nv_dict[(node, vol)] = inst
2473

    
2474
    if not nv_dict:
2475
      return result
2476

    
2477
    node_lvs = self.rpc.call_lv_list(nodes, [])
2478
    for node, node_res in node_lvs.items():
2479
      if node_res.offline:
2480
        continue
2481
      msg = node_res.fail_msg
2482
      if msg:
2483
        logging.warning("Error enumerating LVs on node %s: %s", node, msg)
2484
        res_nodes[node] = msg
2485
        continue
2486

    
2487
      lvs = node_res.payload
2488
      for lv_name, (_, _, lv_online) in lvs.items():
2489
        inst = nv_dict.pop((node, lv_name), None)
2490
        if (not lv_online and inst is not None
2491
            and inst.name not in res_instances):
2492
          res_instances.append(inst.name)
2493

    
2494
    # any leftover items in nv_dict are missing LVs, let's arrange the
2495
    # data better
2496
    for key, inst in nv_dict.iteritems():
2497
      if inst.name not in res_missing:
2498
        res_missing[inst.name] = []
2499
      res_missing[inst.name].append(key)
2500

    
2501
    return result
2502

    
2503

    
2504
class LUClusterRepairDiskSizes(NoHooksLU):
2505
  """Verifies the cluster disks sizes.
2506

2507
  """
2508
  REQ_BGL = False
2509

    
2510
  def ExpandNames(self):
2511
    if self.op.instances:
2512
      self.wanted_names = []
2513
      for name in self.op.instances:
2514
        full_name = _ExpandInstanceName(self.cfg, name)
2515
        self.wanted_names.append(full_name)
2516
      self.needed_locks = {
2517
        locking.LEVEL_NODE: [],
2518
        locking.LEVEL_INSTANCE: self.wanted_names,
2519
        }
2520
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2521
    else:
2522
      self.wanted_names = None
2523
      self.needed_locks = {
2524
        locking.LEVEL_NODE: locking.ALL_SET,
2525
        locking.LEVEL_INSTANCE: locking.ALL_SET,
2526
        }
2527
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
2528

    
2529
  def DeclareLocks(self, level):
2530
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
2531
      self._LockInstancesNodes(primary_only=True)
2532

    
2533
  def CheckPrereq(self):
2534
    """Check prerequisites.
2535

2536
    This only checks the optional instance list against the existing names.
2537

2538
    """
2539
    if self.wanted_names is None:
2540
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
2541

    
2542
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
2543
                             in self.wanted_names]
2544

    
2545
  def _EnsureChildSizes(self, disk):
2546
    """Ensure children of the disk have the needed disk size.
2547

2548
    This is valid mainly for DRBD8 and fixes an issue where the
2549
    children have smaller disk size.
2550

2551
    @param disk: an L{ganeti.objects.Disk} object
2552

2553
    """
2554
    if disk.dev_type == constants.LD_DRBD8:
2555
      assert disk.children, "Empty children for DRBD8?"
2556
      fchild = disk.children[0]
2557
      mismatch = fchild.size < disk.size
2558
      if mismatch:
2559
        self.LogInfo("Child disk has size %d, parent %d, fixing",
2560
                     fchild.size, disk.size)
2561
        fchild.size = disk.size
2562

    
2563
      # and we recurse on this child only, not on the metadev
2564
      return self._EnsureChildSizes(fchild) or mismatch
2565
    else:
2566
      return False
2567

    
2568
  def Exec(self, feedback_fn):
2569
    """Verify the size of cluster disks.
2570

2571
    """
2572
    # TODO: check child disks too
2573
    # TODO: check differences in size between primary/secondary nodes
2574
    per_node_disks = {}
2575
    for instance in self.wanted_instances:
2576
      pnode = instance.primary_node
2577
      if pnode not in per_node_disks:
2578
        per_node_disks[pnode] = []
2579
      for idx, disk in enumerate(instance.disks):
2580
        per_node_disks[pnode].append((instance, idx, disk))
2581

    
2582
    changed = []
2583
    for node, dskl in per_node_disks.items():
2584
      newl = [v[2].Copy() for v in dskl]
2585
      for dsk in newl:
2586
        self.cfg.SetDiskID(dsk, node)
2587
      result = self.rpc.call_blockdev_getsize(node, newl)
2588
      if result.fail_msg:
2589
        self.LogWarning("Failure in blockdev_getsize call to node"
2590
                        " %s, ignoring", node)
2591
        continue
2592
      if len(result.payload) != len(dskl):
2593
        logging.warning("Invalid result from node %s: len(dksl)=%d,"
2594
                        " result.payload=%s", node, len(dskl), result.payload)
2595
        self.LogWarning("Invalid result from node %s, ignoring node results",
2596
                        node)
2597
        continue
2598
      for ((instance, idx, disk), size) in zip(dskl, result.payload):
2599
        if size is None:
2600
          self.LogWarning("Disk %d of instance %s did not return size"
2601
                          " information, ignoring", idx, instance.name)
2602
          continue
2603
        if not isinstance(size, (int, long)):
2604
          self.LogWarning("Disk %d of instance %s did not return valid"
2605
                          " size information, ignoring", idx, instance.name)
2606
          continue
2607
        size = size >> 20
2608
        if size != disk.size:
2609
          self.LogInfo("Disk %d of instance %s has mismatched size,"
2610
                       " correcting: recorded %d, actual %d", idx,
2611
                       instance.name, disk.size, size)
2612
          disk.size = size
2613
          self.cfg.Update(instance, feedback_fn)
2614
          changed.append((instance.name, idx, size))
2615
        if self._EnsureChildSizes(disk):
2616
          self.cfg.Update(instance, feedback_fn)
2617
          changed.append((instance.name, idx, disk.size))
2618
    return changed
2619

    
2620

    
2621
class LUClusterRename(LogicalUnit):
2622
  """Rename the cluster.
2623

2624
  """
2625
  HPATH = "cluster-rename"
2626
  HTYPE = constants.HTYPE_CLUSTER
2627

    
2628
  def BuildHooksEnv(self):
2629
    """Build hooks env.
2630

2631
    """
2632
    env = {
2633
      "OP_TARGET": self.cfg.GetClusterName(),
2634
      "NEW_NAME": self.op.name,
2635
      }
2636
    mn = self.cfg.GetMasterNode()
2637
    all_nodes = self.cfg.GetNodeList()
2638
    return env, [mn], all_nodes
2639

    
2640
  def CheckPrereq(self):
2641
    """Verify that the passed name is a valid one.
2642

2643
    """
2644
    hostname = netutils.GetHostname(name=self.op.name,
2645
                                    family=self.cfg.GetPrimaryIPFamily())
2646

    
2647
    new_name = hostname.name
2648
    self.ip = new_ip = hostname.ip
2649
    old_name = self.cfg.GetClusterName()
2650
    old_ip = self.cfg.GetMasterIP()
2651
    if new_name == old_name and new_ip == old_ip:
2652
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
2653
                                 " cluster has changed",
2654
                                 errors.ECODE_INVAL)
2655
    if new_ip != old_ip:
2656
      if netutils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
2657
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
2658
                                   " reachable on the network" %
2659
                                   new_ip, errors.ECODE_NOTUNIQUE)
2660

    
2661
    self.op.name = new_name
2662

    
2663
  def Exec(self, feedback_fn):
2664
    """Rename the cluster.
2665

2666
    """
2667
    clustername = self.op.name
2668
    ip = self.ip
2669

    
2670
    # shutdown the master IP
2671
    master = self.cfg.GetMasterNode()
2672
    result = self.rpc.call_node_stop_master(master, False)
2673
    result.Raise("Could not disable the master role")
2674

    
2675
    try:
2676
      cluster = self.cfg.GetClusterInfo()
2677
      cluster.cluster_name = clustername
2678
      cluster.master_ip = ip
2679
      self.cfg.Update(cluster, feedback_fn)
2680

    
2681
      # update the known hosts file
2682
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
2683
      node_list = self.cfg.GetOnlineNodeList()
2684
      try:
2685
        node_list.remove(master)
2686
      except ValueError:
2687
        pass
2688
      _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
2689
    finally:
2690
      result = self.rpc.call_node_start_master(master, False, False)
2691
      msg = result.fail_msg
2692
      if msg:
2693
        self.LogWarning("Could not re-enable the master role on"
2694
                        " the master, please restart manually: %s", msg)
2695

    
2696
    return clustername
2697

    
2698

    
2699
class LUClusterSetParams(LogicalUnit):
2700
  """Change the parameters of the cluster.
2701

2702
  """
2703
  HPATH = "cluster-modify"
2704
  HTYPE = constants.HTYPE_CLUSTER
2705
  REQ_BGL = False
2706

    
2707
  def CheckArguments(self):
2708
    """Check parameters
2709

2710
    """
2711
    if self.op.uid_pool:
2712
      uidpool.CheckUidPool(self.op.uid_pool)
2713

    
2714
    if self.op.add_uids:
2715
      uidpool.CheckUidPool(self.op.add_uids)
2716

    
2717
    if self.op.remove_uids:
2718
      uidpool.CheckUidPool(self.op.remove_uids)
2719

    
2720
  def ExpandNames(self):
2721
    # FIXME: in the future maybe other cluster params won't require checking on
2722
    # all nodes to be modified.
2723
    self.needed_locks = {
2724
      locking.LEVEL_NODE: locking.ALL_SET,
2725
    }
2726
    self.share_locks[locking.LEVEL_NODE] = 1
2727

    
2728
  def BuildHooksEnv(self):
2729
    """Build hooks env.
2730

2731
    """
2732
    env = {
2733
      "OP_TARGET": self.cfg.GetClusterName(),
2734
      "NEW_VG_NAME": self.op.vg_name,
2735
      }
2736
    mn = self.cfg.GetMasterNode()
2737
    return env, [mn], [mn]
2738

    
2739
  def CheckPrereq(self):
2740
    """Check prerequisites.
2741

2742
    This checks whether the given params don't conflict and
2743
    if the given volume group is valid.
2744

2745
    """
2746
    if self.op.vg_name is not None and not self.op.vg_name:
2747
      if self.cfg.HasAnyDiskOfType(constants.LD_LV):
2748
        raise errors.OpPrereqError("Cannot disable lvm storage while lvm-based"
2749
                                   " instances exist", errors.ECODE_INVAL)
2750

    
2751
    if self.op.drbd_helper is not None and not self.op.drbd_helper:
2752
      if self.cfg.HasAnyDiskOfType(constants.LD_DRBD8):
2753
        raise errors.OpPrereqError("Cannot disable drbd helper while"
2754
                                   " drbd-based instances exist",
2755
                                   errors.ECODE_INVAL)
2756

    
2757
    node_list = self.acquired_locks[locking.LEVEL_NODE]
2758

    
2759
    # if vg_name not None, checks given volume group on all nodes
2760
    if self.op.vg_name:
2761
      vglist = self.rpc.call_vg_list(node_list)
2762
      for node in node_list:
2763
        msg = vglist[node].fail_msg
2764
        if msg:
2765
          # ignoring down node
2766
          self.LogWarning("Error while gathering data on node %s"
2767
                          " (ignoring node): %s", node, msg)
2768
          continue
2769
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
2770
                                              self.op.vg_name,
2771
                                              constants.MIN_VG_SIZE)
2772
        if vgstatus:
2773
          raise errors.OpPrereqError("Error on node '%s': %s" %
2774
                                     (node, vgstatus), errors.ECODE_ENVIRON)
2775

    
2776
    if self.op.drbd_helper:
2777
      # checks given drbd helper on all nodes
2778
      helpers = self.rpc.call_drbd_helper(node_list)
2779
      for node in node_list:
2780
        ninfo = self.cfg.GetNodeInfo(node)
2781
        if ninfo.offline:
2782
          self.LogInfo("Not checking drbd helper on offline node %s", node)
2783
          continue
2784
        msg = helpers[node].fail_msg
2785
        if msg:
2786
          raise errors.OpPrereqError("Error checking drbd helper on node"
2787
                                     " '%s': %s" % (node, msg),
2788
                                     errors.ECODE_ENVIRON)
2789
        node_helper = helpers[node].payload
2790
        if node_helper != self.op.drbd_helper:
2791
          raise errors.OpPrereqError("Error on node '%s': drbd helper is %s" %
2792
                                     (node, node_helper), errors.ECODE_ENVIRON)
2793

    
2794
    self.cluster = cluster = self.cfg.GetClusterInfo()
2795
    # validate params changes
2796
    if self.op.beparams:
2797
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
2798
      self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
2799

    
2800
    if self.op.ndparams:
2801
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
2802
      self.new_ndparams = cluster.SimpleFillND(self.op.ndparams)
2803

    
2804
      # TODO: we need a more general way to handle resetting
2805
      # cluster-level parameters to default values
2806
      if self.new_ndparams["oob_program"] == "":
2807
        self.new_ndparams["oob_program"] = \
2808
            constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
2809

    
2810
    if self.op.nicparams:
2811
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
2812
      self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
2813
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
2814
      nic_errors = []
2815

    
2816
      # check all instances for consistency
2817
      for instance in self.cfg.GetAllInstancesInfo().values():
2818
        for nic_idx, nic in enumerate(instance.nics):
2819
          params_copy = copy.deepcopy(nic.nicparams)
2820
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
2821

    
2822
          # check parameter syntax
2823
          try:
2824
            objects.NIC.CheckParameterSyntax(params_filled)
2825
          except errors.ConfigurationError, err:
2826
            nic_errors.append("Instance %s, nic/%d: %s" %
2827
                              (instance.name, nic_idx, err))
2828

    
2829
          # if we're moving instances to routed, check that they have an ip
2830
          target_mode = params_filled[constants.NIC_MODE]
2831
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
2832
            nic_errors.append("Instance %s, nic/%d: routed nick with no ip" %
2833
                              (instance.name, nic_idx))
2834
      if nic_errors:
2835
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
2836
                                   "\n".join(nic_errors))
2837

    
2838
    # hypervisor list/parameters
2839
    self.new_hvparams = new_hvp = objects.FillDict(cluster.hvparams, {})
2840
    if self.op.hvparams:
2841
      for hv_name, hv_dict in self.op.hvparams.items():
2842
        if hv_name not in self.new_hvparams:
2843
          self.new_hvparams[hv_name] = hv_dict
2844
        else:
2845
          self.new_hvparams[hv_name].update(hv_dict)
2846

    
2847
    # os hypervisor parameters
2848
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
2849
    if self.op.os_hvp:
2850
      for os_name, hvs in self.op.os_hvp.items():
2851
        if os_name not in self.new_os_hvp:
2852
          self.new_os_hvp[os_name] = hvs
2853
        else:
2854
          for hv_name, hv_dict in hvs.items():
2855
            if hv_name not in self.new_os_hvp[os_name]:
2856
              self.new_os_hvp[os_name][hv_name] = hv_dict
2857
            else:
2858
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
2859

    
2860
    # os parameters
2861
    self.new_osp = objects.FillDict(cluster.osparams, {})
2862
    if self.op.osparams:
2863
      for os_name, osp in self.op.osparams.items():
2864
        if os_name not in self.new_osp:
2865
          self.new_osp[os_name] = {}
2866

    
2867
        self.new_osp[os_name] = _GetUpdatedParams(self.new_osp[os_name], osp,
2868
                                                  use_none=True)
2869

    
2870
        if not self.new_osp[os_name]:
2871
          # we removed all parameters
2872
          del self.new_osp[os_name]
2873
        else:
2874
          # check the parameter validity (remote check)
2875
          _CheckOSParams(self, False, [self.cfg.GetMasterNode()],
2876
                         os_name, self.new_osp[os_name])
2877

    
2878
    # changes to the hypervisor list
2879
    if self.op.enabled_hypervisors is not None:
2880
      self.hv_list = self.op.enabled_hypervisors
2881
      for hv in self.hv_list:
2882
        # if the hypervisor doesn't already exist in the cluster
2883
        # hvparams, we initialize it to empty, and then (in both
2884
        # cases) we make sure to fill the defaults, as we might not
2885
        # have a complete defaults list if the hypervisor wasn't
2886
        # enabled before
2887
        if hv not in new_hvp:
2888
          new_hvp[hv] = {}
2889
        new_hvp[hv] = objects.FillDict(constants.HVC_DEFAULTS[hv], new_hvp[hv])
2890
        utils.ForceDictType(new_hvp[hv], constants.HVS_PARAMETER_TYPES)
2891
    else:
2892
      self.hv_list = cluster.enabled_hypervisors
2893

    
2894
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
2895
      # either the enabled list has changed, or the parameters have, validate
2896
      for hv_name, hv_params in self.new_hvparams.items():
2897
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
2898
            (self.op.enabled_hypervisors and
2899
             hv_name in self.op.enabled_hypervisors)):
2900
          # either this is a new hypervisor, or its parameters have changed
2901
          hv_class = hypervisor.GetHypervisor(hv_name)
2902
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
2903
          hv_class.CheckParameterSyntax(hv_params)
2904
          _CheckHVParams(self, node_list, hv_name, hv_params)
2905

    
2906
    if self.op.os_hvp:
2907
      # no need to check any newly-enabled hypervisors, since the
2908
      # defaults have already been checked in the above code-block
2909
      for os_name, os_hvp in self.new_os_hvp.items():
2910
        for hv_name, hv_params in os_hvp.items():
2911
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
2912
          # we need to fill in the new os_hvp on top of the actual hv_p
2913
          cluster_defaults = self.new_hvparams.get(hv_name, {})
2914
          new_osp = objects.FillDict(cluster_defaults, hv_params)
2915
          hv_class = hypervisor.GetHypervisor(hv_name)
2916
          hv_class.CheckParameterSyntax(new_osp)
2917
          _CheckHVParams(self, node_list, hv_name, new_osp)
2918

    
2919
    if self.op.default_iallocator:
2920
      alloc_script = utils.FindFile(self.op.default_iallocator,
2921
                                    constants.IALLOCATOR_SEARCH_PATH,
2922
                                    os.path.isfile)
2923
      if alloc_script is None:
2924
        raise errors.OpPrereqError("Invalid default iallocator script '%s'"
2925
                                   " specified" % self.op.default_iallocator,
2926
                                   errors.ECODE_INVAL)
2927

    
2928
  def Exec(self, feedback_fn):
2929
    """Change the parameters of the cluster.
2930

2931
    """
2932
    if self.op.vg_name is not None:
2933
      new_volume = self.op.vg_name
2934
      if not new_volume:
2935
        new_volume = None
2936
      if new_volume != self.cfg.GetVGName():
2937
        self.cfg.SetVGName(new_volume)
2938
      else:
2939
        feedback_fn("Cluster LVM configuration already in desired"
2940
                    " state, not changing")
2941
    if self.op.drbd_helper is not None:
2942
      new_helper = self.op.drbd_helper
2943
      if not new_helper:
2944
        new_helper = None
2945
      if new_helper != self.cfg.GetDRBDHelper():
2946
        self.cfg.SetDRBDHelper(new_helper)
2947
      else:
2948
        feedback_fn("Cluster DRBD helper already in desired state,"
2949
                    " not changing")
2950
    if self.op.hvparams:
2951
      self.cluster.hvparams = self.new_hvparams
2952
    if self.op.os_hvp:
2953
      self.cluster.os_hvp = self.new_os_hvp
2954
    if self.op.enabled_hypervisors is not None:
2955
      self.cluster.hvparams = self.new_hvparams
2956
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
2957
    if self.op.beparams:
2958
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
2959
    if self.op.nicparams:
2960
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
2961
    if self.op.osparams:
2962
      self.cluster.osparams = self.new_osp
2963
    if self.op.ndparams:
2964
      self.cluster.ndparams = self.new_ndparams
2965

    
2966
    if self.op.candidate_pool_size is not None:
2967
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
2968
      # we need to update the pool size here, otherwise the save will fail
2969
      _AdjustCandidatePool(self, [])
2970

    
2971
    if self.op.maintain_node_health is not None:
2972
      self.cluster.maintain_node_health = self.op.maintain_node_health
2973

    
2974
    if self.op.prealloc_wipe_disks is not None:
2975
      self.cluster.prealloc_wipe_disks = self.op.prealloc_wipe_disks
2976

    
2977
    if self.op.add_uids is not None:
2978
      uidpool.AddToUidPool(self.cluster.uid_pool, self.op.add_uids)
2979

    
2980
    if self.op.remove_uids is not None:
2981
      uidpool.RemoveFromUidPool(self.cluster.uid_pool, self.op.remove_uids)
2982

    
2983
    if self.op.uid_pool is not None:
2984
      self.cluster.uid_pool = self.op.uid_pool
2985

    
2986
    if self.op.default_iallocator is not None:
2987
      self.cluster.default_iallocator = self.op.default_iallocator
2988

    
2989
    if self.op.reserved_lvs is not None:
2990
      self.cluster.reserved_lvs = self.op.reserved_lvs
2991

    
2992
    def helper_os(aname, mods, desc):
2993
      desc += " OS list"
2994
      lst = getattr(self.cluster, aname)
2995
      for key, val in mods:
2996
        if key == constants.DDM_ADD:
2997
          if val in lst:
2998
            feedback_fn("OS %s already in %s, ignoring" % (val, desc))
2999
          else:
3000
            lst.append(val)
3001
        elif key == constants.DDM_REMOVE:
3002
          if val in lst:
3003
            lst.remove(val)
3004
          else:
3005
            feedback_fn("OS %s not found in %s, ignoring" % (val, desc))
3006
        else:
3007
          raise errors.ProgrammerError("Invalid modification '%s'" % key)
3008

    
3009
    if self.op.hidden_os:
3010
      helper_os("hidden_os", self.op.hidden_os, "hidden")
3011

    
3012
    if self.op.blacklisted_os:
3013
      helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
3014

    
3015
    if self.op.master_netdev:
3016
      master = self.cfg.GetMasterNode()
3017
      feedback_fn("Shutting down master ip on the current netdev (%s)" %
3018
                  self.cluster.master_netdev)
3019
      result = self.rpc.call_node_stop_master(master, False)
3020
      result.Raise("Could not disable the master ip")
3021
      feedback_fn("Changing master_netdev from %s to %s" %
3022
                  (self.cluster.master_netdev, self.op.master_netdev))
3023
      self.cluster.master_netdev = self.op.master_netdev
3024

    
3025
    self.cfg.Update(self.cluster, feedback_fn)
3026

    
3027
    if self.op.master_netdev:
3028
      feedback_fn("Starting the master ip on the new master netdev (%s)" %
3029
                  self.op.master_netdev)
3030
      result = self.rpc.call_node_start_master(master, False, False)
3031
      if result.fail_msg:
3032
        self.LogWarning("Could not re-enable the master ip on"
3033
                        " the master, please restart manually: %s",
3034
                        result.fail_msg)
3035

    
3036

    
3037
def _UploadHelper(lu, nodes, fname):
3038
  """Helper for uploading a file and showing warnings.
3039

3040
  """
3041
  if os.path.exists(fname):
3042
    result = lu.rpc.call_upload_file(nodes, fname)
3043
    for to_node, to_result in result.items():
3044
      msg = to_result.fail_msg
3045
      if msg:
3046
        msg = ("Copy of file %s to node %s failed: %s" %
3047
               (fname, to_node, msg))
3048
        lu.proc.LogWarning(msg)
3049

    
3050

    
3051
def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
3052
  """Distribute additional files which are part of the cluster configuration.
3053

3054
  ConfigWriter takes care of distributing the config and ssconf files, but
3055
  there are more files which should be distributed to all nodes. This function
3056
  makes sure those are copied.
3057

3058
  @param lu: calling logical unit
3059
  @param additional_nodes: list of nodes not in the config to distribute to
3060
  @type additional_vm: boolean
3061
  @param additional_vm: whether the additional nodes are vm-capable or not
3062

3063
  """
3064
  # 1. Gather target nodes
3065
  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
3066
  dist_nodes = lu.cfg.GetOnlineNodeList()
3067
  nvm_nodes = lu.cfg.GetNonVmCapableNodeList()
3068
  vm_nodes = [name for name in dist_nodes if name not in nvm_nodes]
3069
  if additional_nodes is not None:
3070
    dist_nodes.extend(additional_nodes)
3071
    if additional_vm:
3072
      vm_nodes.extend(additional_nodes)
3073
  if myself.name in dist_nodes:
3074
    dist_nodes.remove(myself.name)
3075
  if myself.name in vm_nodes:
3076
    vm_nodes.remove(myself.name)
3077

    
3078
  # 2. Gather files to distribute
3079
  dist_files = set([constants.ETC_HOSTS,
3080
                    constants.SSH_KNOWN_HOSTS_FILE,
3081
                    constants.RAPI_CERT_FILE,
3082
                    constants.RAPI_USERS_FILE,
3083
                    constants.CONFD_HMAC_KEY,
3084
                    constants.CLUSTER_DOMAIN_SECRET_FILE,
3085
                   ])
3086

    
3087
  vm_files = set()
3088
  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
3089
  for hv_name in enabled_hypervisors:
3090
    hv_class = hypervisor.GetHypervisor(hv_name)
3091
    vm_files.update(hv_class.GetAncillaryFiles())
3092

    
3093
  # 3. Perform the files upload
3094
  for fname in dist_files:
3095
    _UploadHelper(lu, dist_nodes, fname)
3096
  for fname in vm_files:
3097
    _UploadHelper(lu, vm_nodes, fname)
3098

    
3099

    
3100
class LUClusterRedistConf(NoHooksLU):
3101
  """Force the redistribution of cluster configuration.
3102

3103
  This is a very simple LU.
3104

3105
  """
3106
  REQ_BGL = False
3107

    
3108
  def ExpandNames(self):
3109
    self.needed_locks = {
3110
      locking.LEVEL_NODE: locking.ALL_SET,
3111
    }
3112
    self.share_locks[locking.LEVEL_NODE] = 1
3113

    
3114
  def Exec(self, feedback_fn):
3115
    """Redistribute the configuration.
3116

3117
    """
3118
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
3119
    _RedistributeAncillaryFiles(self)
3120

    
3121

    
3122
def _WaitForSync(lu, instance, disks=None, oneshot=False):
3123
  """Sleep and poll for an instance's disk to sync.
3124

3125
  """
3126
  if not instance.disks or disks is not None and not disks:
3127
    return True
3128

    
3129
  disks = _ExpandCheckDisks(instance, disks)
3130

    
3131
  if not oneshot:
3132
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
3133

    
3134
  node = instance.primary_node
3135

    
3136
  for dev in disks:
3137
    lu.cfg.SetDiskID(dev, node)
3138

    
3139
  # TODO: Convert to utils.Retry
3140

    
3141
  retries = 0
3142
  degr_retries = 10 # in seconds, as we sleep 1 second each time
3143
  while True:
3144
    max_time = 0
3145
    done = True
3146
    cumul_degraded = False
3147
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, disks)
3148
    msg = rstats.fail_msg
3149
    if msg:
3150
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
3151
      retries += 1
3152
      if retries >= 10:
3153
        raise errors.RemoteError("Can't contact node %s for mirror data,"
3154
                                 " aborting." % node)
3155
      time.sleep(6)
3156
      continue
3157
    rstats = rstats.payload
3158
    retries = 0
3159
    for i, mstat in enumerate(rstats):
3160
      if mstat is None:
3161
        lu.LogWarning("Can't compute data for node %s/%s",
3162
                           node, disks[i].iv_name)
3163
        continue
3164

    
3165
      cumul_degraded = (cumul_degraded or
3166
                        (mstat.is_degraded and mstat.sync_percent is None))
3167
      if mstat.sync_percent is not None:
3168
        done = False
3169
        if mstat.estimated_time is not None:
3170
          rem_time = ("%s remaining (estimated)" %
3171
                      utils.FormatSeconds(mstat.estimated_time))
3172
          max_time = mstat.estimated_time
3173
        else:
3174
          rem_time = "no time estimate"
3175
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
3176
                        (disks[i].iv_name, mstat.sync_percent, rem_time))
3177

    
3178
    # if we're done but degraded, let's do a few small retries, to
3179
    # make sure we see a stable and not transient situation; therefore
3180
    # we force restart of the loop
3181
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
3182
      logging.info("Degraded disks found, %d retries left", degr_retries)
3183
      degr_retries -= 1
3184
      time.sleep(1)
3185
      continue
3186

    
3187
    if done or oneshot:
3188
      break
3189

    
3190
    time.sleep(min(60, max_time))
3191

    
3192
  if done:
3193
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
3194
  return not cumul_degraded
3195

    
3196

    
3197
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
3198
  """Check that mirrors are not degraded.
3199

3200
  The ldisk parameter, if True, will change the test from the
3201
  is_degraded attribute (which represents overall non-ok status for
3202
  the device(s)) to the ldisk (representing the local storage status).
3203

3204
  """
3205
  lu.cfg.SetDiskID(dev, node)
3206

    
3207
  result = True
3208

    
3209
  if on_primary or dev.AssembleOnSecondary():
3210
    rstats = lu.rpc.call_blockdev_find(node, dev)
3211
    msg = rstats.fail_msg
3212
    if msg:
3213
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
3214
      result = False
3215
    elif not rstats.payload:
3216
      lu.LogWarning("Can't find disk on node %s", node)
3217
      result = False
3218
    else:
3219
      if ldisk:
3220
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
3221
      else:
3222
        result = result and not rstats.payload.is_degraded
3223

    
3224
  if dev.children:
3225
    for child in dev.children:
3226
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
3227

    
3228
  return result
3229

    
3230

    
3231
class LUOobCommand(NoHooksLU):
3232
  """Logical unit for OOB handling.
3233

3234
  """
3235
  REG_BGL = False
3236

    
3237
  def CheckPrereq(self):
3238
    """Check prerequisites.
3239

3240
    This checks:
3241
     - the node exists in the configuration
3242
     - OOB is supported
3243

3244
    Any errors are signaled by raising errors.OpPrereqError.
3245

3246
    """
3247
    self.nodes = []
3248
    for node_name in self.op.node_names:
3249
      node = self.cfg.GetNodeInfo(node_name)
3250

    
3251
      if node is None:
3252
        raise errors.OpPrereqError("Node %s not found" % node_name,
3253
                                   errors.ECODE_NOENT)
3254
      else:
3255
        self.nodes.append(node)
3256

    
3257
      if (self.op.command == constants.OOB_POWER_OFF and not node.offline):
3258
        raise errors.OpPrereqError(("Cannot power off node %s because it is"
3259
                                    " not marked offline") % node_name,
3260
                                   errors.ECODE_STATE)
3261

    
3262
  def ExpandNames(self):
3263
    """Gather locks we need.
3264

3265
    """
3266
    if self.op.node_names:
3267
      self.op.node_names = [_ExpandNodeName(self.cfg, name)
3268
                            for name in self.op.node_names]
3269
    else:
3270
      self.op.node_names = self.cfg.GetNodeList()
3271

    
3272
    self.needed_locks = {
3273
      locking.LEVEL_NODE: self.op.node_names,
3274
      }
3275

    
3276
  def Exec(self, feedback_fn):
3277
    """Execute OOB and return result if we expect any.
3278

3279
    """
3280
    master_node = self.cfg.GetMasterNode()
3281
    ret = []
3282

    
3283
    for node in self.nodes:
3284
      node_entry = [(constants.RS_NORMAL, node.name)]
3285
      ret.append(node_entry)
3286

    
3287
      oob_program = _SupportsOob(self.cfg, node)
3288

    
3289
      if not oob_program:
3290
        node_entry.append((constants.RS_UNAVAIL, None))
3291
        continue
3292

    
3293
      logging.info("Executing out-of-band command '%s' using '%s' on %s",
3294
                   self.op.command, oob_program, node.name)
3295
      result = self.rpc.call_run_oob(master_node, oob_program,
3296
                                     self.op.command, node.name,
3297
                                     self.op.timeout)
3298

    
3299
      if result.fail_msg:
3300
        self.LogWarning("On node '%s' out-of-band RPC failed with: %s",
3301
                        node.name, result.fail_msg)
3302
        node_entry.append((constants.RS_NODATA, None))
3303
      else:
3304
        try:
3305
          self._CheckPayload(result)
3306
        except errors.OpExecError, err:
3307
          self.LogWarning("The payload returned by '%s' is not valid: %s",
3308
                          node.name, err)
3309
          node_entry.append((constants.RS_NODATA, None))
3310
        else:
3311
          if self.op.command == constants.OOB_HEALTH:
3312
            # For health we should log important events
3313
            for item, status in result.payload:
3314
              if status in [constants.OOB_STATUS_WARNING,
3315
                            constants.OOB_STATUS_CRITICAL]:
3316
                self.LogWarning("On node '%s' item '%s' has status '%s'",
3317
                                node.name, item, status)
3318

    
3319
          if self.op.command == constants.OOB_POWER_ON:
3320
            node.powered = True
3321
          elif self.op.command == constants.OOB_POWER_OFF:
3322
            node.powered = False
3323
          elif self.op.command == constants.OOB_POWER_STATUS:
3324
            powered = result.payload[constants.OOB_POWER_STATUS_POWERED]
3325
            if powered != node.powered:
3326
              logging.warning(("Recorded power state (%s) of node '%s' does not"
3327
                               " match actual power state (%s)"), node.powered,
3328
                              node.name, powered)
3329

    
3330
          # For configuration changing commands we should update the node
3331
          if self.op.command in (constants.OOB_POWER_ON,
3332
                                 constants.OOB_POWER_OFF):
3333
            self.cfg.Update(node, feedback_fn)
3334

    
3335
          node_entry.append((constants.RS_NORMAL, result.payload))
3336

    
3337
    return ret
3338

    
3339
  def _CheckPayload(self, result):
3340
    """Checks if the payload is valid.
3341

3342
    @param result: RPC result
3343
    @raises errors.OpExecError: If payload is not valid
3344

3345
    """
3346
    errs = []
3347
    if self.op.command == constants.OOB_HEALTH:
3348
      if not isinstance(result.payload, list):
3349
        errs.append("command 'health' is expected to return a list but got %s" %
3350
                    type(result.payload))
3351
      else:
3352
        for item, status in result.payload:
3353
          if status not in constants.OOB_STATUSES:
3354
            errs.append("health item '%s' has invalid status '%s'" %
3355
                        (item, status))
3356

    
3357
    if self.op.command == constants.OOB_POWER_STATUS:
3358
      if not isinstance(result.payload, dict):
3359
        errs.append("power-status is expected to return a dict but got %s" %
3360
                    type(result.payload))
3361

    
3362
    if self.op.command in [
3363
        constants.OOB_POWER_ON,
3364
        constants.OOB_POWER_OFF,
3365
        constants.OOB_POWER_CYCLE,
3366
        ]:
3367
      if result.payload is not None:
3368
        errs.append("%s is expected to not return payload but got '%s'" %
3369
                    (self.op.command, result.payload))
3370

    
3371
    if errs:
3372
      raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
3373
                               utils.CommaJoin(errs))
3374

    
3375

    
3376

    
3377
class LUOsDiagnose(NoHooksLU):
3378
  """Logical unit for OS diagnose/query.
3379

3380
  """
3381
  REQ_BGL = False
3382
  _HID = "hidden"
3383
  _BLK = "blacklisted"
3384
  _VLD = "valid"
3385
  _FIELDS_STATIC = utils.FieldSet()
3386
  _FIELDS_DYNAMIC = utils.FieldSet("name", _VLD, "node_status", "variants",
3387
                                   "parameters", "api_versions", _HID, _BLK)
3388

    
3389
  def CheckArguments(self):
3390
    if self.op.names:
3391
      raise errors.OpPrereqError("Selective OS query not supported",
3392
                                 errors.ECODE_INVAL)
3393

    
3394
    _CheckOutputFields(static=self._FIELDS_STATIC,
3395
                       dynamic=self._FIELDS_DYNAMIC,
3396
                       selected=self.op.output_fields)
3397

    
3398
  def ExpandNames(self):
3399
    # Lock all nodes, in shared mode
3400
    # Temporary removal of locks, should be reverted later
3401
    # TODO: reintroduce locks when they are lighter-weight
3402
    self.needed_locks = {}
3403
    #self.share_locks[locking.LEVEL_NODE] = 1
3404
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3405

    
3406
  @staticmethod
3407
  def _DiagnoseByOS(rlist):
3408
    """Remaps a per-node return list into an a per-os per-node dictionary
3409

3410
    @param rlist: a map with node names as keys and OS objects as values
3411

3412
    @rtype: dict
3413
    @return: a dictionary with osnames as keys and as value another
3414
        map, with nodes as keys and tuples of (path, status, diagnose,
3415
        variants, parameters, api_versions) as values, eg::
3416

3417
          {"debian-etch": {"node1": [(/usr/lib/..., True, "", [], []),
3418
                                     (/srv/..., False, "invalid api")],
3419
                           "node2": [(/srv/..., True, "", [], [])]}
3420
          }
3421

3422
    """
3423
    all_os = {}
3424
    # we build here the list of nodes that didn't fail the RPC (at RPC
3425
    # level), so that nodes with a non-responding node daemon don't
3426
    # make all OSes invalid
3427
    good_nodes = [node_name for node_name in rlist
3428
                  if not rlist[node_name].fail_msg]
3429
    for node_name, nr in rlist.items():
3430
      if nr.fail_msg or not nr.payload:
3431
        continue
3432
      for (name, path, status, diagnose, variants,
3433
           params, api_versions) in nr.payload:
3434
        if name not in all_os:
3435
          # build a list of nodes for this os containing empty lists
3436
          # for each node in node_list
3437
          all_os[name] = {}
3438
          for nname in good_nodes:
3439
            all_os[name][nname] = []
3440
        # convert params from [name, help] to (name, help)
3441
        params = [tuple(v) for v in params]
3442
        all_os[name][node_name].append((path, status, diagnose,
3443
                                        variants, params, api_versions))
3444
    return all_os
3445

    
3446
  def Exec(self, feedback_fn):
3447
    """Compute the list of OSes.
3448

3449
    """
3450
    valid_nodes = [node.name
3451
                   for node in self.cfg.GetAllNodesInfo().values()
3452
                   if not node.offline and node.vm_capable]
3453
    node_data = self.rpc.call_os_diagnose(valid_nodes)
3454
    pol = self._DiagnoseByOS(node_data)
3455
    output = []
3456
    cluster = self.cfg.GetClusterInfo()
3457

    
3458
    for os_name in utils.NiceSort(pol.keys()):
3459
      os_data = pol[os_name]
3460
      row = []
3461
      valid = True
3462
      (variants, params, api_versions) = null_state = (set(), set(), set())
3463
      for idx, osl in enumerate(os_data.values()):
3464
        valid = bool(valid and osl and osl[0][1])
3465
        if not valid:
3466
          (variants, params, api_versions) = null_state
3467
          break
3468
        node_variants, node_params, node_api = osl[0][3:6]
3469
        if idx == 0: # first entry
3470
          variants = set(node_variants)
3471
          params = set(node_params)
3472
          api_versions = set(node_api)
3473
        else: # keep consistency
3474
          variants.intersection_update(node_variants)
3475
          params.intersection_update(node_params)
3476
          api_versions.intersection_update(node_api)
3477

    
3478
      is_hid = os_name in cluster.hidden_os
3479
      is_blk = os_name in cluster.blacklisted_os
3480
      if ((self._HID not in self.op.output_fields and is_hid) or
3481
          (self._BLK not in self.op.output_fields and is_blk) or
3482
          (self._VLD not in self.op.output_fields and not valid)):
3483
        continue
3484

    
3485
      for field in self.op.output_fields:
3486
        if field == "name":
3487
          val = os_name
3488
        elif field == self._VLD:
3489
          val = valid
3490
        elif field == "node_status":
3491
          # this is just a copy of the dict
3492
          val = {}
3493
          for node_name, nos_list in os_data.items():
3494
            val[node_name] = nos_list
3495
        elif field == "variants":
3496
          val = utils.NiceSort(list(variants))
3497
        elif field == "parameters":
3498
          val = list(params)
3499
        elif field == "api_versions":
3500
          val = list(api_versions)
3501
        elif field == self._HID:
3502
          val = is_hid
3503
        elif field == self._BLK:
3504
          val = is_blk
3505
        else:
3506
          raise errors.ParameterError(field)
3507
        row.append(val)
3508
      output.append(row)
3509

    
3510
    return output
3511

    
3512

    
3513
class LUNodeRemove(LogicalUnit):
3514
  """Logical unit for removing a node.
3515

3516
  """
3517
  HPATH = "node-remove"
3518
  HTYPE = constants.HTYPE_NODE
3519

    
3520
  def BuildHooksEnv(self):
3521
    """Build hooks env.
3522

3523
    This doesn't run on the target node in the pre phase as a failed
3524
    node would then be impossible to remove.
3525

3526
    """
3527
    env = {
3528
      "OP_TARGET": self.op.node_name,
3529
      "NODE_NAME": self.op.node_name,
3530
      }
3531
    all_nodes = self.cfg.GetNodeList()
3532
    try:
3533
      all_nodes.remove(self.op.node_name)
3534
    except ValueError:
3535
      logging.warning("Node %s which is about to be removed not found"
3536
                      " in the all nodes list", self.op.node_name)
3537
    return env, all_nodes, all_nodes
3538

    
3539
  def CheckPrereq(self):
3540
    """Check prerequisites.
3541

3542
    This checks:
3543
     - the node exists in the configuration
3544
     - it does not have primary or secondary instances
3545
     - it's not the master
3546

3547
    Any errors are signaled by raising errors.OpPrereqError.
3548

3549
    """
3550
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
3551
    node = self.cfg.GetNodeInfo(self.op.node_name)
3552
    assert node is not None
3553

    
3554
    instance_list = self.cfg.GetInstanceList()
3555

    
3556
    masternode = self.cfg.GetMasterNode()
3557
    if node.name == masternode:
3558
      raise errors.OpPrereqError("Node is the master node,"
3559
                                 " you need to failover first.",
3560
                                 errors.ECODE_INVAL)
3561

    
3562
    for instance_name in instance_list:
3563
      instance = self.cfg.GetInstanceInfo(instance_name)
3564
      if node.name in instance.all_nodes:
3565
        raise errors.OpPrereqError("Instance %s is still running on the node,"
3566
                                   " please remove first." % instance_name,
3567
                                   errors.ECODE_INVAL)
3568
    self.op.node_name = node.name
3569
    self.node = node
3570

    
3571
  def Exec(self, feedback_fn):
3572
    """Removes the node from the cluster.
3573

3574
    """
3575
    node = self.node
3576
    logging.info("Stopping the node daemon and removing configs from node %s",
3577
                 node.name)
3578

    
3579
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
3580

    
3581
    # Promote nodes to master candidate as needed
3582
    _AdjustCandidatePool(self, exceptions=[node.name])
3583
    self.context.RemoveNode(node.name)
3584

    
3585
    # Run post hooks on the node before it's removed
3586
    hm = self.proc.hmclass(self.rpc.call_hooks_runner, self)
3587
    try:
3588
      hm.RunPhase(constants.HOOKS_PHASE_POST, [node.name])
3589
    except:
3590
      # pylint: disable-msg=W0702
3591
      self.LogWarning("Errors occurred running hooks on %s" % node.name)
3592

    
3593
    result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
3594
    msg = result.fail_msg
3595
    if msg:
3596
      self.LogWarning("Errors encountered on the remote node while leaving"
3597
                      " the cluster: %s", msg)
3598

    
3599
    # Remove node from our /etc/hosts
3600
    if self.cfg.GetClusterInfo().modify_etc_hosts:
3601
      master_node = self.cfg.GetMasterNode()
3602
      result = self.rpc.call_etc_hosts_modify(master_node,
3603
                                              constants.ETC_HOSTS_REMOVE,
3604
                                              node.name, None)
3605
      result.Raise("Can't update hosts file with new host data")
3606
      _RedistributeAncillaryFiles(self)
3607

    
3608

    
3609
class _NodeQuery(_QueryBase):
3610
  FIELDS = query.NODE_FIELDS
3611

    
3612
  def ExpandNames(self, lu):
3613
    lu.needed_locks = {}
3614
    lu.share_locks[locking.LEVEL_NODE] = 1
3615

    
3616
    if self.names:
3617
      self.wanted = _GetWantedNodes(lu, self.names)
3618
    else:
3619
      self.wanted = locking.ALL_SET
3620

    
3621
    self.do_locking = (self.use_locking and
3622
                       query.NQ_LIVE in self.requested_data)
3623

    
3624
    if self.do_locking:
3625
      # if we don't request only static fields, we need to lock the nodes
3626
      lu.needed_locks[locking.LEVEL_NODE] = self.wanted
3627

    
3628
  def DeclareLocks(self, lu, level):
3629
    pass
3630

    
3631
  def _GetQueryData(self, lu):
3632
    """Computes the list of nodes and their attributes.
3633

3634
    """
3635
    all_info = lu.cfg.GetAllNodesInfo()
3636

    
3637
    nodenames = self._GetNames(lu, all_info.keys(), locking.LEVEL_NODE)
3638

    
3639
    # Gather data as requested
3640
    if query.NQ_LIVE in self.requested_data:
3641
      # filter out non-vm_capable nodes
3642
      toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
3643

    
3644
      node_data = lu.rpc.call_node_info(toquery_nodes, lu.cfg.GetVGName(),
3645
                                        lu.cfg.GetHypervisorType())
3646
      live_data = dict((name, nresult.payload)
3647
                       for (name, nresult) in node_data.items()
3648
                       if not nresult.fail_msg and nresult.payload)
3649
    else:
3650
      live_data = None
3651

    
3652
    if query.NQ_INST in self.requested_data:
3653
      node_to_primary = dict([(name, set()) for name in nodenames])
3654
      node_to_secondary = dict([(name, set()) for name in nodenames])
3655

    
3656
      inst_data = lu.cfg.GetAllInstancesInfo()
3657

    
3658
      for inst in inst_data.values():
3659
        if inst.primary_node in node_to_primary:
3660
          node_to_primary[inst.primary_node].add(inst.name)
3661
        for secnode in inst.secondary_nodes:
3662
          if secnode in node_to_secondary:
3663
            node_to_secondary[secnode].add(inst.name)
3664
    else:
3665
      node_to_primary = None
3666
      node_to_secondary = None
3667

    
3668
    if query.NQ_OOB in self.requested_data:
3669
      oob_support = dict((name, bool(_SupportsOob(lu.cfg, node)))
3670
                         for name, node in all_info.iteritems())
3671
    else:
3672
      oob_support = None
3673

    
3674
    if query.NQ_GROUP in self.requested_data:
3675
      groups = lu.cfg.GetAllNodeGroupsInfo()
3676
    else:
3677
      groups = {}
3678

    
3679
    return query.NodeQueryData([all_info[name] for name in nodenames],
3680
                               live_data, lu.cfg.GetMasterNode(),
3681
                               node_to_primary, node_to_secondary, groups,
3682
                               oob_support, lu.cfg.GetClusterInfo())
3683

    
3684

    
3685
class LUNodeQuery(NoHooksLU):
3686
  """Logical unit for querying nodes.
3687

3688
  """
3689
  # pylint: disable-msg=W0142
3690
  REQ_BGL = False
3691

    
3692
  def CheckArguments(self):
3693
    self.nq = _NodeQuery(self.op.names, self.op.output_fields,
3694
                         self.op.use_locking)
3695

    
3696
  def ExpandNames(self):
3697
    self.nq.ExpandNames(self)
3698

    
3699
  def Exec(self, feedback_fn):
3700
    return self.nq.OldStyleQuery(self)
3701

    
3702

    
3703
class LUNodeQueryvols(NoHooksLU):
3704
  """Logical unit for getting volumes on node(s).
3705

3706
  """
3707
  REQ_BGL = False
3708
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
3709
  _FIELDS_STATIC = utils.FieldSet("node")
3710

    
3711
  def CheckArguments(self):
3712
    _CheckOutputFields(static=self._FIELDS_STATIC,
3713
                       dynamic=self._FIELDS_DYNAMIC,
3714
                       selected=self.op.output_fields)
3715

    
3716
  def ExpandNames(self):
3717
    self.needed_locks = {}
3718
    self.share_locks[locking.LEVEL_NODE] = 1
3719
    if not self.op.nodes:
3720
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3721
    else:
3722
      self.needed_locks[locking.LEVEL_NODE] = \
3723
        _GetWantedNodes(self, self.op.nodes)
3724

    
3725
  def Exec(self, feedback_fn):
3726
    """Computes the list of nodes and their attributes.
3727

3728
    """
3729
    nodenames = self.acquired_locks[locking.LEVEL_NODE]
3730
    volumes = self.rpc.call_node_volumes(nodenames)
3731

    
3732
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
3733
             in self.cfg.GetInstanceList()]
3734

    
3735
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
3736

    
3737
    output = []
3738
    for node in nodenames:
3739
      nresult = volumes[node]
3740
      if nresult.offline:
3741
        continue
3742
      msg = nresult.fail_msg
3743
      if msg:
3744
        self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
3745
        continue
3746

    
3747
      node_vols = nresult.payload[:]
3748
      node_vols.sort(key=lambda vol: vol['dev'])
3749

    
3750
      for vol in node_vols:
3751
        node_output = []
3752
        for field in self.op.output_fields:
3753
          if field == "node":
3754
            val = node
3755
          elif field == "phys":
3756
            val = vol['dev']
3757
          elif field == "vg":
3758
            val = vol['vg']
3759
          elif field == "name":
3760
            val = vol['name']
3761
          elif field == "size":
3762
            val = int(float(vol['size']))
3763
          elif field == "instance":
3764
            for inst in ilist:
3765
              if node not in lv_by_node[inst]:
3766
                continue
3767
              if vol['name'] in lv_by_node[inst][node]:
3768
                val = inst.name
3769
                break
3770
            else:
3771
              val = '-'
3772
          else:
3773
            raise errors.ParameterError(field)
3774
          node_output.append(str(val))
3775

    
3776
        output.append(node_output)
3777

    
3778
    return output
3779

    
3780

    
3781
class LUNodeQueryStorage(NoHooksLU):
3782
  """Logical unit for getting information on storage units on node(s).
3783

3784
  """
3785
  _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
3786
  REQ_BGL = False
3787

    
3788
  def CheckArguments(self):
3789
    _CheckOutputFields(static=self._FIELDS_STATIC,
3790
                       dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
3791
                       selected=self.op.output_fields)
3792

    
3793
  def ExpandNames(self):
3794
    self.needed_locks = {}
3795
    self.share_locks[locking.LEVEL_NODE] = 1
3796

    
3797
    if self.op.nodes:
3798
      self.needed_locks[locking.LEVEL_NODE] = \
3799
        _GetWantedNodes(self, self.op.nodes)
3800
    else:
3801
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3802

    
3803
  def Exec(self, feedback_fn):
3804
    """Computes the list of nodes and their attributes.
3805

3806
    """
3807
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
3808

    
3809
    # Always get name to sort by
3810
    if constants.SF_NAME in self.op.output_fields:
3811
      fields = self.op.output_fields[:]
3812
    else:
3813
      fields = [constants.SF_NAME] + self.op.output_fields
3814

    
3815
    # Never ask for node or type as it's only known to the LU
3816
    for extra in [constants.SF_NODE, constants.SF_TYPE]:
3817
      while extra in fields:
3818
        fields.remove(extra)
3819

    
3820
    field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
3821
    name_idx = field_idx[constants.SF_NAME]
3822

    
3823
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
3824
    data = self.rpc.call_storage_list(self.nodes,
3825
                                      self.op.storage_type, st_args,
3826
                                      self.op.name, fields)
3827

    
3828
    result = []
3829

    
3830
    for node in utils.NiceSort(self.nodes):
3831
      nresult = data[node]
3832
      if nresult.offline:
3833
        continue
3834

    
3835
      msg = nresult.fail_msg
3836
      if msg:
3837
        self.LogWarning("Can't get storage data from node %s: %s", node, msg)
3838
        continue
3839

    
3840
      rows = dict([(row[name_idx], row) for row in nresult.payload])
3841

    
3842
      for name in utils.NiceSort(rows.keys()):
3843
        row = rows[name]
3844

    
3845
        out = []
3846

    
3847
        for field in self.op.output_fields:
3848
          if field == constants.SF_NODE:
3849
            val = node
3850
          elif field == constants.SF_TYPE:
3851
            val = self.op.storage_type
3852
          elif field in field_idx:
3853
            val = row[field_idx[field]]
3854
          else:
3855
            raise errors.ParameterError(field)
3856

    
3857
          out.append(val)
3858

    
3859
        result.append(out)
3860

    
3861
    return result
3862

    
3863

    
3864
class _InstanceQuery(_QueryBase):
3865
  FIELDS = query.INSTANCE_FIELDS
3866

    
3867
  def ExpandNames(self, lu):
3868
    lu.needed_locks = {}
3869
    lu.share_locks[locking.LEVEL_INSTANCE] = 1
3870
    lu.share_locks[locking.LEVEL_NODE] = 1
3871

    
3872
    if self.names:
3873
      self.wanted = _GetWantedInstances(lu, self.names)
3874
    else:
3875
      self.wanted = locking.ALL_SET
3876

    
3877
    self.do_locking = (self.use_locking and
3878
                       query.IQ_LIVE in self.requested_data)
3879
    if self.do_locking:
3880
      lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3881
      lu.needed_locks[locking.LEVEL_NODE] = []
3882
      lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3883

    
3884
  def DeclareLocks(self, lu, level):
3885
    if level == locking.LEVEL_NODE and self.do_locking:
3886
      lu._LockInstancesNodes() # pylint: disable-msg=W0212
3887

    
3888
  def _GetQueryData(self, lu):
3889
    """Computes the list of instances and their attributes.
3890

3891
    """
3892
    cluster = lu.cfg.GetClusterInfo()
3893
    all_info = lu.cfg.GetAllInstancesInfo()
3894

    
3895
    instance_names = self._GetNames(lu, all_info.keys(), locking.LEVEL_INSTANCE)
3896

    
3897
    instance_list = [all_info[name] for name in instance_names]
3898
    nodes = frozenset(itertools.chain(*(inst.all_nodes
3899
                                        for inst in instance_list)))
3900
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3901
    bad_nodes = []
3902
    offline_nodes = []
3903
    wrongnode_inst = set()
3904

    
3905
    # Gather data as requested
3906
    if self.requested_data & set([query.IQ_LIVE, query.IQ_CONSOLE]):
3907
      live_data = {}
3908
      node_data = lu.rpc.call_all_instances_info(nodes, hv_list)
3909
      for name in nodes:
3910
        result = node_data[name]
3911
        if result.offline:
3912
          # offline nodes will be in both lists
3913
          assert result.fail_msg
3914
          offline_nodes.append(name)
3915
        if result.fail_msg:
3916
          bad_nodes.append(name)
3917
        elif result.payload:
3918
          for inst in result.payload:
3919
            if inst in all_info:
3920
              if all_info[inst].primary_node == name:
3921
                live_data.update(result.payload)
3922
              else:
3923
                wrongnode_inst.add(inst)
3924
            else:
3925
              # orphan instance; we don't list it here as we don't
3926
              # handle this case yet in the output of instance listing
3927
              logging.warning("Orphan instance '%s' found on node %s",
3928
                              inst, name)
3929
        # else no instance is alive
3930
    else:
3931
      live_data = {}
3932

    
3933
    if query.IQ_DISKUSAGE in self.requested_data:
3934
      disk_usage = dict((inst.name,
3935
                         _ComputeDiskSize(inst.disk_template,
3936
                                          [{"size": disk.size}
3937
                                           for disk in inst.disks]))
3938
                        for inst in instance_list)
3939
    else:
3940
      disk_usage = None
3941

    
3942
    if query.IQ_CONSOLE in self.requested_data:
3943
      consinfo = {}
3944
      for inst in instance_list:
3945
        if inst.name in live_data:
3946
          # Instance is running
3947
          consinfo[inst.name] = _GetInstanceConsole(cluster, inst)
3948
        else:
3949
          consinfo[inst.name] = None
3950
      assert set(consinfo.keys()) == set(instance_names)
3951
    else:
3952
      consinfo = None
3953

    
3954
    return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
3955
                                   disk_usage, offline_nodes, bad_nodes,
3956
                                   live_data, wrongnode_inst, consinfo)
3957

    
3958

    
3959
class LUQuery(NoHooksLU):
3960
  """Query for resources/items of a certain kind.
3961

3962
  """
3963
  # pylint: disable-msg=W0142
3964
  REQ_BGL = False
3965

    
3966
  def CheckArguments(self):
3967
    qcls = _GetQueryImplementation(self.op.what)
3968
    names = qlang.ReadSimpleFilter("name", self.op.filter)
3969

    
3970
    self.impl = qcls(names, self.op.fields, False)
3971

    
3972
  def ExpandNames(self):
3973
    self.impl.ExpandNames(self)
3974

    
3975
  def DeclareLocks(self, level):
3976
    self.impl.DeclareLocks(self, level)
3977

    
3978
  def Exec(self, feedback_fn):
3979
    return self.impl.NewStyleQuery(self)
3980

    
3981

    
3982
class LUQueryFields(NoHooksLU):
3983
  """Query for resources/items of a certain kind.
3984

3985
  """
3986
  # pylint: disable-msg=W0142
3987
  REQ_BGL = False
3988

    
3989
  def CheckArguments(self):
3990
    self.qcls = _GetQueryImplementation(self.op.what)
3991

    
3992
  def ExpandNames(self):
3993
    self.needed_locks = {}
3994

    
3995
  def Exec(self, feedback_fn):
3996
    return self.qcls.FieldsQuery(self.op.fields)
3997

    
3998

    
3999
class LUNodeModifyStorage(NoHooksLU):
4000
  """Logical unit for modifying a storage volume on a node.
4001

4002
  """
4003
  REQ_BGL = False
4004

    
4005
  def CheckArguments(self):
4006
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4007

    
4008
    storage_type = self.op.storage_type
4009

    
4010
    try:
4011
      modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
4012
    except KeyError:
4013
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
4014
                                 " modified" % storage_type,
4015
                                 errors.ECODE_INVAL)
4016

    
4017
    diff = set(self.op.changes.keys()) - modifiable
4018
    if diff:
4019
      raise errors.OpPrereqError("The following fields can not be modified for"
4020
                                 " storage units of type '%s': %r" %
4021
                                 (storage_type, list(diff)),
4022
                                 errors.ECODE_INVAL)
4023

    
4024
  def ExpandNames(self):
4025
    self.needed_locks = {
4026
      locking.LEVEL_NODE: self.op.node_name,
4027
      }
4028

    
4029
  def Exec(self, feedback_fn):
4030
    """Computes the list of nodes and their attributes.
4031

4032
    """
4033
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
4034
    result = self.rpc.call_storage_modify(self.op.node_name,
4035
                                          self.op.storage_type, st_args,
4036
                                          self.op.name, self.op.changes)
4037
    result.Raise("Failed to modify storage unit '%s' on %s" %
4038
                 (self.op.name, self.op.node_name))
4039

    
4040

    
4041
class LUNodeAdd(LogicalUnit):
4042
  """Logical unit for adding node to the cluster.
4043

4044
  """
4045
  HPATH = "node-add"
4046
  HTYPE = constants.HTYPE_NODE
4047
  _NFLAGS = ["master_capable", "vm_capable"]
4048

    
4049
  def CheckArguments(self):
4050
    self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
4051
    # validate/normalize the node name
4052
    self.hostname = netutils.GetHostname(name=self.op.node_name,
4053
                                         family=self.primary_ip_family)
4054
    self.op.node_name = self.hostname.name
4055
    if self.op.readd and self.op.group:
4056
      raise errors.OpPrereqError("Cannot pass a node group when a node is"
4057
                                 " being readded", errors.ECODE_INVAL)
4058

    
4059
  def BuildHooksEnv(self):
4060
    """Build hooks env.
4061

4062
    This will run on all nodes before, and on all nodes + the new node after.
4063

4064
    """
4065
    env = {
4066
      "OP_TARGET": self.op.node_name,
4067
      "NODE_NAME": self.op.node_name,
4068
      "NODE_PIP": self.op.primary_ip,
4069
      "NODE_SIP": self.op.secondary_ip,
4070
      "MASTER_CAPABLE": str(self.op.master_capable),
4071
      "VM_CAPABLE": str(self.op.vm_capable),
4072
      }
4073
    nodes_0 = self.cfg.GetNodeList()
4074
    nodes_1 = nodes_0 + [self.op.node_name, ]
4075
    return env, nodes_0, nodes_1
4076

    
4077
  def CheckPrereq(self):
4078
    """Check prerequisites.
4079

4080
    This checks:
4081
     - the new node is not already in the config
4082
     - it is resolvable
4083
     - its parameters (single/dual homed) matches the cluster
4084

4085
    Any errors are signaled by raising errors.OpPrereqError.
4086

4087
    """
4088
    cfg = self.cfg
4089
    hostname = self.hostname
4090
    node = hostname.name
4091
    primary_ip = self.op.primary_ip = hostname.ip
4092
    if self.op.secondary_ip is None:
4093
      if self.primary_ip_family == netutils.IP6Address.family:
4094
        raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
4095
                                   " IPv4 address must be given as secondary",
4096
                                   errors.ECODE_INVAL)
4097
      self.op.secondary_ip = primary_ip
4098

    
4099
    secondary_ip = self.op.secondary_ip
4100
    if not netutils.IP4Address.IsValid(secondary_ip):
4101
      raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
4102
                                 " address" % secondary_ip, errors.ECODE_INVAL)
4103

    
4104
    node_list = cfg.GetNodeList()
4105
    if not self.op.readd and node in node_list:
4106
      raise errors.OpPrereqError("Node %s is already in the configuration" %
4107
                                 node, errors.ECODE_EXISTS)
4108
    elif self.op.readd and node not in node_list:
4109
      raise errors.OpPrereqError("Node %s is not in the configuration" % node,
4110
                                 errors.ECODE_NOENT)
4111

    
4112
    self.changed_primary_ip = False
4113

    
4114
    for existing_node_name in node_list:
4115
      existing_node = cfg.GetNodeInfo(existing_node_name)
4116

    
4117
      if self.op.readd and node == existing_node_name:
4118
        if existing_node.secondary_ip != secondary_ip:
4119
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
4120
                                     " address configuration as before",
4121
                                     errors.ECODE_INVAL)
4122
        if existing_node.primary_ip != primary_ip:
4123
          self.changed_primary_ip = True
4124

    
4125
        continue
4126

    
4127
      if (existing_node.primary_ip == primary_ip or
4128
          existing_node.secondary_ip == primary_ip or
4129
          existing_node.primary_ip == secondary_ip or
4130
          existing_node.secondary_ip == secondary_ip):
4131
        raise errors.OpPrereqError("New node ip address(es) conflict with"
4132
                                   " existing node %s" % existing_node.name,
4133
                                   errors.ECODE_NOTUNIQUE)
4134

    
4135
    # After this 'if' block, None is no longer a valid value for the
4136
    # _capable op attributes
4137
    if self.op.readd:
4138
      old_node = self.cfg.GetNodeInfo(node)
4139
      assert old_node is not None, "Can't retrieve locked node %s" % node
4140
      for attr in self._NFLAGS:
4141
        if getattr(self.op, attr) is None:
4142
          setattr(self.op, attr, getattr(old_node, attr))
4143
    else:
4144
      for attr in self._NFLAGS:
4145
        if getattr(self.op, attr) is None:
4146
          setattr(self.op, attr, True)
4147

    
4148
    if self.op.readd and not self.op.vm_capable:
4149
      pri, sec = cfg.GetNodeInstances(node)
4150
      if pri or sec:
4151
        raise errors.OpPrereqError("Node %s being re-added with vm_capable"
4152
                                   " flag set to false, but it already holds"
4153
                                   " instances" % node,
4154
                                   errors.ECODE_STATE)
4155

    
4156
    # check that the type of the node (single versus dual homed) is the
4157
    # same as for the master
4158
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
4159
    master_singlehomed = myself.secondary_ip == myself.primary_ip
4160
    newbie_singlehomed = secondary_ip == primary_ip
4161
    if master_singlehomed != newbie_singlehomed:
4162
      if master_singlehomed:
4163
        raise errors.OpPrereqError("The master has no secondary ip but the"
4164
                                   " new node has one",
4165
                                   errors.ECODE_INVAL)
4166
      else:
4167
        raise errors.OpPrereqError("The master has a secondary ip but the"
4168
                                   " new node doesn't have one",
4169
                                   errors.ECODE_INVAL)
4170

    
4171
    # checks reachability
4172
    if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
4173
      raise errors.OpPrereqError("Node not reachable by ping",
4174
                                 errors.ECODE_ENVIRON)
4175

    
4176
    if not newbie_singlehomed:
4177
      # check reachability from my secondary ip to newbie's secondary ip
4178
      if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
4179
                           source=myself.secondary_ip):
4180
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
4181
                                   " based ping to node daemon port",
4182
                                   errors.ECODE_ENVIRON)
4183

    
4184
    if self.op.readd:
4185
      exceptions = [node]
4186
    else:
4187
      exceptions = []
4188

    
4189
    if self.op.master_capable:
4190
      self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
4191
    else:
4192
      self.master_candidate = False
4193

    
4194
    if self.op.readd:
4195
      self.new_node = old_node
4196
    else:
4197
      node_group = cfg.LookupNodeGroup(self.op.group)
4198
      self.new_node = objects.Node(name=node,
4199
                                   primary_ip=primary_ip,
4200
                                   secondary_ip=secondary_ip,
4201
                                   master_candidate=self.master_candidate,
4202
                                   offline=False, drained=False,
4203
                                   group=node_group)
4204

    
4205
    if self.op.ndparams:
4206
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
4207

    
4208
  def Exec(self, feedback_fn):
4209
    """Adds the new node to the cluster.
4210

4211
    """
4212
    new_node = self.new_node
4213
    node = new_node.name
4214

    
4215
    # We adding a new node so we assume it's powered
4216
    new_node.powered = True
4217

    
4218
    # for re-adds, reset the offline/drained/master-candidate flags;
4219
    # we need to reset here, otherwise offline would prevent RPC calls
4220
    # later in the procedure; this also means that if the re-add
4221
    # fails, we are left with a non-offlined, broken node
4222
    if self.op.readd:
4223
      new_node.drained = new_node.offline = False # pylint: disable-msg=W0201
4224
      self.LogInfo("Readding a node, the offline/drained flags were reset")
4225
      # if we demote the node, we do cleanup later in the procedure
4226
      new_node.master_candidate = self.master_candidate
4227
      if self.changed_primary_ip:
4228
        new_node.primary_ip = self.op.primary_ip
4229

    
4230
    # copy the master/vm_capable flags
4231
    for attr in self._NFLAGS:
4232
      setattr(new_node, attr, getattr(self.op, attr))
4233

    
4234
    # notify the user about any possible mc promotion
4235
    if new_node.master_candidate:
4236
      self.LogInfo("Node will be a master candidate")
4237

    
4238
    if self.op.ndparams:
4239
      new_node.ndparams = self.op.ndparams
4240
    else:
4241
      new_node.ndparams = {}
4242

    
4243
    # check connectivity
4244
    result = self.rpc.call_version([node])[node]
4245
    result.Raise("Can't get version information from node %s" % node)
4246
    if constants.PROTOCOL_VERSION == result.payload:
4247
      logging.info("Communication to node %s fine, sw version %s match",
4248
                   node, result.payload)
4249
    else:
4250
      raise errors.OpExecError("Version mismatch master version %s,"
4251
                               " node version %s" %
4252
                               (constants.PROTOCOL_VERSION, result.payload))
4253

    
4254
    # Add node to our /etc/hosts, and add key to known_hosts
4255
    if self.cfg.GetClusterInfo().modify_etc_hosts:
4256
      master_node = self.cfg.GetMasterNode()
4257
      result = self.rpc.call_etc_hosts_modify(master_node,
4258
                                              constants.ETC_HOSTS_ADD,
4259
                                              self.hostname.name,
4260
                                              self.hostname.ip)
4261
      result.Raise("Can't update hosts file with new host data")
4262

    
4263
    if new_node.secondary_ip != new_node.primary_ip:
4264
      _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
4265
                               False)
4266

    
4267
    node_verify_list = [self.cfg.GetMasterNode()]
4268
    node_verify_param = {
4269
      constants.NV_NODELIST: [node],
4270
      # TODO: do a node-net-test as well?
4271
    }
4272

    
4273
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
4274
                                       self.cfg.GetClusterName())
4275
    for verifier in node_verify_list:
4276
      result[verifier].Raise("Cannot communicate with node %s" % verifier)
4277
      nl_payload = result[verifier].payload[constants.NV_NODELIST]
4278
      if nl_payload:
4279
        for failed in nl_payload:
4280
          feedback_fn("ssh/hostname verification failed"
4281
                      " (checking from %s): %s" %
4282
                      (verifier, nl_payload[failed]))
4283
        raise errors.OpExecError("ssh/hostname verification failed.")
4284

    
4285
    if self.op.readd:
4286
      _RedistributeAncillaryFiles(self)
4287
      self.context.ReaddNode(new_node)
4288
      # make sure we redistribute the config
4289
      self.cfg.Update(new_node, feedback_fn)
4290
      # and make sure the new node will not have old files around
4291
      if not new_node.master_candidate:
4292
        result = self.rpc.call_node_demote_from_mc(new_node.name)
4293
        msg = result.fail_msg
4294
        if msg:
4295
          self.LogWarning("Node failed to demote itself from master"
4296
                          " candidate status: %s" % msg)
4297
    else:
4298
      _RedistributeAncillaryFiles(self, additional_nodes=[node],
4299
                                  additional_vm=self.op.vm_capable)
4300
      self.context.AddNode(new_node, self.proc.GetECId())
4301

    
4302

    
4303
class LUNodeSetParams(LogicalUnit):
4304
  """Modifies the parameters of a node.
4305

4306
  @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
4307
      to the node role (as _ROLE_*)
4308
  @cvar _R2F: a dictionary from node role to tuples of flags
4309
  @cvar _FLAGS: a list of attribute names corresponding to the flags
4310

4311
  """
4312
  HPATH = "node-modify"
4313
  HTYPE = constants.HTYPE_NODE
4314
  REQ_BGL = False
4315
  (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
4316
  _F2R = {
4317
    (True, False, False): _ROLE_CANDIDATE,
4318
    (False, True, False): _ROLE_DRAINED,
4319
    (False, False, True): _ROLE_OFFLINE,
4320
    (False, False, False): _ROLE_REGULAR,
4321
    }
4322
  _R2F = dict((v, k) for k, v in _F2R.items())
4323
  _FLAGS = ["master_candidate", "drained", "offline"]
4324

    
4325
  def CheckArguments(self):
4326
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4327
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
4328
                self.op.master_capable, self.op.vm_capable,
4329
                self.op.secondary_ip, self.op.ndparams]
4330
    if all_mods.count(None) == len(all_mods):
4331
      raise errors.OpPrereqError("Please pass at least one modification",
4332
                                 errors.ECODE_INVAL)
4333
    if all_mods.count(True) > 1:
4334
      raise errors.OpPrereqError("Can't set the node into more than one"
4335
                                 " state at the same time",
4336
                                 errors.ECODE_INVAL)
4337

    
4338
    # Boolean value that tells us whether we might be demoting from MC
4339
    self.might_demote = (self.op.master_candidate == False or
4340
                         self.op.offline == True or
4341
                         self.op.drained == True or
4342
                         self.op.master_capable == False)
4343

    
4344
    if self.op.secondary_ip:
4345
      if not netutils.IP4Address.IsValid(self.op.secondary_ip):
4346
        raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
4347
                                   " address" % self.op.secondary_ip,
4348
                                   errors.ECODE_INVAL)
4349

    
4350
    self.lock_all = self.op.auto_promote and self.might_demote
4351
    self.lock_instances = self.op.secondary_ip is not None
4352

    
4353
  def ExpandNames(self):
4354
    if self.lock_all:
4355
      self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
4356
    else:
4357
      self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
4358

    
4359
    if self.lock_instances:
4360
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
4361

    
4362
  def DeclareLocks(self, level):
4363
    # If we have locked all instances, before waiting to lock nodes, release
4364
    # all the ones living on nodes unrelated to the current operation.
4365
    if level == locking.LEVEL_NODE and self.lock_instances:
4366
      instances_release = []
4367
      instances_keep = []
4368
      self.affected_instances = []
4369
      if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4370
        for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
4371
          instance = self.context.cfg.GetInstanceInfo(instance_name)
4372
          i_mirrored = instance.disk_template in constants.DTS_INT_MIRROR
4373
          if i_mirrored and self.op.node_name in instance.all_nodes:
4374
            instances_keep.append(instance_name)
4375
            self.affected_instances.append(instance)
4376
          else:
4377
            instances_release.append(instance_name)
4378
        if instances_release:
4379
          self.context.glm.release(locking.LEVEL_INSTANCE, instances_release)
4380
          self.acquired_locks[locking.LEVEL_INSTANCE] = instances_keep
4381

    
4382
  def BuildHooksEnv(self):
4383
    """Build hooks env.
4384

4385
    This runs on the master node.
4386

4387
    """
4388
    env = {
4389
      "OP_TARGET": self.op.node_name,
4390
      "MASTER_CANDIDATE": str(self.op.master_candidate),
4391
      "OFFLINE": str(self.op.offline),
4392
      "DRAINED": str(self.op.drained),
4393
      "MASTER_CAPABLE": str(self.op.master_capable),
4394
      "VM_CAPABLE": str(self.op.vm_capable),
4395
      }
4396
    nl = [self.cfg.GetMasterNode(),
4397
          self.op.node_name]
4398
    return env, nl, nl
4399

    
4400
  def CheckPrereq(self):
4401
    """Check prerequisites.
4402

4403
    This only checks the instance list against the existing names.
4404

4405
    """
4406
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
4407

    
4408
    if (self.op.master_candidate is not None or
4409
        self.op.drained is not None or
4410
        self.op.offline is not None):
4411
      # we can't change the master's node flags
4412
      if self.op.node_name == self.cfg.GetMasterNode():
4413
        raise errors.OpPrereqError("The master role can be changed"
4414
                                   " only via master-failover",
4415
                                   errors.ECODE_INVAL)
4416

    
4417
    if self.op.master_candidate and not node.master_capable:
4418
      raise errors.OpPrereqError("Node %s is not master capable, cannot make"
4419
                                 " it a master candidate" % node.name,
4420
                                 errors.ECODE_STATE)
4421

    
4422
    if self.op.vm_capable == False:
4423
      (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
4424
      if ipri or isec:
4425
        raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
4426
                                   " the vm_capable flag" % node.name,
4427
                                   errors.ECODE_STATE)
4428

    
4429
    if node.master_candidate and self.might_demote and not self.lock_all:
4430
      assert not self.op.auto_promote, "auto_promote set but lock_all not"
4431
      # check if after removing the current node, we're missing master
4432
      # candidates
4433
      (mc_remaining, mc_should, _) = \
4434
          self.cfg.GetMasterCandidateStats(exceptions=[node.name])
4435
      if mc_remaining < mc_should:
4436
        raise errors.OpPrereqError("Not enough master candidates, please"
4437
                                   " pass auto promote option to allow"
4438
                                   " promotion", errors.ECODE_STATE)
4439

    
4440
    self.old_flags = old_flags = (node.master_candidate,
4441
                                  node.drained, node.offline)
4442
    assert old_flags in self._F2R, "Un-handled old flags  %s" % str(old_flags)
4443
    self.old_role = old_role = self._F2R[old_flags]
4444

    
4445
    # Check for ineffective changes
4446
    for attr in self._FLAGS:
4447
      if (getattr(self.op, attr) == False and getattr(node, attr) == False):
4448
        self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
4449
        setattr(self.op, attr, None)
4450

    
4451
    # Past this point, any flag change to False means a transition
4452
    # away from the respective state, as only real changes are kept
4453

    
4454
    # TODO: We might query the real power state if it supports OOB
4455
    if _SupportsOob(self.cfg, node):
4456
      if self.op.offline is False and not (node.powered or
4457
                                           self.op.powered == True):
4458
        raise errors.OpPrereqError(("Please power on node %s first before you"
4459
                                    " can reset offline state") %
4460
                                   self.op.node_name)
4461
    elif self.op.powered is not None:
4462
      raise errors.OpPrereqError(("Unable to change powered state for node %s"
4463
                                  " which does not support out-of-band"
4464
                                  " handling") % self.op.node_name)
4465

    
4466
    # If we're being deofflined/drained, we'll MC ourself if needed
4467
    if (self.op.drained == False or self.op.offline == False or
4468
        (self.op.master_capable and not node.master_capable)):
4469
      if _DecideSelfPromotion(self):
4470
        self.op.master_candidate = True
4471
        self.LogInfo("Auto-promoting node to master candidate")
4472

    
4473
    # If we're no longer master capable, we'll demote ourselves from MC
4474
    if self.op.master_capable == False and node.master_candidate:
4475
      self.LogInfo("Demoting from master candidate")
4476
      self.op.master_candidate = False
4477

    
4478
    # Compute new role
4479
    assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
4480
    if self.op.master_candidate:
4481
      new_role = self._ROLE_CANDIDATE
4482
    elif self.op.drained:
4483
      new_role = self._ROLE_DRAINED
4484
    elif self.op.offline:
4485
      new_role = self._ROLE_OFFLINE
4486
    elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
4487
      # False is still in new flags, which means we're un-setting (the
4488
      # only) True flag
4489
      new_role = self._ROLE_REGULAR
4490
    else: # no new flags, nothing, keep old role
4491
      new_role = old_role
4492

    
4493
    self.new_role = new_role
4494

    
4495
    if old_role == self._ROLE_OFFLINE and new_role != old_role:
4496
      # Trying to transition out of offline status
4497
      result = self.rpc.call_version([node.name])[node.name]
4498
      if result.fail_msg:
4499
        raise errors.OpPrereqError("Node %s is being de-offlined but fails"
4500
                                   " to report its version: %s" %
4501
                                   (node.name, result.fail_msg),
4502
                                   errors.ECODE_STATE)
4503
      else:
4504
        self.LogWarning("Transitioning node from offline to online state"
4505
                        " without using re-add. Please make sure the node"
4506
                        " is healthy!")
4507

    
4508
    if self.op.secondary_ip:
4509
      # Ok even without locking, because this can't be changed by any LU
4510
      master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
4511
      master_singlehomed = master.secondary_ip == master.primary_ip
4512
      if master_singlehomed and self.op.secondary_ip:
4513
        raise errors.OpPrereqError("Cannot change the secondary ip on a single"
4514
                                   " homed cluster", errors.ECODE_INVAL)
4515

    
4516
      if node.offline:
4517
        if self.affected_instances:
4518
          raise errors.OpPrereqError("Cannot change secondary ip: offline"
4519
                                     " node has instances (%s) configured"
4520
                                     " to use it" % self.affected_instances)
4521
      else:
4522
        # On online nodes, check that no instances are running, and that
4523
        # the node has the new ip and we can reach it.
4524
        for instance in self.affected_instances:
4525
          _CheckInstanceDown(self, instance, "cannot change secondary ip")
4526

    
4527
        _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
4528
        if master.name != node.name:
4529
          # check reachability from master secondary ip to new secondary ip
4530
          if not netutils.TcpPing(self.op.secondary_ip,
4531
                                  constants.DEFAULT_NODED_PORT,
4532
                                  source=master.secondary_ip):
4533
            raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
4534
                                       " based ping to node daemon port",
4535
                                       errors.ECODE_ENVIRON)
4536

    
4537
    if self.op.ndparams:
4538
      new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
4539
      utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
4540
      self.new_ndparams = new_ndparams
4541

    
4542
  def Exec(self, feedback_fn):
4543
    """Modifies a node.
4544

4545
    """
4546
    node = self.node
4547
    old_role = self.old_role
4548
    new_role = self.new_role
4549

    
4550
    result = []
4551

    
4552
    if self.op.ndparams:
4553
      node.ndparams = self.new_ndparams
4554

    
4555
    if self.op.powered is not None:
4556
      node.powered = self.op.powered
4557

    
4558
    for attr in ["master_capable", "vm_capable"]:
4559
      val = getattr(self.op, attr)
4560
      if val is not None:
4561
        setattr(node, attr, val)
4562
        result.append((attr, str(val)))
4563

    
4564
    if new_role != old_role:
4565
      # Tell the node to demote itself, if no longer MC and not offline
4566
      if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
4567
        msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
4568
        if msg:
4569
          self.LogWarning("Node failed to demote itself: %s", msg)
4570

    
4571
      new_flags = self._R2F[new_role]
4572
      for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
4573
        if of != nf:
4574
          result.append((desc, str(nf)))
4575
      (node.master_candidate, node.drained, node.offline) = new_flags
4576

    
4577
      # we locked all nodes, we adjust the CP before updating this node
4578
      if self.lock_all:
4579
        _AdjustCandidatePool(self, [node.name])
4580

    
4581
    if self.op.secondary_ip:
4582
      node.secondary_ip = self.op.secondary_ip
4583
      result.append(("secondary_ip", self.op.secondary_ip))
4584

    
4585
    # this will trigger configuration file update, if needed
4586
    self.cfg.Update(node, feedback_fn)
4587

    
4588
    # this will trigger job queue propagation or cleanup if the mc
4589
    # flag changed
4590
    if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
4591
      self.context.ReaddNode(node)
4592

    
4593
    return result
4594

    
4595

    
4596
class LUNodePowercycle(NoHooksLU):
4597
  """Powercycles a node.
4598

4599
  """
4600
  REQ_BGL = False
4601

    
4602
  def CheckArguments(self):
4603
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4604
    if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
4605
      raise errors.OpPrereqError("The node is the master and the force"
4606
                                 " parameter was not set",
4607
                                 errors.ECODE_INVAL)
4608

    
4609
  def ExpandNames(self):
4610
    """Locking for PowercycleNode.
4611

4612
    This is a last-resort option and shouldn't block on other
4613
    jobs. Therefore, we grab no locks.
4614

4615
    """
4616
    self.needed_locks = {}
4617

    
4618
  def Exec(self, feedback_fn):
4619
    """Reboots a node.
4620

4621
    """
4622
    result = self.rpc.call_node_powercycle(self.op.node_name,
4623
                                           self.cfg.GetHypervisorType())
4624
    result.Raise("Failed to schedule the reboot")
4625
    return result.payload
4626

    
4627

    
4628
class LUClusterQuery(NoHooksLU):
4629
  """Query cluster configuration.
4630

4631
  """
4632
  REQ_BGL = False
4633

    
4634
  def ExpandNames(self):
4635
    self.needed_locks = {}
4636

    
4637
  def Exec(self, feedback_fn):
4638
    """Return cluster config.
4639

4640
    """
4641
    cluster = self.cfg.GetClusterInfo()
4642
    os_hvp = {}
4643

    
4644
    # Filter just for enabled hypervisors
4645
    for os_name, hv_dict in cluster.os_hvp.items():
4646
      os_hvp[os_name] = {}
4647
      for hv_name, hv_params in hv_dict.items():
4648
        if hv_name in cluster.enabled_hypervisors:
4649
          os_hvp[os_name][hv_name] = hv_params
4650

    
4651
    # Convert ip_family to ip_version
4652
    primary_ip_version = constants.IP4_VERSION
4653
    if cluster.primary_ip_family == netutils.IP6Address.family:
4654
      primary_ip_version = constants.IP6_VERSION
4655

    
4656
    result = {
4657
      "software_version": constants.RELEASE_VERSION,
4658
      "protocol_version": constants.PROTOCOL_VERSION,
4659
      "config_version": constants.CONFIG_VERSION,
4660
      "os_api_version": max(constants.OS_API_VERSIONS),
4661
      "export_version": constants.EXPORT_VERSION,
4662
      "architecture": (platform.architecture()[0], platform.machine()),
4663
      "name": cluster.cluster_name,
4664
      "master": cluster.master_node,
4665
      "default_hypervisor": cluster.enabled_hypervisors[0],
4666
      "enabled_hypervisors": cluster.enabled_hypervisors,
4667
      "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
4668
                        for hypervisor_name in cluster.enabled_hypervisors]),
4669
      "os_hvp": os_hvp,
4670
      "beparams": cluster.beparams,
4671
      "osparams": cluster.osparams,
4672
      "nicparams": cluster.nicparams,
4673
      "ndparams": cluster.ndparams,
4674
      "candidate_pool_size": cluster.candidate_pool_size,
4675
      "master_netdev": cluster.master_netdev,
4676
      "volume_group_name": cluster.volume_group_name,
4677
      "drbd_usermode_helper": cluster.drbd_usermode_helper,
4678
      "file_storage_dir": cluster.file_storage_dir,
4679
      "shared_file_storage_dir": cluster.shared_file_storage_dir,
4680
      "maintain_node_health": cluster.maintain_node_health,
4681
      "ctime": cluster.ctime,
4682
      "mtime": cluster.mtime,
4683
      "uuid": cluster.uuid,
4684
      "tags": list(cluster.GetTags()),
4685
      "uid_pool": cluster.uid_pool,
4686
      "default_iallocator": cluster.default_iallocator,
4687
      "reserved_lvs": cluster.reserved_lvs,
4688
      "primary_ip_version": primary_ip_version,
4689
      "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
4690
      "hidden_os": cluster.hidden_os,
4691
      "blacklisted_os": cluster.blacklisted_os,
4692
      }
4693

    
4694
    return result
4695

    
4696

    
4697
class LUClusterConfigQuery(NoHooksLU):
4698
  """Return configuration values.
4699

4700
  """
4701
  REQ_BGL = False
4702
  _FIELDS_DYNAMIC = utils.FieldSet()
4703
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag",
4704
                                  "watcher_pause", "volume_group_name")
4705

    
4706
  def CheckArguments(self):
4707
    _CheckOutputFields(static=self._FIELDS_STATIC,
4708
                       dynamic=self._FIELDS_DYNAMIC,
4709
                       selected=self.op.output_fields)
4710

    
4711
  def ExpandNames(self):
4712
    self.needed_locks = {}
4713

    
4714
  def Exec(self, feedback_fn):
4715
    """Dump a representation of the cluster config to the standard output.
4716

4717
    """
4718
    values = []
4719
    for field in self.op.output_fields:
4720
      if field == "cluster_name":
4721
        entry = self.cfg.GetClusterName()
4722
      elif field == "master_node":
4723
        entry = self.cfg.GetMasterNode()
4724
      elif field == "drain_flag":
4725
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
4726
      elif field == "watcher_pause":
4727
        entry = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
4728
      elif field == "volume_group_name":
4729
        entry = self.cfg.GetVGName()
4730
      else:
4731
        raise errors.ParameterError(field)
4732
      values.append(entry)
4733
    return values
4734

    
4735

    
4736
class LUInstanceActivateDisks(NoHooksLU):
4737
  """Bring up an instance's disks.
4738

4739
  """
4740
  REQ_BGL = False
4741

    
4742
  def ExpandNames(self):
4743
    self._ExpandAndLockInstance()
4744
    self.needed_locks[locking.LEVEL_NODE] = []
4745
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4746

    
4747
  def DeclareLocks(self, level):
4748
    if level == locking.LEVEL_NODE:
4749
      self._LockInstancesNodes()
4750

    
4751
  def CheckPrereq(self):
4752
    """Check prerequisites.
4753

4754
    This checks that the instance is in the cluster.
4755

4756
    """
4757
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4758
    assert self.instance is not None, \
4759
      "Cannot retrieve locked instance %s" % self.op.instance_name
4760
    _CheckNodeOnline(self, self.instance.primary_node)
4761

    
4762
  def Exec(self, feedback_fn):
4763
    """Activate the disks.
4764

4765
    """
4766
    disks_ok, disks_info = \
4767
              _AssembleInstanceDisks(self, self.instance,
4768
                                     ignore_size=self.op.ignore_size)
4769
    if not disks_ok:
4770
      raise errors.OpExecError("Cannot activate block devices")
4771

    
4772
    return disks_info
4773

    
4774

    
4775
def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
4776
                           ignore_size=False):
4777
  """Prepare the block devices for an instance.
4778

4779
  This sets up the block devices on all nodes.
4780

4781
  @type lu: L{LogicalUnit}
4782
  @param lu: the logical unit on whose behalf we execute
4783
  @type instance: L{objects.Instance}
4784
  @param instance: the instance for whose disks we assemble
4785
  @type disks: list of L{objects.Disk} or None
4786
  @param disks: which disks to assemble (or all, if None)
4787
  @type ignore_secondaries: boolean
4788
  @param ignore_secondaries: if true, errors on secondary nodes
4789
      won't result in an error return from the function
4790
  @type ignore_size: boolean
4791
  @param ignore_size: if true, the current known size of the disk
4792
      will not be used during the disk activation, useful for cases
4793
      when the size is wrong
4794
  @return: False if the operation failed, otherwise a list of
4795
      (host, instance_visible_name, node_visible_name)
4796
      with the mapping from node devices to instance devices
4797

4798
  """
4799
  device_info = []
4800
  disks_ok = True
4801
  iname = instance.name
4802
  disks = _ExpandCheckDisks(instance, disks)
4803

    
4804
  # With the two passes mechanism we try to reduce the window of
4805
  # opportunity for the race condition of switching DRBD to primary
4806
  # before handshaking occured, but we do not eliminate it
4807

    
4808
  # The proper fix would be to wait (with some limits) until the
4809
  # connection has been made and drbd transitions from WFConnection
4810
  # into any other network-connected state (Connected, SyncTarget,
4811
  # SyncSource, etc.)
4812

    
4813
  # 1st pass, assemble on all nodes in secondary mode
4814
  for idx, inst_disk in enumerate(disks):
4815
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
4816
      if ignore_size:
4817
        node_disk = node_disk.Copy()
4818
        node_disk.UnsetSize()
4819
      lu.cfg.SetDiskID(node_disk, node)
4820
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False, idx)
4821
      msg = result.fail_msg
4822
      if msg:
4823
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
4824
                           " (is_primary=False, pass=1): %s",
4825
                           inst_disk.iv_name, node, msg)
4826
        if not ignore_secondaries:
4827
          disks_ok = False
4828

    
4829
  # FIXME: race condition on drbd migration to primary
4830

    
4831
  # 2nd pass, do only the primary node
4832
  for idx, inst_disk in enumerate(disks):
4833
    dev_path = None
4834

    
4835
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
4836
      if node != instance.primary_node:
4837
        continue
4838
      if ignore_size:
4839
        node_disk = node_disk.Copy()
4840
        node_disk.UnsetSize()
4841
      lu.cfg.SetDiskID(node_disk, node)
4842
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True, idx)
4843
      msg = result.fail_msg
4844
      if msg:
4845
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
4846
                           " (is_primary=True, pass=2): %s",
4847
                           inst_disk.iv_name, node, msg)
4848
        disks_ok = False
4849
      else:
4850
        dev_path = result.payload
4851

    
4852
    device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
4853

    
4854
  # leave the disks configured for the primary node
4855
  # this is a workaround that would be fixed better by
4856
  # improving the logical/physical id handling
4857
  for disk in disks:
4858
    lu.cfg.SetDiskID(disk, instance.primary_node)
4859

    
4860
  return disks_ok, device_info
4861

    
4862

    
4863
def _StartInstanceDisks(lu, instance, force):
4864
  """Start the disks of an instance.
4865

4866
  """
4867
  disks_ok, _ = _AssembleInstanceDisks(lu, instance,
4868
                                           ignore_secondaries=force)
4869
  if not disks_ok:
4870
    _ShutdownInstanceDisks(lu, instance)
4871
    if force is not None and not force:
4872
      lu.proc.LogWarning("", hint="If the message above refers to a"
4873
                         " secondary node,"
4874
                         " you can retry the operation using '--force'.")
4875
    raise errors.OpExecError("Disk consistency error")
4876

    
4877

    
4878
class LUInstanceDeactivateDisks(NoHooksLU):
4879
  """Shutdown an instance's disks.
4880

4881
  """
4882
  REQ_BGL = False
4883

    
4884
  def ExpandNames(self):
4885
    self._ExpandAndLockInstance()
4886
    self.needed_locks[locking.LEVEL_NODE] = []
4887
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4888

    
4889
  def DeclareLocks(self, level):
4890
    if level == locking.LEVEL_NODE:
4891
      self._LockInstancesNodes()
4892

    
4893
  def CheckPrereq(self):
4894
    """Check prerequisites.
4895

4896
    This checks that the instance is in the cluster.
4897

4898
    """
4899
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4900
    assert self.instance is not None, \
4901
      "Cannot retrieve locked instance %s" % self.op.instance_name
4902

    
4903
  def Exec(self, feedback_fn):
4904
    """Deactivate the disks
4905

4906
    """
4907
    instance = self.instance
4908
    if self.op.force:
4909
      _ShutdownInstanceDisks(self, instance)
4910
    else:
4911
      _SafeShutdownInstanceDisks(self, instance)
4912

    
4913

    
4914
def _SafeShutdownInstanceDisks(lu, instance, disks=None):
4915
  """Shutdown block devices of an instance.
4916

4917
  This function checks if an instance is running, before calling
4918
  _ShutdownInstanceDisks.
4919

4920
  """
4921
  _CheckInstanceDown(lu, instance, "cannot shutdown disks")
4922
  _ShutdownInstanceDisks(lu, instance, disks=disks)
4923

    
4924

    
4925
def _ExpandCheckDisks(instance, disks):
4926
  """Return the instance disks selected by the disks list
4927

4928
  @type disks: list of L{objects.Disk} or None
4929
  @param disks: selected disks
4930
  @rtype: list of L{objects.Disk}
4931
  @return: selected instance disks to act on
4932

4933
  """
4934
  if disks is None:
4935
    return instance.disks
4936
  else:
4937
    if not set(disks).issubset(instance.disks):
4938
      raise errors.ProgrammerError("Can only act on disks belonging to the"
4939
                                   " target instance")
4940
    return disks
4941

    
4942

    
4943
def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
4944
  """Shutdown block devices of an instance.
4945

4946
  This does the shutdown on all nodes of the instance.
4947

4948
  If the ignore_primary is false, errors on the primary node are
4949
  ignored.
4950

4951
  """
4952
  all_result = True
4953
  disks = _ExpandCheckDisks(instance, disks)
4954

    
4955
  for disk in disks:
4956
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
4957
      lu.cfg.SetDiskID(top_disk, node)
4958
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
4959
      msg = result.fail_msg
4960
      if msg:
4961
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
4962
                      disk.iv_name, node, msg)
4963
        if ((node == instance.primary_node and not ignore_primary) or
4964
            (node != instance.primary_node and not result.offline)):
4965
          all_result = False
4966
  return all_result
4967

    
4968

    
4969
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
4970
  """Checks if a node has enough free memory.
4971

4972
  This function check if a given node has the needed amount of free
4973
  memory. In case the node has less memory or we cannot get the
4974
  information from the node, this function raise an OpPrereqError
4975
  exception.
4976

4977
  @type lu: C{LogicalUnit}
4978
  @param lu: a logical unit from which we get configuration data
4979
  @type node: C{str}
4980
  @param node: the node to check
4981
  @type reason: C{str}
4982
  @param reason: string to use in the error message
4983
  @type requested: C{int}
4984
  @param requested: the amount of memory in MiB to check for
4985
  @type hypervisor_name: C{str}
4986
  @param hypervisor_name: the hypervisor to ask for memory stats
4987
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
4988
      we cannot check the node
4989

4990
  """
4991
  nodeinfo = lu.rpc.call_node_info([node], None, hypervisor_name)
4992
  nodeinfo[node].Raise("Can't get data from node %s" % node,
4993
                       prereq=True, ecode=errors.ECODE_ENVIRON)
4994
  free_mem = nodeinfo[node].payload.get('memory_free', None)
4995
  if not isinstance(free_mem, int):
4996
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
4997
                               " was '%s'" % (node, free_mem),
4998
                               errors.ECODE_ENVIRON)
4999
  if requested > free_mem:
5000
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
5001
                               " needed %s MiB, available %s MiB" %
5002
                               (node, reason, requested, free_mem),
5003
                               errors.ECODE_NORES)
5004

    
5005

    
5006
def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
5007
  """Checks if nodes have enough free disk space in the all VGs.
5008

5009
  This function check if all given nodes have the needed amount of
5010
  free disk. In case any node has less disk or we cannot get the
5011
  information from the node, this function raise an OpPrereqError
5012
  exception.
5013

5014
  @type lu: C{LogicalUnit}
5015
  @param lu: a logical unit from which we get configuration data
5016
  @type nodenames: C{list}
5017
  @param nodenames: the list of node names to check
5018
  @type req_sizes: C{dict}
5019
  @param req_sizes: the hash of vg and corresponding amount of disk in
5020
      MiB to check for
5021
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5022
      or we cannot check the node
5023

5024
  """
5025
  for vg, req_size in req_sizes.items():
5026
    _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
5027

    
5028

    
5029
def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
5030
  """Checks if nodes have enough free disk space in the specified VG.
5031

5032
  This function check if all given nodes have the needed amount of
5033
  free disk. In case any node has less disk or we cannot get the
5034
  information from the node, this function raise an OpPrereqError
5035
  exception.
5036

5037
  @type lu: C{LogicalUnit}
5038
  @param lu: a logical unit from which we get configuration data
5039
  @type nodenames: C{list}
5040
  @param nodenames: the list of node names to check
5041
  @type vg: C{str}
5042
  @param vg: the volume group to check
5043
  @type requested: C{int}
5044
  @param requested: the amount of disk in MiB to check for
5045
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5046
      or we cannot check the node
5047

5048
  """
5049
  nodeinfo = lu.rpc.call_node_info(nodenames, vg, None)
5050
  for node in nodenames:
5051
    info = nodeinfo[node]
5052
    info.Raise("Cannot get current information from node %s" % node,
5053
               prereq=True, ecode=errors.ECODE_ENVIRON)
5054
    vg_free = info.payload.get("vg_free", None)
5055
    if not isinstance(vg_free, int):
5056
      raise errors.OpPrereqError("Can't compute free disk space on node"
5057
                                 " %s for vg %s, result was '%s'" %
5058
                                 (node, vg, vg_free), errors.ECODE_ENVIRON)
5059
    if requested > vg_free:
5060
      raise errors.OpPrereqError("Not enough disk space on target node %s"
5061
                                 " vg %s: required %d MiB, available %d MiB" %
5062
                                 (node, vg, requested, vg_free),
5063
                                 errors.ECODE_NORES)
5064

    
5065

    
5066
class LUInstanceStartup(LogicalUnit):
5067
  """Starts an instance.
5068

5069
  """
5070
  HPATH = "instance-start"
5071
  HTYPE = constants.HTYPE_INSTANCE
5072
  REQ_BGL = False
5073

    
5074
  def CheckArguments(self):
5075
    # extra beparams
5076
    if self.op.beparams:
5077
      # fill the beparams dict
5078
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
5079

    
5080
  def ExpandNames(self):
5081
    self._ExpandAndLockInstance()
5082

    
5083
  def BuildHooksEnv(self):
5084
    """Build hooks env.
5085

5086
    This runs on master, primary and secondary nodes of the instance.
5087

5088
    """
5089
    env = {
5090
      "FORCE": self.op.force,
5091
      }
5092
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5093
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5094
    return env, nl, nl
5095

    
5096
  def CheckPrereq(self):
5097
    """Check prerequisites.
5098

5099
    This checks that the instance is in the cluster.
5100

5101
    """
5102
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5103
    assert self.instance is not None, \
5104
      "Cannot retrieve locked instance %s" % self.op.instance_name
5105

    
5106
    # extra hvparams
5107
    if self.op.hvparams:
5108
      # check hypervisor parameter syntax (locally)
5109
      cluster = self.cfg.GetClusterInfo()
5110
      utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
5111
      filled_hvp = cluster.FillHV(instance)
5112
      filled_hvp.update(self.op.hvparams)
5113
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
5114
      hv_type.CheckParameterSyntax(filled_hvp)
5115
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
5116

    
5117
    self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
5118

    
5119
    if self.primary_offline and self.op.ignore_offline_nodes:
5120
      self.proc.LogWarning("Ignoring offline primary node")
5121

    
5122
      if self.op.hvparams or self.op.beparams:
5123
        self.proc.LogWarning("Overridden parameters are ignored")
5124
    else:
5125
      _CheckNodeOnline(self, instance.primary_node)
5126

    
5127
      bep = self.cfg.GetClusterInfo().FillBE(instance)
5128

    
5129
      # check bridges existence
5130
      _CheckInstanceBridgesExist(self, instance)
5131

    
5132
      remote_info = self.rpc.call_instance_info(instance.primary_node,
5133
                                                instance.name,
5134
                                                instance.hypervisor)
5135
      remote_info.Raise("Error checking node %s" % instance.primary_node,
5136
                        prereq=True, ecode=errors.ECODE_ENVIRON)
5137
      if not remote_info.payload: # not running already
5138
        _CheckNodeFreeMemory(self, instance.primary_node,
5139
                             "starting instance %s" % instance.name,
5140
                             bep[constants.BE_MEMORY], instance.hypervisor)
5141

    
5142
  def Exec(self, feedback_fn):
5143
    """Start the instance.
5144

5145
    """
5146
    instance = self.instance
5147
    force = self.op.force
5148

    
5149
    self.cfg.MarkInstanceUp(instance.name)
5150

    
5151
    if self.primary_offline:
5152
      assert self.op.ignore_offline_nodes
5153
      self.proc.LogInfo("Primary node offline, marked instance as started")
5154
    else:
5155
      node_current = instance.primary_node
5156

    
5157
      _StartInstanceDisks(self, instance, force)
5158

    
5159
      result = self.rpc.call_instance_start(node_current, instance,
5160
                                            self.op.hvparams, self.op.beparams)
5161
      msg = result.fail_msg
5162
      if msg:
5163
        _ShutdownInstanceDisks(self, instance)
5164
        raise errors.OpExecError("Could not start instance: %s" % msg)
5165

    
5166

    
5167
class LUInstanceReboot(LogicalUnit):
5168
  """Reboot an instance.
5169

5170
  """
5171
  HPATH = "instance-reboot"
5172
  HTYPE = constants.HTYPE_INSTANCE
5173
  REQ_BGL = False
5174

    
5175
  def ExpandNames(self):
5176
    self._ExpandAndLockInstance()
5177

    
5178
  def BuildHooksEnv(self):
5179
    """Build hooks env.
5180

5181
    This runs on master, primary and secondary nodes of the instance.
5182

5183
    """
5184
    env = {
5185
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
5186
      "REBOOT_TYPE": self.op.reboot_type,
5187
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5188
      }
5189
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5190
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5191
    return env, nl, nl
5192

    
5193
  def CheckPrereq(self):
5194
    """Check prerequisites.
5195

5196
    This checks that the instance is in the cluster.
5197

5198
    """
5199
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5200
    assert self.instance is not None, \
5201
      "Cannot retrieve locked instance %s" % self.op.instance_name
5202

    
5203
    _CheckNodeOnline(self, instance.primary_node)
5204

    
5205
    # check bridges existence
5206
    _CheckInstanceBridgesExist(self, instance)
5207

    
5208
  def Exec(self, feedback_fn):
5209
    """Reboot the instance.
5210

5211
    """
5212
    instance = self.instance
5213
    ignore_secondaries = self.op.ignore_secondaries
5214
    reboot_type = self.op.reboot_type
5215

    
5216
    node_current = instance.primary_node
5217

    
5218
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
5219
                       constants.INSTANCE_REBOOT_HARD]:
5220
      for disk in instance.disks:
5221
        self.cfg.SetDiskID(disk, node_current)
5222
      result = self.rpc.call_instance_reboot(node_current, instance,
5223
                                             reboot_type,
5224
                                             self.op.shutdown_timeout)
5225
      result.Raise("Could not reboot instance")
5226
    else:
5227
      result = self.rpc.call_instance_shutdown(node_current, instance,
5228
                                               self.op.shutdown_timeout)
5229
      result.Raise("Could not shutdown instance for full reboot")
5230
      _ShutdownInstanceDisks(self, instance)
5231
      _StartInstanceDisks(self, instance, ignore_secondaries)
5232
      result = self.rpc.call_instance_start(node_current, instance, None, None)
5233
      msg = result.fail_msg
5234
      if msg:
5235
        _ShutdownInstanceDisks(self, instance)
5236
        raise errors.OpExecError("Could not start instance for"
5237
                                 " full reboot: %s" % msg)
5238

    
5239
    self.cfg.MarkInstanceUp(instance.name)
5240

    
5241

    
5242
class LUInstanceShutdown(LogicalUnit):
5243
  """Shutdown an instance.
5244

5245
  """
5246
  HPATH = "instance-stop"
5247
  HTYPE = constants.HTYPE_INSTANCE
5248
  REQ_BGL = False
5249

    
5250
  def ExpandNames(self):
5251
    self._ExpandAndLockInstance()
5252

    
5253
  def BuildHooksEnv(self):
5254
    """Build hooks env.
5255

5256
    This runs on master, primary and secondary nodes of the instance.
5257

5258
    """
5259
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5260
    env["TIMEOUT"] = self.op.timeout
5261
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5262
    return env, nl, nl
5263

    
5264
  def CheckPrereq(self):
5265
    """Check prerequisites.
5266

5267
    This checks that the instance is in the cluster.
5268

5269
    """
5270
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5271
    assert self.instance is not None, \
5272
      "Cannot retrieve locked instance %s" % self.op.instance_name
5273

    
5274
    self.primary_offline = \
5275
      self.cfg.GetNodeInfo(self.instance.primary_node).offline
5276

    
5277
    if self.primary_offline and self.op.ignore_offline_nodes:
5278
      self.proc.LogWarning("Ignoring offline primary node")
5279
    else:
5280
      _CheckNodeOnline(self, self.instance.primary_node)
5281

    
5282
  def Exec(self, feedback_fn):
5283
    """Shutdown the instance.
5284

5285
    """
5286
    instance = self.instance
5287
    node_current = instance.primary_node
5288
    timeout = self.op.timeout
5289

    
5290
    self.cfg.MarkInstanceDown(instance.name)
5291

    
5292
    if self.primary_offline:
5293
      assert self.op.ignore_offline_nodes
5294
      self.proc.LogInfo("Primary node offline, marked instance as stopped")
5295
    else:
5296
      result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
5297
      msg = result.fail_msg
5298
      if msg:
5299
        self.proc.LogWarning("Could not shutdown instance: %s" % msg)
5300

    
5301
      _ShutdownInstanceDisks(self, instance)
5302

    
5303

    
5304
class LUInstanceReinstall(LogicalUnit):
5305
  """Reinstall an instance.
5306

5307
  """
5308
  HPATH = "instance-reinstall"
5309
  HTYPE = constants.HTYPE_INSTANCE
5310
  REQ_BGL = False
5311

    
5312
  def ExpandNames(self):
5313
    self._ExpandAndLockInstance()
5314

    
5315
  def BuildHooksEnv(self):
5316
    """Build hooks env.
5317

5318
    This runs on master, primary and secondary nodes of the instance.
5319

5320
    """
5321
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5322
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5323
    return env, nl, nl
5324

    
5325
  def CheckPrereq(self):
5326
    """Check prerequisites.
5327

5328
    This checks that the instance is in the cluster and is not running.
5329

5330
    """
5331
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5332
    assert instance is not None, \
5333
      "Cannot retrieve locked instance %s" % self.op.instance_name
5334
    _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
5335
                     " offline, cannot reinstall")
5336
    for node in instance.secondary_nodes:
5337
      _CheckNodeOnline(self, node, "Instance secondary node offline,"
5338
                       " cannot reinstall")
5339

    
5340
    if instance.disk_template == constants.DT_DISKLESS:
5341
      raise errors.OpPrereqError("Instance '%s' has no disks" %
5342
                                 self.op.instance_name,
5343
                                 errors.ECODE_INVAL)
5344
    _CheckInstanceDown(self, instance, "cannot reinstall")
5345

    
5346
    if self.op.os_type is not None:
5347
      # OS verification
5348
      pnode = _ExpandNodeName(self.cfg, instance.primary_node)
5349
      _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
5350
      instance_os = self.op.os_type
5351
    else:
5352
      instance_os = instance.os
5353

    
5354
    nodelist = list(instance.all_nodes)
5355

    
5356
    if self.op.osparams:
5357
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
5358
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
5359
      self.os_inst = i_osdict # the new dict (without defaults)
5360
    else:
5361
      self.os_inst = None
5362

    
5363
    self.instance = instance
5364

    
5365
  def Exec(self, feedback_fn):
5366
    """Reinstall the instance.
5367

5368
    """
5369
    inst = self.instance
5370

    
5371
    if self.op.os_type is not None:
5372
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
5373
      inst.os = self.op.os_type
5374
      # Write to configuration
5375
      self.cfg.Update(inst, feedback_fn)
5376

    
5377
    _StartInstanceDisks(self, inst, None)
5378
    try:
5379
      feedback_fn("Running the instance OS create scripts...")
5380
      # FIXME: pass debug option from opcode to backend
5381
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True,
5382
                                             self.op.debug_level,
5383
                                             osparams=self.os_inst)
5384
      result.Raise("Could not install OS for instance %s on node %s" %
5385
                   (inst.name, inst.primary_node))
5386
    finally:
5387
      _ShutdownInstanceDisks(self, inst)
5388

    
5389

    
5390
class LUInstanceRecreateDisks(LogicalUnit):
5391
  """Recreate an instance's missing disks.
5392

5393
  """
5394
  HPATH = "instance-recreate-disks"
5395
  HTYPE = constants.HTYPE_INSTANCE
5396
  REQ_BGL = False
5397

    
5398
  def ExpandNames(self):
5399
    self._ExpandAndLockInstance()
5400

    
5401
  def BuildHooksEnv(self):
5402
    """Build hooks env.
5403

5404
    This runs on master, primary and secondary nodes of the instance.
5405

5406
    """
5407
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5408
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5409
    return env, nl, nl
5410

    
5411
  def CheckPrereq(self):
5412
    """Check prerequisites.
5413

5414
    This checks that the instance is in the cluster and is not running.
5415

5416
    """
5417
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5418
    assert instance is not None, \
5419
      "Cannot retrieve locked instance %s" % self.op.instance_name
5420
    _CheckNodeOnline(self, instance.primary_node)
5421

    
5422
    if instance.disk_template == constants.DT_DISKLESS:
5423
      raise errors.OpPrereqError("Instance '%s' has no disks" %
5424
                                 self.op.instance_name, errors.ECODE_INVAL)
5425
    _CheckInstanceDown(self, instance, "cannot recreate disks")
5426

    
5427
    if not self.op.disks:
5428
      self.op.disks = range(len(instance.disks))
5429
    else:
5430
      for idx in self.op.disks:
5431
        if idx >= len(instance.disks):
5432
          raise errors.OpPrereqError("Invalid disk index passed '%s'" % idx,
5433
                                     errors.ECODE_INVAL)
5434

    
5435
    self.instance = instance
5436

    
5437
  def Exec(self, feedback_fn):
5438
    """Recreate the disks.
5439

5440
    """
5441
    to_skip = []
5442
    for idx, _ in enumerate(self.instance.disks):
5443
      if idx not in self.op.disks: # disk idx has not been passed in
5444
        to_skip.append(idx)
5445
        continue
5446

    
5447
    _CreateDisks(self, self.instance, to_skip=to_skip)
5448

    
5449

    
5450
class LUInstanceRename(LogicalUnit):
5451
  """Rename an instance.
5452

5453
  """
5454
  HPATH = "instance-rename"
5455
  HTYPE = constants.HTYPE_INSTANCE
5456

    
5457
  def CheckArguments(self):
5458
    """Check arguments.
5459

5460
    """
5461
    if self.op.ip_check and not self.op.name_check:
5462
      # TODO: make the ip check more flexible and not depend on the name check
5463
      raise errors.OpPrereqError("Cannot do ip check without a name check",
5464
                                 errors.ECODE_INVAL)
5465

    
5466
  def BuildHooksEnv(self):
5467
    """Build hooks env.
5468

5469
    This runs on master, primary and secondary nodes of the instance.
5470

5471
    """
5472
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5473
    env["INSTANCE_NEW_NAME"] = self.op.new_name
5474
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5475
    return env, nl, nl
5476

    
5477
  def CheckPrereq(self):
5478
    """Check prerequisites.
5479

5480
    This checks that the instance is in the cluster and is not running.
5481

5482
    """
5483
    self.op.instance_name = _ExpandInstanceName(self.cfg,
5484
                                                self.op.instance_name)
5485
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5486
    assert instance is not None
5487
    _CheckNodeOnline(self, instance.primary_node)
5488
    _CheckInstanceDown(self, instance, "cannot rename")
5489
    self.instance = instance
5490

    
5491
    new_name = self.op.new_name
5492
    if self.op.name_check:
5493
      hostname = netutils.GetHostname(name=new_name)
5494
      self.LogInfo("Resolved given name '%s' to '%s'", new_name,
5495
                   hostname.name)
5496
      new_name = self.op.new_name = hostname.name
5497
      if (self.op.ip_check and
5498
          netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
5499
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
5500
                                   (hostname.ip, new_name),
5501
                                   errors.ECODE_NOTUNIQUE)
5502

    
5503
    instance_list = self.cfg.GetInstanceList()
5504
    if new_name in instance_list and new_name != instance.name:
5505
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
5506
                                 new_name, errors.ECODE_EXISTS)
5507

    
5508
  def Exec(self, feedback_fn):
5509
    """Rename the instance.
5510

5511
    """
5512
    inst = self.instance
5513
    old_name = inst.name
5514

    
5515
    rename_file_storage = False
5516
    if (inst.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE) and
5517
        self.op.new_name != inst.name):
5518
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
5519
      rename_file_storage = True
5520

    
5521
    self.cfg.RenameInstance(inst.name, self.op.new_name)
5522
    # Change the instance lock. This is definitely safe while we hold the BGL
5523
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
5524
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
5525

    
5526
    # re-read the instance from the configuration after rename
5527
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
5528

    
5529
    if rename_file_storage:
5530
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
5531
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
5532
                                                     old_file_storage_dir,
5533
                                                     new_file_storage_dir)
5534
      result.Raise("Could not rename on node %s directory '%s' to '%s'"
5535
                   " (but the instance has been renamed in Ganeti)" %
5536
                   (inst.primary_node, old_file_storage_dir,
5537
                    new_file_storage_dir))
5538

    
5539
    _StartInstanceDisks(self, inst, None)
5540
    try:
5541
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
5542
                                                 old_name, self.op.debug_level)
5543
      msg = result.fail_msg
5544
      if msg:
5545
        msg = ("Could not run OS rename script for instance %s on node %s"
5546
               " (but the instance has been renamed in Ganeti): %s" %
5547
               (inst.name, inst.primary_node, msg))
5548
        self.proc.LogWarning(msg)
5549
    finally:
5550
      _ShutdownInstanceDisks(self, inst)
5551

    
5552
    return inst.name
5553

    
5554

    
5555
class LUInstanceRemove(LogicalUnit):
5556
  """Remove an instance.
5557

5558
  """
5559
  HPATH = "instance-remove"
5560
  HTYPE = constants.HTYPE_INSTANCE
5561
  REQ_BGL = False
5562

    
5563
  def ExpandNames(self):
5564
    self._ExpandAndLockInstance()
5565
    self.needed_locks[locking.LEVEL_NODE] = []
5566
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5567

    
5568
  def DeclareLocks(self, level):
5569
    if level == locking.LEVEL_NODE:
5570
      self._LockInstancesNodes()
5571

    
5572
  def BuildHooksEnv(self):
5573
    """Build hooks env.
5574

5575
    This runs on master, primary and secondary nodes of the instance.
5576

5577
    """
5578
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5579
    env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
5580
    nl = [self.cfg.GetMasterNode()]
5581
    nl_post = list(self.instance.all_nodes) + nl
5582
    return env, nl, nl_post
5583

    
5584
  def CheckPrereq(self):
5585
    """Check prerequisites.
5586

5587
    This checks that the instance is in the cluster.
5588

5589
    """
5590
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5591
    assert self.instance is not None, \
5592
      "Cannot retrieve locked instance %s" % self.op.instance_name
5593

    
5594
  def Exec(self, feedback_fn):
5595
    """Remove the instance.
5596

5597
    """
5598
    instance = self.instance
5599
    logging.info("Shutting down instance %s on node %s",
5600
                 instance.name, instance.primary_node)
5601

    
5602
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
5603
                                             self.op.shutdown_timeout)
5604
    msg = result.fail_msg
5605
    if msg:
5606
      if self.op.ignore_failures:
5607
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
5608
      else:
5609
        raise errors.OpExecError("Could not shutdown instance %s on"
5610
                                 " node %s: %s" %
5611
                                 (instance.name, instance.primary_node, msg))
5612

    
5613
    _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
5614

    
5615

    
5616
def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
5617
  """Utility function to remove an instance.
5618

5619
  """
5620
  logging.info("Removing block devices for instance %s", instance.name)
5621

    
5622
  if not _RemoveDisks(lu, instance):
5623
    if not ignore_failures:
5624
      raise errors.OpExecError("Can't remove instance's disks")
5625
    feedback_fn("Warning: can't remove instance's disks")
5626

    
5627
  logging.info("Removing instance %s out of cluster config", instance.name)
5628

    
5629
  lu.cfg.RemoveInstance(instance.name)
5630

    
5631
  assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
5632
    "Instance lock removal conflict"
5633

    
5634
  # Remove lock for the instance
5635
  lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
5636

    
5637

    
5638
class LUInstanceQuery(NoHooksLU):
5639
  """Logical unit for querying instances.
5640

5641
  """
5642
  # pylint: disable-msg=W0142
5643
  REQ_BGL = False
5644

    
5645
  def CheckArguments(self):
5646
    self.iq = _InstanceQuery(self.op.names, self.op.output_fields,
5647
                             self.op.use_locking)
5648

    
5649
  def ExpandNames(self):
5650
    self.iq.ExpandNames(self)
5651

    
5652
  def DeclareLocks(self, level):
5653
    self.iq.DeclareLocks(self, level)
5654

    
5655
  def Exec(self, feedback_fn):
5656
    return self.iq.OldStyleQuery(self)
5657

    
5658

    
5659
class LUInstanceFailover(LogicalUnit):
5660
  """Failover an instance.
5661

5662
  """
5663
  HPATH = "instance-failover"
5664
  HTYPE = constants.HTYPE_INSTANCE
5665
  REQ_BGL = False
5666

    
5667
  def CheckArguments(self):
5668
    """Check the arguments.
5669

5670
    """
5671
    self.iallocator = getattr(self.op, "iallocator", None)
5672
    self.target_node = getattr(self.op, "target_node", None)
5673

    
5674
  def ExpandNames(self):
5675
    self._ExpandAndLockInstance()
5676

    
5677
    if self.op.target_node is not None:
5678
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
5679

    
5680
    self.needed_locks[locking.LEVEL_NODE] = []
5681
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5682

    
5683
  def DeclareLocks(self, level):
5684
    if level == locking.LEVEL_NODE:
5685
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
5686
      if instance.disk_template in constants.DTS_EXT_MIRROR:
5687
        if self.op.target_node is None:
5688
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5689
        else:
5690
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
5691
                                                   self.op.target_node]
5692
        del self.recalculate_locks[locking.LEVEL_NODE]
5693
      else:
5694
        self._LockInstancesNodes()
5695

    
5696
  def BuildHooksEnv(self):
5697
    """Build hooks env.
5698

5699
    This runs on master, primary and secondary nodes of the instance.
5700

5701
    """
5702
    instance = self.instance
5703
    source_node = instance.primary_node
5704
    env = {
5705
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
5706
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5707
      "OLD_PRIMARY": source_node,
5708
      "NEW_PRIMARY": self.op.target_node,
5709
      }
5710

    
5711
    if instance.disk_template in constants.DTS_INT_MIRROR:
5712
      env["OLD_SECONDARY"] = instance.secondary_nodes[0]
5713
      env["NEW_SECONDARY"] = source_node
5714
    else:
5715
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = ""
5716

    
5717
    env.update(_BuildInstanceHookEnvByObject(self, instance))
5718
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
5719
    nl_post = list(nl)
5720
    nl_post.append(source_node)
5721
    return env, nl, nl_post
5722

    
5723
  def CheckPrereq(self):
5724
    """Check prerequisites.
5725

5726
    This checks that the instance is in the cluster.
5727

5728
    """
5729
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5730
    assert self.instance is not None, \
5731
      "Cannot retrieve locked instance %s" % self.op.instance_name
5732

    
5733
    bep = self.cfg.GetClusterInfo().FillBE(instance)
5734
    if instance.disk_template not in constants.DTS_MIRRORED:
5735
      raise errors.OpPrereqError("Instance's disk layout is not"
5736
                                 " mirrored, cannot failover.",
5737
                                 errors.ECODE_STATE)
5738

    
5739
    if instance.disk_template in constants.DTS_EXT_MIRROR:
5740
      _CheckIAllocatorOrNode(self, "iallocator", "target_node")
5741
      if self.op.iallocator:
5742
        self._RunAllocator()
5743
        # Release all unnecessary node locks
5744
        nodes_keep = [instance.primary_node, self.op.target_node]
5745
        nodes_rel = [node for node in self.acquired_locks[locking.LEVEL_NODE]
5746
                     if node not in nodes_keep]
5747
        self.context.glm.release(locking.LEVEL_NODE, nodes_rel)
5748
        self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
5749

    
5750
      # self.op.target_node is already populated, either directly or by the
5751
      # iallocator run
5752
      target_node = self.op.target_node
5753

    
5754
    else:
5755
      secondary_nodes = instance.secondary_nodes
5756
      if not secondary_nodes:
5757
        raise errors.ConfigurationError("No secondary node but using"
5758
                                        " %s disk template" %
5759
                                        instance.disk_template)
5760
      target_node = secondary_nodes[0]
5761

    
5762
      if self.op.iallocator or (self.op.target_node and
5763
                                self.op.target_node != target_node):
5764
        raise errors.OpPrereqError("Instances with disk template %s cannot"
5765
                                   " be failed over to arbitrary nodes"
5766
                                   " (neither an iallocator nor a target"
5767
                                   " node can be passed)" %
5768
                                   instance.disk_template, errors.ECODE_INVAL)
5769
    _CheckNodeOnline(self, target_node)
5770
    _CheckNodeNotDrained(self, target_node)
5771

    
5772
    # Save target_node so that we can use it in BuildHooksEnv
5773
    self.op.target_node = target_node
5774

    
5775
    if instance.admin_up:
5776
      # check memory requirements on the secondary node
5777
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
5778
                           instance.name, bep[constants.BE_MEMORY],
5779
                           instance.hypervisor)
5780
    else:
5781
      self.LogInfo("Not checking memory on the secondary node as"
5782
                   " instance will not be started")
5783

    
5784
    # check bridge existance
5785
    _CheckInstanceBridgesExist(self, instance, node=target_node)
5786

    
5787
  def Exec(self, feedback_fn):
5788
    """Failover an instance.
5789

5790
    The failover is done by shutting it down on its present node and
5791
    starting it on the secondary.
5792

5793
    """
5794
    instance = self.instance
5795
    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
5796

    
5797
    source_node = instance.primary_node
5798
    target_node = self.op.target_node
5799

    
5800
    if instance.admin_up:
5801
      feedback_fn("* checking disk consistency between source and target")
5802
      for dev in instance.disks:
5803
        # for drbd, these are drbd over lvm
5804
        if not _CheckDiskConsistency(self, dev, target_node, False):
5805
          if not self.op.ignore_consistency:
5806
            raise errors.OpExecError("Disk %s is degraded on target node,"
5807
                                     " aborting failover." % dev.iv_name)
5808
    else:
5809
      feedback_fn("* not checking disk consistency as instance is not running")
5810

    
5811
    feedback_fn("* shutting down instance on source node")
5812
    logging.info("Shutting down instance %s on node %s",
5813
                 instance.name, source_node)
5814

    
5815
    result = self.rpc.call_instance_shutdown(source_node, instance,
5816
                                             self.op.shutdown_timeout)
5817
    msg = result.fail_msg
5818
    if msg:
5819
      if self.op.ignore_consistency or primary_node.offline:
5820
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
5821
                             " Proceeding anyway. Please make sure node"
5822
                             " %s is down. Error details: %s",
5823
                             instance.name, source_node, source_node, msg)
5824
      else:
5825
        raise errors.OpExecError("Could not shutdown instance %s on"
5826
                                 " node %s: %s" %
5827
                                 (instance.name, source_node, msg))
5828

    
5829
    feedback_fn("* deactivating the instance's disks on source node")
5830
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
5831
      raise errors.OpExecError("Can't shut down the instance's disks.")
5832

    
5833
    instance.primary_node = target_node
5834
    # distribute new instance config to the other nodes
5835
    self.cfg.Update(instance, feedback_fn)
5836

    
5837
    # Only start the instance if it's marked as up
5838
    if instance.admin_up:
5839
      feedback_fn("* activating the instance's disks on target node")
5840
      logging.info("Starting instance %s on node %s",
5841
                   instance.name, target_node)
5842

    
5843
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
5844
                                           ignore_secondaries=True)
5845
      if not disks_ok:
5846
        _ShutdownInstanceDisks(self, instance)
5847
        raise errors.OpExecError("Can't activate the instance's disks")
5848

    
5849
      feedback_fn("* starting the instance on the target node")
5850
      result = self.rpc.call_instance_start(target_node, instance, None, None)
5851
      msg = result.fail_msg
5852
      if msg:
5853
        _ShutdownInstanceDisks(self, instance)
5854
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
5855
                                 (instance.name, target_node, msg))
5856

    
5857
  def _RunAllocator(self):
5858
    """Run the allocator based on input opcode.
5859

5860
    """
5861
    ial = IAllocator(self.cfg, self.rpc,
5862
                     mode=constants.IALLOCATOR_MODE_RELOC,
5863
                     name=self.instance.name,
5864
                     # TODO See why hail breaks with a single node below
5865
                     relocate_from=[self.instance.primary_node,
5866
                                    self.instance.primary_node],
5867
                     )
5868

    
5869
    ial.Run(self.op.iallocator)
5870

    
5871
    if not ial.success:
5872
      raise errors.OpPrereqError("Can't compute nodes using"
5873
                                 " iallocator '%s': %s" %
5874
                                 (self.op.iallocator, ial.info),
5875
                                 errors.ECODE_NORES)
5876
    if len(ial.result) != ial.required_nodes:
5877
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
5878
                                 " of nodes (%s), required %s" %
5879
                                 (self.op.iallocator, len(ial.result),
5880
                                  ial.required_nodes), errors.ECODE_FAULT)
5881
    self.op.target_node = ial.result[0]
5882
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
5883
                 self.instance.name, self.op.iallocator,
5884
                 utils.CommaJoin(ial.result))
5885

    
5886

    
5887
class LUInstanceMigrate(LogicalUnit):
5888
  """Migrate an instance.
5889

5890
  This is migration without shutting down, compared to the failover,
5891
  which is done with shutdown.
5892

5893
  """
5894
  HPATH = "instance-migrate"
5895
  HTYPE = constants.HTYPE_INSTANCE
5896
  REQ_BGL = False
5897

    
5898
  def ExpandNames(self):
5899
    self._ExpandAndLockInstance()
5900

    
5901
    if self.op.target_node is not None:
5902
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
5903

    
5904
    self.needed_locks[locking.LEVEL_NODE] = []
5905
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5906

    
5907
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
5908
                                       self.op.cleanup)
5909
    self.tasklets = [self._migrater]
5910

    
5911
  def DeclareLocks(self, level):
5912
    if level == locking.LEVEL_NODE:
5913
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
5914
      if instance.disk_template in constants.DTS_EXT_MIRROR:
5915
        if self.op.target_node is None:
5916
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5917
        else:
5918
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
5919
                                                   self.op.target_node]
5920
        del self.recalculate_locks[locking.LEVEL_NODE]
5921
      else:
5922
        self._LockInstancesNodes()
5923

    
5924
  def BuildHooksEnv(self):
5925
    """Build hooks env.
5926

5927
    This runs on master, primary and secondary nodes of the instance.
5928

5929
    """
5930
    instance = self._migrater.instance
5931
    source_node = instance.primary_node
5932
    target_node = self.op.target_node
5933
    env = _BuildInstanceHookEnvByObject(self, instance)
5934
    env["MIGRATE_LIVE"] = self._migrater.live
5935
    env["MIGRATE_CLEANUP"] = self.op.cleanup
5936
    env.update({
5937
        "OLD_PRIMARY": source_node,
5938
        "NEW_PRIMARY": target_node,
5939
        })
5940

    
5941
    if instance.disk_template in constants.DTS_INT_MIRROR:
5942
      env["OLD_SECONDARY"] = target_node
5943
      env["NEW_SECONDARY"] = source_node
5944
    else:
5945
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
5946

    
5947
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
5948
    nl_post = list(nl)
5949
    nl_post.append(source_node)
5950
    return env, nl, nl_post
5951

    
5952

    
5953
class LUInstanceMove(LogicalUnit):
5954
  """Move an instance by data-copying.
5955

5956
  """
5957
  HPATH = "instance-move"
5958
  HTYPE = constants.HTYPE_INSTANCE
5959
  REQ_BGL = False
5960

    
5961
  def ExpandNames(self):
5962
    self._ExpandAndLockInstance()
5963
    target_node = _ExpandNodeName(self.cfg, self.op.target_node)
5964
    self.op.target_node = target_node
5965
    self.needed_locks[locking.LEVEL_NODE] = [target_node]
5966
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5967

    
5968
  def DeclareLocks(self, level):
5969
    if level == locking.LEVEL_NODE:
5970
      self._LockInstancesNodes(primary_only=True)
5971

    
5972
  def BuildHooksEnv(self):
5973
    """Build hooks env.
5974

5975
    This runs on master, primary and secondary nodes of the instance.
5976

5977
    """
5978
    env = {
5979
      "TARGET_NODE": self.op.target_node,
5980
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5981
      }
5982
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5983
    nl = [self.cfg.GetMasterNode()] + [self.instance.primary_node,
5984
                                       self.op.target_node]
5985
    return env, nl, nl
5986

    
5987
  def CheckPrereq(self):
5988
    """Check prerequisites.
5989

5990
    This checks that the instance is in the cluster.
5991

5992
    """
5993
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5994
    assert self.instance is not None, \
5995
      "Cannot retrieve locked instance %s" % self.op.instance_name
5996

    
5997
    node = self.cfg.GetNodeInfo(self.op.target_node)
5998
    assert node is not None, \
5999
      "Cannot retrieve locked node %s" % self.op.target_node
6000

    
6001
    self.target_node = target_node = node.name
6002

    
6003
    if target_node == instance.primary_node:
6004
      raise errors.OpPrereqError("Instance %s is already on the node %s" %
6005
                                 (instance.name, target_node),
6006
                                 errors.ECODE_STATE)
6007

    
6008
    bep = self.cfg.GetClusterInfo().FillBE(instance)
6009

    
6010
    for idx, dsk in enumerate(instance.disks):
6011
      if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
6012
        raise errors.OpPrereqError("Instance disk %d has a complex layout,"
6013
                                   " cannot copy" % idx, errors.ECODE_STATE)
6014

    
6015
    _CheckNodeOnline(self, target_node)
6016
    _CheckNodeNotDrained(self, target_node)
6017
    _CheckNodeVmCapable(self, target_node)
6018

    
6019
    if instance.admin_up:
6020
      # check memory requirements on the secondary node
6021
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
6022
                           instance.name, bep[constants.BE_MEMORY],
6023
                           instance.hypervisor)
6024
    else:
6025
      self.LogInfo("Not checking memory on the secondary node as"
6026
                   " instance will not be started")
6027

    
6028
    # check bridge existance
6029
    _CheckInstanceBridgesExist(self, instance, node=target_node)
6030

    
6031
  def Exec(self, feedback_fn):
6032
    """Move an instance.
6033

6034
    The move is done by shutting it down on its present node, copying
6035
    the data over (slow) and starting it on the new node.
6036

6037
    """
6038
    instance = self.instance
6039

    
6040
    source_node = instance.primary_node
6041
    target_node = self.target_node
6042

    
6043
    self.LogInfo("Shutting down instance %s on source node %s",
6044
                 instance.name, source_node)
6045

    
6046
    result = self.rpc.call_instance_shutdown(source_node, instance,
6047
                                             self.op.shutdown_timeout)
6048
    msg = result.fail_msg
6049
    if msg:
6050
      if self.op.ignore_consistency:
6051
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
6052
                             " Proceeding anyway. Please make sure node"
6053
                             " %s is down. Error details: %s",
6054
                             instance.name, source_node, source_node, msg)
6055
      else:
6056
        raise errors.OpExecError("Could not shutdown instance %s on"
6057
                                 " node %s: %s" %
6058
                                 (instance.name, source_node, msg))
6059

    
6060
    # create the target disks
6061
    try:
6062
      _CreateDisks(self, instance, target_node=target_node)
6063
    except errors.OpExecError:
6064
      self.LogWarning("Device creation failed, reverting...")
6065
      try:
6066
        _RemoveDisks(self, instance, target_node=target_node)
6067
      finally:
6068
        self.cfg.ReleaseDRBDMinors(instance.name)
6069
        raise
6070

    
6071
    cluster_name = self.cfg.GetClusterInfo().cluster_name
6072

    
6073
    errs = []
6074
    # activate, get path, copy the data over
6075
    for idx, disk in enumerate(instance.disks):
6076
      self.LogInfo("Copying data for disk %d", idx)
6077
      result = self.rpc.call_blockdev_assemble(target_node, disk,
6078
                                               instance.name, True, idx)
6079
      if result.fail_msg:
6080
        self.LogWarning("Can't assemble newly created disk %d: %s",
6081
                        idx, result.fail_msg)
6082
        errs.append(result.fail_msg)
6083
        break
6084
      dev_path = result.payload
6085
      result = self.rpc.call_blockdev_export(source_node, disk,
6086
                                             target_node, dev_path,
6087
                                             cluster_name)
6088
      if result.fail_msg:
6089
        self.LogWarning("Can't copy data over for disk %d: %s",
6090
                        idx, result.fail_msg)
6091
        errs.append(result.fail_msg)
6092
        break
6093

    
6094
    if errs:
6095
      self.LogWarning("Some disks failed to copy, aborting")
6096
      try:
6097
        _RemoveDisks(self, instance, target_node=target_node)
6098
      finally:
6099
        self.cfg.ReleaseDRBDMinors(instance.name)
6100
        raise errors.OpExecError("Errors during disk copy: %s" %
6101
                                 (",".join(errs),))
6102

    
6103
    instance.primary_node = target_node
6104
    self.cfg.Update(instance, feedback_fn)
6105

    
6106
    self.LogInfo("Removing the disks on the original node")
6107
    _RemoveDisks(self, instance, target_node=source_node)
6108

    
6109
    # Only start the instance if it's marked as up
6110
    if instance.admin_up:
6111
      self.LogInfo("Starting instance %s on node %s",
6112
                   instance.name, target_node)
6113

    
6114
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
6115
                                           ignore_secondaries=True)
6116
      if not disks_ok:
6117
        _ShutdownInstanceDisks(self, instance)
6118
        raise errors.OpExecError("Can't activate the instance's disks")
6119

    
6120
      result = self.rpc.call_instance_start(target_node, instance, None, None)
6121
      msg = result.fail_msg
6122
      if msg:
6123
        _ShutdownInstanceDisks(self, instance)
6124
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
6125
                                 (instance.name, target_node, msg))
6126

    
6127

    
6128
class LUNodeMigrate(LogicalUnit):
6129
  """Migrate all instances from a node.
6130

6131
  """
6132
  HPATH = "node-migrate"
6133
  HTYPE = constants.HTYPE_NODE
6134
  REQ_BGL = False
6135

    
6136
  def CheckArguments(self):
6137
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
6138

    
6139
  def ExpandNames(self):
6140
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6141

    
6142
    self.needed_locks = {}
6143

    
6144
    # Create tasklets for migrating instances for all instances on this node
6145
    names = []
6146
    tasklets = []
6147

    
6148
    self.lock_all_nodes = False
6149

    
6150
    for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name):
6151
      logging.debug("Migrating instance %s", inst.name)
6152
      names.append(inst.name)
6153

    
6154
      tasklets.append(TLMigrateInstance(self, inst.name, False))
6155

    
6156
      if inst.disk_template in constants.DTS_EXT_MIRROR:
6157
        # We need to lock all nodes, as the iallocator will choose the
6158
        # destination nodes afterwards
6159
        self.lock_all_nodes = True
6160

    
6161
    self.tasklets = tasklets
6162

    
6163
    # Declare node locks
6164
    if self.lock_all_nodes:
6165
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6166
    else:
6167
      self.needed_locks[locking.LEVEL_NODE] = [self.op.node_name]
6168
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6169

    
6170
    # Declare instance locks
6171
    self.needed_locks[locking.LEVEL_INSTANCE] = names
6172

    
6173
  def DeclareLocks(self, level):
6174
    if level == locking.LEVEL_NODE and not self.lock_all_nodes:
6175
      self._LockInstancesNodes()
6176

    
6177
  def BuildHooksEnv(self):
6178
    """Build hooks env.
6179

6180
    This runs on the master, the primary and all the secondaries.
6181

6182
    """
6183
    env = {
6184
      "NODE_NAME": self.op.node_name,
6185
      }
6186

    
6187
    nl = [self.cfg.GetMasterNode()]
6188

    
6189
    return (env, nl, nl)
6190

    
6191

    
6192
class TLMigrateInstance(Tasklet):
6193
  """Tasklet class for instance migration.
6194

6195
  @type live: boolean
6196
  @ivar live: whether the migration will be done live or non-live;
6197
      this variable is initalized only after CheckPrereq has run
6198

6199
  """
6200
  def __init__(self, lu, instance_name, cleanup):
6201
    """Initializes this class.
6202

6203
    """
6204
    Tasklet.__init__(self, lu)
6205

    
6206
    # Parameters
6207
    self.instance_name = instance_name
6208
    self.cleanup = cleanup
6209
    self.live = False # will be overridden later
6210

    
6211
  def CheckPrereq(self):
6212
    """Check prerequisites.
6213

6214
    This checks that the instance is in the cluster.
6215

6216
    """
6217
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
6218
    instance = self.cfg.GetInstanceInfo(instance_name)
6219
    assert instance is not None
6220
    self.instance = instance
6221

    
6222
    if instance.disk_template not in constants.DTS_MIRRORED:
6223
      raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
6224
                                 " migrations" % instance.disk_template,
6225
                                 errors.ECODE_STATE)
6226

    
6227
    if instance.disk_template in constants.DTS_EXT_MIRROR:
6228
      _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
6229

    
6230
      if self.lu.op.iallocator:
6231
        self._RunAllocator()
6232
      else:
6233
        # We set set self.target_node as it is required by
6234
        # BuildHooksEnv
6235
        self.target_node = self.lu.op.target_node
6236

    
6237
      target_node = self.target_node
6238
      if self.target_node == instance.primary_node:
6239
        raise errors.OpPrereqError("Cannot migrate instance %s"
6240
                                   " to its primary (%s)" %
6241
                                   (instance.name, instance.primary_node))
6242

    
6243
      if len(self.lu.tasklets) == 1:
6244
        # It is safe to remove locks only when we're the only tasklet in the LU
6245
        nodes_keep = [instance.primary_node, target_node]
6246
        nodes_rel = [node for node in self.lu.acquired_locks[locking.LEVEL_NODE]
6247
                     if node not in nodes_keep]
6248
        self.lu.context.glm.release(locking.LEVEL_NODE, nodes_rel)
6249
        self.lu.acquired_locks[locking.LEVEL_NODE] = nodes_keep
6250

    
6251
    else:
6252
      secondary_nodes = instance.secondary_nodes
6253
      if not secondary_nodes:
6254
        raise errors.ConfigurationError("No secondary node but using"
6255
                                        " %s disk template" %
6256
                                        instance.disk_template)
6257
      target_node = secondary_nodes[0]
6258
      if self.lu.op.iallocator or (self.lu.op.target_node and
6259
                                   self.lu.op.target_node != target_node):
6260
        raise errors.OpPrereqError("Instances with disk template %s cannot"
6261
                                   " be migrated over to arbitrary nodes"
6262
                                   " (neither an iallocator nor a target"
6263
                                   " node can be passed)" %
6264
                                   instance.disk_template, errors.ECODE_INVAL)
6265

    
6266
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
6267

    
6268
    # check memory requirements on the secondary node
6269
    _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
6270
                         instance.name, i_be[constants.BE_MEMORY],
6271
                         instance.hypervisor)
6272

    
6273
    # check bridge existance
6274
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
6275

    
6276
    if not self.cleanup:
6277
      _CheckNodeNotDrained(self.lu, target_node)
6278
      result = self.rpc.call_instance_migratable(instance.primary_node,
6279
                                                 instance)
6280
      result.Raise("Can't migrate, please use failover",
6281
                   prereq=True, ecode=errors.ECODE_STATE)
6282

    
6283

    
6284
    if self.lu.op.live is not None and self.lu.op.mode is not None:
6285
      raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
6286
                                 " parameters are accepted",
6287
                                 errors.ECODE_INVAL)
6288
    if self.lu.op.live is not None:
6289
      if self.lu.op.live:
6290
        self.lu.op.mode = constants.HT_MIGRATION_LIVE
6291
      else:
6292
        self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
6293
      # reset the 'live' parameter to None so that repeated
6294
      # invocations of CheckPrereq do not raise an exception
6295
      self.lu.op.live = None
6296
    elif self.lu.op.mode is None:
6297
      # read the default value from the hypervisor
6298
      i_hv = self.cfg.GetClusterInfo().FillHV(self.instance, skip_globals=False)
6299
      self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
6300

    
6301
    self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
6302

    
6303
  def _RunAllocator(self):
6304
    """Run the allocator based on input opcode.
6305

6306
    """
6307
    ial = IAllocator(self.cfg, self.rpc,
6308
                     mode=constants.IALLOCATOR_MODE_RELOC,
6309
                     name=self.instance_name,
6310
                     # TODO See why hail breaks with a single node below
6311
                     relocate_from=[self.instance.primary_node,
6312
                                    self.instance.primary_node],
6313
                     )
6314

    
6315
    ial.Run(self.lu.op.iallocator)
6316

    
6317
    if not ial.success:
6318
      raise errors.OpPrereqError("Can't compute nodes using"
6319
                                 " iallocator '%s': %s" %
6320
                                 (self.lu.op.iallocator, ial.info),
6321
                                 errors.ECODE_NORES)
6322
    if len(ial.result) != ial.required_nodes:
6323
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
6324
                                 " of nodes (%s), required %s" %
6325
                                 (self.lu.op.iallocator, len(ial.result),
6326
                                  ial.required_nodes), errors.ECODE_FAULT)
6327
    self.target_node = ial.result[0]
6328
    self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
6329
                 self.instance_name, self.lu.op.iallocator,
6330
                 utils.CommaJoin(ial.result))
6331

    
6332
  def _WaitUntilSync(self):
6333
    """Poll with custom rpc for disk sync.
6334

6335
    This uses our own step-based rpc call.
6336

6337
    """
6338
    self.feedback_fn("* wait until resync is done")
6339
    all_done = False
6340
    while not all_done:
6341
      all_done = True
6342
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
6343
                                            self.nodes_ip,
6344
                                            self.instance.disks)
6345
      min_percent = 100
6346
      for node, nres in result.items():
6347
        nres.Raise("Cannot resync disks on node %s" % node)
6348
        node_done, node_percent = nres.payload
6349
        all_done = all_done and node_done
6350
        if node_percent is not None:
6351
          min_percent = min(min_percent, node_percent)
6352
      if not all_done:
6353
        if min_percent < 100:
6354
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
6355
        time.sleep(2)
6356

    
6357
  def _EnsureSecondary(self, node):
6358
    """Demote a node to secondary.
6359

6360
    """
6361
    self.feedback_fn("* switching node %s to secondary mode" % node)
6362

    
6363
    for dev in self.instance.disks:
6364
      self.cfg.SetDiskID(dev, node)
6365

    
6366
    result = self.rpc.call_blockdev_close(node, self.instance.name,
6367
                                          self.instance.disks)
6368
    result.Raise("Cannot change disk to secondary on node %s" % node)
6369

    
6370
  def _GoStandalone(self):
6371
    """Disconnect from the network.
6372

6373
    """
6374
    self.feedback_fn("* changing into standalone mode")
6375
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
6376
                                               self.instance.disks)
6377
    for node, nres in result.items():
6378
      nres.Raise("Cannot disconnect disks node %s" % node)
6379

    
6380
  def _GoReconnect(self, multimaster):
6381
    """Reconnect to the network.
6382

6383
    """
6384
    if multimaster:
6385
      msg = "dual-master"
6386
    else:
6387
      msg = "single-master"
6388
    self.feedback_fn("* changing disks into %s mode" % msg)
6389
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
6390
                                           self.instance.disks,
6391
                                           self.instance.name, multimaster)
6392
    for node, nres in result.items():
6393
      nres.Raise("Cannot change disks config on node %s" % node)
6394

    
6395
  def _ExecCleanup(self):
6396
    """Try to cleanup after a failed migration.
6397

6398
    The cleanup is done by:
6399
      - check that the instance is running only on one node
6400
        (and update the config if needed)
6401
      - change disks on its secondary node to secondary
6402
      - wait until disks are fully synchronized
6403
      - disconnect from the network
6404
      - change disks into single-master mode
6405
      - wait again until disks are fully synchronized
6406

6407
    """
6408
    instance = self.instance
6409
    target_node = self.target_node
6410
    source_node = self.source_node
6411

    
6412
    # check running on only one node
6413
    self.feedback_fn("* checking where the instance actually runs"
6414
                     " (if this hangs, the hypervisor might be in"
6415
                     " a bad state)")
6416
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
6417
    for node, result in ins_l.items():
6418
      result.Raise("Can't contact node %s" % node)
6419

    
6420
    runningon_source = instance.name in ins_l[source_node].payload
6421
    runningon_target = instance.name in ins_l[target_node].payload
6422

    
6423
    if runningon_source and runningon_target:
6424
      raise errors.OpExecError("Instance seems to be running on two nodes,"
6425
                               " or the hypervisor is confused. You will have"
6426
                               " to ensure manually that it runs only on one"
6427
                               " and restart this operation.")
6428

    
6429
    if not (runningon_source or runningon_target):
6430
      raise errors.OpExecError("Instance does not seem to be running at all."
6431
                               " In this case, it's safer to repair by"
6432
                               " running 'gnt-instance stop' to ensure disk"
6433
                               " shutdown, and then restarting it.")
6434

    
6435
    if runningon_target:
6436
      # the migration has actually succeeded, we need to update the config
6437
      self.feedback_fn("* instance running on secondary node (%s),"
6438
                       " updating config" % target_node)
6439
      instance.primary_node = target_node
6440
      self.cfg.Update(instance, self.feedback_fn)
6441
      demoted_node = source_node
6442
    else:
6443
      self.feedback_fn("* instance confirmed to be running on its"
6444
                       " primary node (%s)" % source_node)
6445
      demoted_node = target_node
6446

    
6447
    if instance.disk_template in constants.DTS_INT_MIRROR:
6448
      self._EnsureSecondary(demoted_node)
6449
      try:
6450
        self._WaitUntilSync()
6451
      except errors.OpExecError:
6452
        # we ignore here errors, since if the device is standalone, it
6453
        # won't be able to sync
6454
        pass
6455
      self._GoStandalone()
6456
      self._GoReconnect(False)
6457
      self._WaitUntilSync()
6458

    
6459
    self.feedback_fn("* done")
6460

    
6461
  def _RevertDiskStatus(self):
6462
    """Try to revert the disk status after a failed migration.
6463

6464
    """
6465
    target_node = self.target_node
6466
    if self.instance.disk_template in constants.DTS_EXT_MIRROR:
6467
      return
6468

    
6469
    try:
6470
      self._EnsureSecondary(target_node)
6471
      self._GoStandalone()
6472
      self._GoReconnect(False)
6473
      self._WaitUntilSync()
6474
    except errors.OpExecError, err:
6475
      self.lu.LogWarning("Migration failed and I can't reconnect the"
6476
                         " drives: error '%s'\n"
6477
                         "Please look and recover the instance status" %
6478
                         str(err))
6479

    
6480
  def _AbortMigration(self):
6481
    """Call the hypervisor code to abort a started migration.
6482

6483
    """
6484
    instance = self.instance
6485
    target_node = self.target_node
6486
    migration_info = self.migration_info
6487

    
6488
    abort_result = self.rpc.call_finalize_migration(target_node,
6489
                                                    instance,
6490
                                                    migration_info,
6491
                                                    False)
6492
    abort_msg = abort_result.fail_msg
6493
    if abort_msg:
6494
      logging.error("Aborting migration failed on target node %s: %s",
6495
                    target_node, abort_msg)
6496
      # Don't raise an exception here, as we stil have to try to revert the
6497
      # disk status, even if this step failed.
6498

    
6499
  def _ExecMigration(self):
6500
    """Migrate an instance.
6501

6502
    The migrate is done by:
6503
      - change the disks into dual-master mode
6504
      - wait until disks are fully synchronized again
6505
      - migrate the instance
6506
      - change disks on the new secondary node (the old primary) to secondary
6507
      - wait until disks are fully synchronized
6508
      - change disks into single-master mode
6509

6510
    """
6511
    instance = self.instance
6512
    target_node = self.target_node
6513
    source_node = self.source_node
6514

    
6515
    self.feedback_fn("* checking disk consistency between source and target")
6516
    for dev in instance.disks:
6517
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
6518
        raise errors.OpExecError("Disk %s is degraded or not fully"
6519
                                 " synchronized on target node,"
6520
                                 " aborting migrate." % dev.iv_name)
6521

    
6522
    # First get the migration information from the remote node
6523
    result = self.rpc.call_migration_info(source_node, instance)
6524
    msg = result.fail_msg
6525
    if msg:
6526
      log_err = ("Failed fetching source migration information from %s: %s" %
6527
                 (source_node, msg))
6528
      logging.error(log_err)
6529
      raise errors.OpExecError(log_err)
6530

    
6531
    self.migration_info = migration_info = result.payload
6532

    
6533
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
6534
      # Then switch the disks to master/master mode
6535
      self._EnsureSecondary(target_node)
6536
      self._GoStandalone()
6537
      self._GoReconnect(True)
6538
      self._WaitUntilSync()
6539

    
6540
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
6541
    result = self.rpc.call_accept_instance(target_node,
6542
                                           instance,
6543
                                           migration_info,
6544
                                           self.nodes_ip[target_node])
6545

    
6546
    msg = result.fail_msg
6547
    if msg:
6548
      logging.error("Instance pre-migration failed, trying to revert"
6549
                    " disk status: %s", msg)
6550
      self.feedback_fn("Pre-migration failed, aborting")
6551
      self._AbortMigration()
6552
      self._RevertDiskStatus()
6553
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
6554
                               (instance.name, msg))
6555

    
6556
    self.feedback_fn("* migrating instance to %s" % target_node)
6557
    result = self.rpc.call_instance_migrate(source_node, instance,
6558
                                            self.nodes_ip[target_node],
6559
                                            self.live)
6560
    msg = result.fail_msg
6561
    if msg:
6562
      logging.error("Instance migration failed, trying to revert"
6563
                    " disk status: %s", msg)
6564
      self.feedback_fn("Migration failed, aborting")
6565
      self._AbortMigration()
6566
      self._RevertDiskStatus()
6567
      raise errors.OpExecError("Could not migrate instance %s: %s" %
6568
                               (instance.name, msg))
6569

    
6570
    instance.primary_node = target_node
6571
    # distribute new instance config to the other nodes
6572
    self.cfg.Update(instance, self.feedback_fn)
6573

    
6574
    result = self.rpc.call_finalize_migration(target_node,
6575
                                              instance,
6576
                                              migration_info,
6577
                                              True)
6578
    msg = result.fail_msg
6579
    if msg:
6580
      logging.error("Instance migration succeeded, but finalization failed:"
6581
                    " %s", msg)
6582
      raise errors.OpExecError("Could not finalize instance migration: %s" %
6583
                               msg)
6584

    
6585
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
6586
      self._EnsureSecondary(source_node)
6587
      self._WaitUntilSync()
6588
      self._GoStandalone()
6589
      self._GoReconnect(False)
6590
      self._WaitUntilSync()
6591

    
6592
    self.feedback_fn("* done")
6593

    
6594
  def Exec(self, feedback_fn):
6595
    """Perform the migration.
6596

6597
    """
6598
    feedback_fn("Migrating instance %s" % self.instance.name)
6599

    
6600
    self.feedback_fn = feedback_fn
6601

    
6602
    self.source_node = self.instance.primary_node
6603

    
6604
    # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
6605
    if self.instance.disk_template in constants.DTS_INT_MIRROR:
6606
      self.target_node = self.instance.secondary_nodes[0]
6607
      # Otherwise self.target_node has been populated either
6608
      # directly, or through an iallocator.
6609

    
6610
    self.all_nodes = [self.source_node, self.target_node]
6611
    self.nodes_ip = {
6612
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
6613
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
6614
      }
6615

    
6616
    if self.cleanup:
6617
      return self._ExecCleanup()
6618
    else:
6619
      return self._ExecMigration()
6620

    
6621

    
6622
def _CreateBlockDev(lu, node, instance, device, force_create,
6623
                    info, force_open):
6624
  """Create a tree of block devices on a given node.
6625

6626
  If this device type has to be created on secondaries, create it and
6627
  all its children.
6628

6629
  If not, just recurse to children keeping the same 'force' value.
6630

6631
  @param lu: the lu on whose behalf we execute
6632
  @param node: the node on which to create the device
6633
  @type instance: L{objects.Instance}
6634
  @param instance: the instance which owns the device
6635
  @type device: L{objects.Disk}
6636
  @param device: the device to create
6637
  @type force_create: boolean
6638
  @param force_create: whether to force creation of this device; this
6639
      will be change to True whenever we find a device which has
6640
      CreateOnSecondary() attribute
6641
  @param info: the extra 'metadata' we should attach to the device
6642
      (this will be represented as a LVM tag)
6643
  @type force_open: boolean
6644
  @param force_open: this parameter will be passes to the
6645
      L{backend.BlockdevCreate} function where it specifies
6646
      whether we run on primary or not, and it affects both
6647
      the child assembly and the device own Open() execution
6648

6649
  """
6650
  if device.CreateOnSecondary():
6651
    force_create = True
6652

    
6653
  if device.children:
6654
    for child in device.children:
6655
      _CreateBlockDev(lu, node, instance, child, force_create,
6656
                      info, force_open)
6657

    
6658
  if not force_create:
6659
    return
6660

    
6661
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
6662

    
6663

    
6664
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
6665
  """Create a single block device on a given node.
6666

6667
  This will not recurse over children of the device, so they must be
6668
  created in advance.
6669

6670
  @param lu: the lu on whose behalf we execute
6671
  @param node: the node on which to create the device
6672
  @type instance: L{objects.Instance}
6673
  @param instance: the instance which owns the device
6674
  @type device: L{objects.Disk}
6675
  @param device: the device to create
6676
  @param info: the extra 'metadata' we should attach to the device
6677
      (this will be represented as a LVM tag)
6678
  @type force_open: boolean
6679
  @param force_open: this parameter will be passes to the
6680
      L{backend.BlockdevCreate} function where it specifies
6681
      whether we run on primary or not, and it affects both
6682
      the child assembly and the device own Open() execution
6683

6684
  """
6685
  lu.cfg.SetDiskID(device, node)
6686
  result = lu.rpc.call_blockdev_create(node, device, device.size,
6687
                                       instance.name, force_open, info)
6688
  result.Raise("Can't create block device %s on"
6689
               " node %s for instance %s" % (device, node, instance.name))
6690
  if device.physical_id is None:
6691
    device.physical_id = result.payload
6692

    
6693

    
6694
def _GenerateUniqueNames(lu, exts):
6695
  """Generate a suitable LV name.
6696

6697
  This will generate a logical volume name for the given instance.
6698

6699
  """
6700
  results = []
6701
  for val in exts:
6702
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
6703
    results.append("%s%s" % (new_id, val))
6704
  return results
6705

    
6706

    
6707
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgname, names, iv_name,
6708
                         p_minor, s_minor):
6709
  """Generate a drbd8 device complete with its children.
6710

6711
  """
6712
  port = lu.cfg.AllocatePort()
6713
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
6714
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
6715
                          logical_id=(vgname, names[0]))
6716
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
6717
                          logical_id=(vgname, names[1]))
6718
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
6719
                          logical_id=(primary, secondary, port,
6720
                                      p_minor, s_minor,
6721
                                      shared_secret),
6722
                          children=[dev_data, dev_meta],
6723
                          iv_name=iv_name)
6724
  return drbd_dev
6725

    
6726

    
6727
def _GenerateDiskTemplate(lu, template_name,
6728
                          instance_name, primary_node,
6729
                          secondary_nodes, disk_info,
6730
                          file_storage_dir, file_driver,
6731
                          base_index, feedback_fn):
6732
  """Generate the entire disk layout for a given template type.
6733

6734
  """
6735
  #TODO: compute space requirements
6736

    
6737
  vgname = lu.cfg.GetVGName()
6738
  disk_count = len(disk_info)
6739
  disks = []
6740
  if template_name == constants.DT_DISKLESS:
6741
    pass
6742
  elif template_name == constants.DT_PLAIN:
6743
    if len(secondary_nodes) != 0:
6744
      raise errors.ProgrammerError("Wrong template configuration")
6745

    
6746
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6747
                                      for i in range(disk_count)])
6748
    for idx, disk in enumerate(disk_info):
6749
      disk_index = idx + base_index
6750
      vg = disk.get("vg", vgname)
6751
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
6752
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
6753
                              logical_id=(vg, names[idx]),
6754
                              iv_name="disk/%d" % disk_index,
6755
                              mode=disk["mode"])
6756
      disks.append(disk_dev)
6757
  elif template_name == constants.DT_DRBD8:
6758
    if len(secondary_nodes) != 1:
6759
      raise errors.ProgrammerError("Wrong template configuration")
6760
    remote_node = secondary_nodes[0]
6761
    minors = lu.cfg.AllocateDRBDMinor(
6762
      [primary_node, remote_node] * len(disk_info), instance_name)
6763

    
6764
    names = []
6765
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6766
                                               for i in range(disk_count)]):
6767
      names.append(lv_prefix + "_data")
6768
      names.append(lv_prefix + "_meta")
6769
    for idx, disk in enumerate(disk_info):
6770
      disk_index = idx + base_index
6771
      vg = disk.get("vg", vgname)
6772
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
6773
                                      disk["size"], vg, names[idx*2:idx*2+2],
6774
                                      "disk/%d" % disk_index,
6775
                                      minors[idx*2], minors[idx*2+1])
6776
      disk_dev.mode = disk["mode"]
6777
      disks.append(disk_dev)
6778
  elif template_name == constants.DT_FILE:
6779
    if len(secondary_nodes) != 0:
6780
      raise errors.ProgrammerError("Wrong template configuration")
6781

    
6782
    opcodes.RequireFileStorage()
6783

    
6784
    for idx, disk in enumerate(disk_info):
6785
      disk_index = idx + base_index
6786
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6787
                              iv_name="disk/%d" % disk_index,
6788
                              logical_id=(file_driver,
6789
                                          "%s/disk%d" % (file_storage_dir,
6790
                                                         disk_index)),
6791
                              mode=disk["mode"])
6792
      disks.append(disk_dev)
6793
  elif template_name == constants.DT_SHARED_FILE:
6794
    if len(secondary_nodes) != 0:
6795
      raise errors.ProgrammerError("Wrong template configuration")
6796

    
6797
    opcodes.RequireSharedFileStorage()
6798

    
6799
    for idx, disk in enumerate(disk_info):
6800
      disk_index = idx + base_index
6801
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6802
                              iv_name="disk/%d" % disk_index,
6803
                              logical_id=(file_driver,
6804
                                          "%s/disk%d" % (file_storage_dir,
6805
                                                         disk_index)),
6806
                              mode=disk["mode"])
6807
      disks.append(disk_dev)
6808
  elif template_name == constants.DT_BLOCK:
6809
    if len(secondary_nodes) != 0:
6810
      raise errors.ProgrammerError("Wrong template configuration")
6811

    
6812
    for idx, disk in enumerate(disk_info):
6813
      disk_index = idx + base_index
6814
      disk_dev = objects.Disk(dev_type=constants.LD_BLOCKDEV, size=disk["size"],
6815
                              logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
6816
                                          disk["adopt"]),
6817
                              iv_name="disk/%d" % disk_index,
6818
                              mode=disk["mode"])
6819
      disks.append(disk_dev)
6820

    
6821
  else:
6822
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
6823
  return disks
6824

    
6825

    
6826
def _GetInstanceInfoText(instance):
6827
  """Compute that text that should be added to the disk's metadata.
6828

6829
  """
6830
  return "originstname+%s" % instance.name
6831

    
6832

    
6833
def _CalcEta(time_taken, written, total_size):
6834
  """Calculates the ETA based on size written and total size.
6835

6836
  @param time_taken: The time taken so far
6837
  @param written: amount written so far
6838
  @param total_size: The total size of data to be written
6839
  @return: The remaining time in seconds
6840

6841
  """
6842
  avg_time = time_taken / float(written)
6843
  return (total_size - written) * avg_time
6844

    
6845

    
6846
def _WipeDisks(lu, instance):
6847
  """Wipes instance disks.
6848

6849
  @type lu: L{LogicalUnit}
6850
  @param lu: the logical unit on whose behalf we execute
6851
  @type instance: L{objects.Instance}
6852
  @param instance: the instance whose disks we should create
6853
  @return: the success of the wipe
6854

6855
  """
6856
  node = instance.primary_node
6857

    
6858
  for device in instance.disks:
6859
    lu.cfg.SetDiskID(device, node)
6860

    
6861
  logging.info("Pause sync of instance %s disks", instance.name)
6862
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
6863

    
6864
  for idx, success in enumerate(result.payload):
6865
    if not success:
6866
      logging.warn("pause-sync of instance %s for disks %d failed",
6867
                   instance.name, idx)
6868

    
6869
  try:
6870
    for idx, device in enumerate(instance.disks):
6871
      lu.LogInfo("* Wiping disk %d", idx)
6872
      logging.info("Wiping disk %d for instance %s, node %s",
6873
                   idx, instance.name, node)
6874

    
6875
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
6876
      # MAX_WIPE_CHUNK at max
6877
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
6878
                            constants.MIN_WIPE_CHUNK_PERCENT)
6879

    
6880
      offset = 0
6881
      size = device.size
6882
      last_output = 0
6883
      start_time = time.time()
6884

    
6885
      while offset < size:
6886
        wipe_size = min(wipe_chunk_size, size - offset)
6887
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
6888
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
6889
                     (idx, offset, wipe_size))
6890
        now = time.time()
6891
        offset += wipe_size
6892
        if now - last_output >= 60:
6893
          eta = _CalcEta(now - start_time, offset, size)
6894
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
6895
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
6896
          last_output = now
6897
  finally:
6898
    logging.info("Resume sync of instance %s disks", instance.name)
6899

    
6900
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
6901

    
6902
    for idx, success in enumerate(result.payload):
6903
      if not success:
6904
        lu.LogWarning("Warning: Resume sync of disk %d failed. Please have a"
6905
                      " look at the status and troubleshoot the issue.", idx)
6906
        logging.warn("resume-sync of instance %s for disks %d failed",
6907
                     instance.name, idx)
6908

    
6909

    
6910
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
6911
  """Create all disks for an instance.
6912

6913
  This abstracts away some work from AddInstance.
6914

6915
  @type lu: L{LogicalUnit}
6916
  @param lu: the logical unit on whose behalf we execute
6917
  @type instance: L{objects.Instance}
6918
  @param instance: the instance whose disks we should create
6919
  @type to_skip: list
6920
  @param to_skip: list of indices to skip
6921
  @type target_node: string
6922
  @param target_node: if passed, overrides the target node for creation
6923
  @rtype: boolean
6924
  @return: the success of the creation
6925

6926
  """
6927
  info = _GetInstanceInfoText(instance)
6928
  if target_node is None:
6929
    pnode = instance.primary_node
6930
    all_nodes = instance.all_nodes
6931
  else:
6932
    pnode = target_node
6933
    all_nodes = [pnode]
6934

    
6935
  if instance.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE):
6936
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6937
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
6938

    
6939
    result.Raise("Failed to create directory '%s' on"
6940
                 " node %s" % (file_storage_dir, pnode))
6941

    
6942
  # Note: this needs to be kept in sync with adding of disks in
6943
  # LUInstanceSetParams
6944
  for idx, device in enumerate(instance.disks):
6945
    if to_skip and idx in to_skip:
6946
      continue
6947
    logging.info("Creating volume %s for instance %s",
6948
                 device.iv_name, instance.name)
6949
    #HARDCODE
6950
    for node in all_nodes:
6951
      f_create = node == pnode
6952
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
6953

    
6954

    
6955
def _RemoveDisks(lu, instance, target_node=None):
6956
  """Remove all disks for an instance.
6957

6958
  This abstracts away some work from `AddInstance()` and
6959
  `RemoveInstance()`. Note that in case some of the devices couldn't
6960
  be removed, the removal will continue with the other ones (compare
6961
  with `_CreateDisks()`).
6962

6963
  @type lu: L{LogicalUnit}
6964
  @param lu: the logical unit on whose behalf we execute
6965
  @type instance: L{objects.Instance}
6966
  @param instance: the instance whose disks we should remove
6967
  @type target_node: string
6968
  @param target_node: used to override the node on which to remove the disks
6969
  @rtype: boolean
6970
  @return: the success of the removal
6971

6972
  """
6973
  logging.info("Removing block devices for instance %s", instance.name)
6974

    
6975
  all_result = True
6976
  for device in instance.disks:
6977
    if target_node:
6978
      edata = [(target_node, device)]
6979
    else:
6980
      edata = device.ComputeNodeTree(instance.primary_node)
6981
    for node, disk in edata:
6982
      lu.cfg.SetDiskID(disk, node)
6983
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
6984
      if msg:
6985
        lu.LogWarning("Could not remove block device %s on node %s,"
6986
                      " continuing anyway: %s", device.iv_name, node, msg)
6987
        all_result = False
6988

    
6989
  if instance.disk_template == constants.DT_FILE:
6990
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6991
    if target_node:
6992
      tgt = target_node
6993
    else:
6994
      tgt = instance.primary_node
6995
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
6996
    if result.fail_msg:
6997
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
6998
                    file_storage_dir, instance.primary_node, result.fail_msg)
6999
      all_result = False
7000

    
7001
  return all_result
7002

    
7003

    
7004
def _ComputeDiskSizePerVG(disk_template, disks):
7005
  """Compute disk size requirements in the volume group
7006

7007
  """
7008
  def _compute(disks, payload):
7009
    """Universal algorithm
7010

7011
    """
7012
    vgs = {}
7013
    for disk in disks:
7014
      vgs[disk["vg"]] = vgs.get("vg", 0) + disk["size"] + payload
7015

    
7016
    return vgs
7017

    
7018
  # Required free disk space as a function of disk and swap space
7019
  req_size_dict = {
7020
    constants.DT_DISKLESS: {},
7021
    constants.DT_PLAIN: _compute(disks, 0),
7022
    # 128 MB are added for drbd metadata for each disk
7023
    constants.DT_DRBD8: _compute(disks, 128),
7024
    constants.DT_FILE: {},
7025
    constants.DT_SHARED_FILE: {},
7026
  }
7027

    
7028
  if disk_template not in req_size_dict:
7029
    raise errors.ProgrammerError("Disk template '%s' size requirement"
7030
                                 " is unknown" %  disk_template)
7031

    
7032
  return req_size_dict[disk_template]
7033

    
7034

    
7035
def _ComputeDiskSize(disk_template, disks):
7036
  """Compute disk size requirements in the volume group
7037

7038
  """
7039
  # Required free disk space as a function of disk and swap space
7040
  req_size_dict = {
7041
    constants.DT_DISKLESS: None,
7042
    constants.DT_PLAIN: sum(d["size"] for d in disks),
7043
    # 128 MB are added for drbd metadata for each disk
7044
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
7045
    constants.DT_FILE: None,
7046
    constants.DT_SHARED_FILE: 0,
7047
    constants.DT_BLOCK: 0,
7048
  }
7049

    
7050
  if disk_template not in req_size_dict:
7051
    raise errors.ProgrammerError("Disk template '%s' size requirement"
7052
                                 " is unknown" %  disk_template)
7053

    
7054
  return req_size_dict[disk_template]
7055

    
7056

    
7057
def _FilterVmNodes(lu, nodenames):
7058
  """Filters out non-vm_capable nodes from a list.
7059

7060
  @type lu: L{LogicalUnit}
7061
  @param lu: the logical unit for which we check
7062
  @type nodenames: list
7063
  @param nodenames: the list of nodes on which we should check
7064
  @rtype: list
7065
  @return: the list of vm-capable nodes
7066

7067
  """
7068
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
7069
  return [name for name in nodenames if name not in vm_nodes]
7070

    
7071

    
7072
def _CheckHVParams(lu, nodenames, hvname, hvparams):
7073
  """Hypervisor parameter validation.
7074

7075
  This function abstract the hypervisor parameter validation to be
7076
  used in both instance create and instance modify.
7077

7078
  @type lu: L{LogicalUnit}
7079
  @param lu: the logical unit for which we check
7080
  @type nodenames: list
7081
  @param nodenames: the list of nodes on which we should check
7082
  @type hvname: string
7083
  @param hvname: the name of the hypervisor we should use
7084
  @type hvparams: dict
7085
  @param hvparams: the parameters which we need to check
7086
  @raise errors.OpPrereqError: if the parameters are not valid
7087

7088
  """
7089
  nodenames = _FilterVmNodes(lu, nodenames)
7090
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
7091
                                                  hvname,
7092
                                                  hvparams)
7093
  for node in nodenames:
7094
    info = hvinfo[node]
7095
    if info.offline:
7096
      continue
7097
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
7098

    
7099

    
7100
def _CheckOSParams(lu, required, nodenames, osname, osparams):
7101
  """OS parameters validation.
7102

7103
  @type lu: L{LogicalUnit}
7104
  @param lu: the logical unit for which we check
7105
  @type required: boolean
7106
  @param required: whether the validation should fail if the OS is not
7107
      found
7108
  @type nodenames: list
7109
  @param nodenames: the list of nodes on which we should check
7110
  @type osname: string
7111
  @param osname: the name of the hypervisor we should use
7112
  @type osparams: dict
7113
  @param osparams: the parameters which we need to check
7114
  @raise errors.OpPrereqError: if the parameters are not valid
7115

7116
  """
7117
  nodenames = _FilterVmNodes(lu, nodenames)
7118
  result = lu.rpc.call_os_validate(required, nodenames, osname,
7119
                                   [constants.OS_VALIDATE_PARAMETERS],
7120
                                   osparams)
7121
  for node, nres in result.items():
7122
    # we don't check for offline cases since this should be run only
7123
    # against the master node and/or an instance's nodes
7124
    nres.Raise("OS Parameters validation failed on node %s" % node)
7125
    if not nres.payload:
7126
      lu.LogInfo("OS %s not found on node %s, validation skipped",
7127
                 osname, node)
7128

    
7129

    
7130
class LUInstanceCreate(LogicalUnit):
7131
  """Create an instance.
7132

7133
  """
7134
  HPATH = "instance-add"
7135
  HTYPE = constants.HTYPE_INSTANCE
7136
  REQ_BGL = False
7137

    
7138
  def CheckArguments(self):
7139
    """Check arguments.
7140

7141
    """
7142
    # do not require name_check to ease forward/backward compatibility
7143
    # for tools
7144
    if self.op.no_install and self.op.start:
7145
      self.LogInfo("No-installation mode selected, disabling startup")
7146
      self.op.start = False
7147
    # validate/normalize the instance name
7148
    self.op.instance_name = \
7149
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
7150

    
7151
    if self.op.ip_check and not self.op.name_check:
7152
      # TODO: make the ip check more flexible and not depend on the name check
7153
      raise errors.OpPrereqError("Cannot do ip check without a name check",
7154
                                 errors.ECODE_INVAL)
7155

    
7156
    # check nics' parameter names
7157
    for nic in self.op.nics:
7158
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
7159

    
7160
    # check disks. parameter names and consistent adopt/no-adopt strategy
7161
    has_adopt = has_no_adopt = False
7162
    for disk in self.op.disks:
7163
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
7164
      if "adopt" in disk:
7165
        has_adopt = True
7166
      else:
7167
        has_no_adopt = True
7168
    if has_adopt and has_no_adopt:
7169
      raise errors.OpPrereqError("Either all disks are adopted or none is",
7170
                                 errors.ECODE_INVAL)
7171
    if has_adopt:
7172
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
7173
        raise errors.OpPrereqError("Disk adoption is not supported for the"
7174
                                   " '%s' disk template" %
7175
                                   self.op.disk_template,
7176
                                   errors.ECODE_INVAL)
7177
      if self.op.iallocator is not None:
7178
        raise errors.OpPrereqError("Disk adoption not allowed with an"
7179
                                   " iallocator script", errors.ECODE_INVAL)
7180
      if self.op.mode == constants.INSTANCE_IMPORT:
7181
        raise errors.OpPrereqError("Disk adoption not allowed for"
7182
                                   " instance import", errors.ECODE_INVAL)
7183
    else:
7184
      if self.op.disk_template in constants.DTS_MUST_ADOPT:
7185
        raise errors.OpPrereqError("Disk template %s requires disk adoption,"
7186
                                   " but no 'adopt' parameter given" %
7187
                                   self.op.disk_template,
7188
                                   errors.ECODE_INVAL)
7189

    
7190
    self.adopt_disks = has_adopt
7191

    
7192
    # instance name verification
7193
    if self.op.name_check:
7194
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
7195
      self.op.instance_name = self.hostname1.name
7196
      # used in CheckPrereq for ip ping check
7197
      self.check_ip = self.hostname1.ip
7198
    else:
7199
      self.check_ip = None
7200

    
7201
    # file storage checks
7202
    if (self.op.file_driver and
7203
        not self.op.file_driver in constants.FILE_DRIVER):
7204
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
7205
                                 self.op.file_driver, errors.ECODE_INVAL)
7206

    
7207
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
7208
      raise errors.OpPrereqError("File storage directory path not absolute",
7209
                                 errors.ECODE_INVAL)
7210

    
7211
    ### Node/iallocator related checks
7212
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
7213

    
7214
    if self.op.pnode is not None:
7215
      if self.op.disk_template in constants.DTS_INT_MIRROR:
7216
        if self.op.snode is None:
7217
          raise errors.OpPrereqError("The networked disk templates need"
7218
                                     " a mirror node", errors.ECODE_INVAL)
7219
      elif self.op.snode:
7220
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
7221
                        " template")
7222
        self.op.snode = None
7223

    
7224
    self._cds = _GetClusterDomainSecret()
7225

    
7226
    if self.op.mode == constants.INSTANCE_IMPORT:
7227
      # On import force_variant must be True, because if we forced it at
7228
      # initial install, our only chance when importing it back is that it
7229
      # works again!
7230
      self.op.force_variant = True
7231

    
7232
      if self.op.no_install:
7233
        self.LogInfo("No-installation mode has no effect during import")
7234

    
7235
    elif self.op.mode == constants.INSTANCE_CREATE:
7236
      if self.op.os_type is None:
7237
        raise errors.OpPrereqError("No guest OS specified",
7238
                                   errors.ECODE_INVAL)
7239
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
7240
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
7241
                                   " installation" % self.op.os_type,
7242
                                   errors.ECODE_STATE)
7243
      if self.op.disk_template is None:
7244
        raise errors.OpPrereqError("No disk template specified",
7245
                                   errors.ECODE_INVAL)
7246

    
7247
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7248
      # Check handshake to ensure both clusters have the same domain secret
7249
      src_handshake = self.op.source_handshake
7250
      if not src_handshake:
7251
        raise errors.OpPrereqError("Missing source handshake",
7252
                                   errors.ECODE_INVAL)
7253

    
7254
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
7255
                                                           src_handshake)
7256
      if errmsg:
7257
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
7258
                                   errors.ECODE_INVAL)
7259

    
7260
      # Load and check source CA
7261
      self.source_x509_ca_pem = self.op.source_x509_ca
7262
      if not self.source_x509_ca_pem:
7263
        raise errors.OpPrereqError("Missing source X509 CA",
7264
                                   errors.ECODE_INVAL)
7265

    
7266
      try:
7267
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
7268
                                                    self._cds)
7269
      except OpenSSL.crypto.Error, err:
7270
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
7271
                                   (err, ), errors.ECODE_INVAL)
7272

    
7273
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
7274
      if errcode is not None:
7275
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
7276
                                   errors.ECODE_INVAL)
7277

    
7278
      self.source_x509_ca = cert
7279

    
7280
      src_instance_name = self.op.source_instance_name
7281
      if not src_instance_name:
7282
        raise errors.OpPrereqError("Missing source instance name",
7283
                                   errors.ECODE_INVAL)
7284

    
7285
      self.source_instance_name = \
7286
          netutils.GetHostname(name=src_instance_name).name
7287

    
7288
    else:
7289
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
7290
                                 self.op.mode, errors.ECODE_INVAL)
7291

    
7292
  def ExpandNames(self):
7293
    """ExpandNames for CreateInstance.
7294

7295
    Figure out the right locks for instance creation.
7296

7297
    """
7298
    self.needed_locks = {}
7299

    
7300
    instance_name = self.op.instance_name
7301
    # this is just a preventive check, but someone might still add this
7302
    # instance in the meantime, and creation will fail at lock-add time
7303
    if instance_name in self.cfg.GetInstanceList():
7304
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7305
                                 instance_name, errors.ECODE_EXISTS)
7306

    
7307
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
7308

    
7309
    if self.op.iallocator:
7310
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7311
    else:
7312
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
7313
      nodelist = [self.op.pnode]
7314
      if self.op.snode is not None:
7315
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
7316
        nodelist.append(self.op.snode)
7317
      self.needed_locks[locking.LEVEL_NODE] = nodelist
7318

    
7319
    # in case of import lock the source node too
7320
    if self.op.mode == constants.INSTANCE_IMPORT:
7321
      src_node = self.op.src_node
7322
      src_path = self.op.src_path
7323

    
7324
      if src_path is None:
7325
        self.op.src_path = src_path = self.op.instance_name
7326

    
7327
      if src_node is None:
7328
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7329
        self.op.src_node = None
7330
        if os.path.isabs(src_path):
7331
          raise errors.OpPrereqError("Importing an instance from an absolute"
7332
                                     " path requires a source node option.",
7333
                                     errors.ECODE_INVAL)
7334
      else:
7335
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
7336
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
7337
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
7338
        if not os.path.isabs(src_path):
7339
          self.op.src_path = src_path = \
7340
            utils.PathJoin(constants.EXPORT_DIR, src_path)
7341

    
7342
  def _RunAllocator(self):
7343
    """Run the allocator based on input opcode.
7344

7345
    """
7346
    nics = [n.ToDict() for n in self.nics]
7347
    ial = IAllocator(self.cfg, self.rpc,
7348
                     mode=constants.IALLOCATOR_MODE_ALLOC,
7349
                     name=self.op.instance_name,
7350
                     disk_template=self.op.disk_template,
7351
                     tags=[],
7352
                     os=self.op.os_type,
7353
                     vcpus=self.be_full[constants.BE_VCPUS],
7354
                     mem_size=self.be_full[constants.BE_MEMORY],
7355
                     disks=self.disks,
7356
                     nics=nics,
7357
                     hypervisor=self.op.hypervisor,
7358
                     )
7359

    
7360
    ial.Run(self.op.iallocator)
7361

    
7362
    if not ial.success:
7363
      raise errors.OpPrereqError("Can't compute nodes using"
7364
                                 " iallocator '%s': %s" %
7365
                                 (self.op.iallocator, ial.info),
7366
                                 errors.ECODE_NORES)
7367
    if len(ial.result) != ial.required_nodes:
7368
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
7369
                                 " of nodes (%s), required %s" %
7370
                                 (self.op.iallocator, len(ial.result),
7371
                                  ial.required_nodes), errors.ECODE_FAULT)
7372
    self.op.pnode = ial.result[0]
7373
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
7374
                 self.op.instance_name, self.op.iallocator,
7375
                 utils.CommaJoin(ial.result))
7376
    if ial.required_nodes == 2:
7377
      self.op.snode = ial.result[1]
7378

    
7379
  def BuildHooksEnv(self):
7380
    """Build hooks env.
7381

7382
    This runs on master, primary and secondary nodes of the instance.
7383

7384
    """
7385
    env = {
7386
      "ADD_MODE": self.op.mode,
7387
      }
7388
    if self.op.mode == constants.INSTANCE_IMPORT:
7389
      env["SRC_NODE"] = self.op.src_node
7390
      env["SRC_PATH"] = self.op.src_path
7391
      env["SRC_IMAGES"] = self.src_images
7392

    
7393
    env.update(_BuildInstanceHookEnv(
7394
      name=self.op.instance_name,
7395
      primary_node=self.op.pnode,
7396
      secondary_nodes=self.secondaries,
7397
      status=self.op.start,
7398
      os_type=self.op.os_type,
7399
      memory=self.be_full[constants.BE_MEMORY],
7400
      vcpus=self.be_full[constants.BE_VCPUS],
7401
      nics=_NICListToTuple(self, self.nics),
7402
      disk_template=self.op.disk_template,
7403
      disks=[(d["size"], d["mode"]) for d in self.disks],
7404
      bep=self.be_full,
7405
      hvp=self.hv_full,
7406
      hypervisor_name=self.op.hypervisor,
7407
    ))
7408

    
7409
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
7410
          self.secondaries)
7411
    return env, nl, nl
7412

    
7413
  def _ReadExportInfo(self):
7414
    """Reads the export information from disk.
7415

7416
    It will override the opcode source node and path with the actual
7417
    information, if these two were not specified before.
7418

7419
    @return: the export information
7420

7421
    """
7422
    assert self.op.mode == constants.INSTANCE_IMPORT
7423

    
7424
    src_node = self.op.src_node
7425
    src_path = self.op.src_path
7426

    
7427
    if src_node is None:
7428
      locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
7429
      exp_list = self.rpc.call_export_list(locked_nodes)
7430
      found = False
7431
      for node in exp_list:
7432
        if exp_list[node].fail_msg:
7433
          continue
7434
        if src_path in exp_list[node].payload:
7435
          found = True
7436
          self.op.src_node = src_node = node
7437
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
7438
                                                       src_path)
7439
          break
7440
      if not found:
7441
        raise errors.OpPrereqError("No export found for relative path %s" %
7442
                                    src_path, errors.ECODE_INVAL)
7443

    
7444
    _CheckNodeOnline(self, src_node)
7445
    result = self.rpc.call_export_info(src_node, src_path)
7446
    result.Raise("No export or invalid export found in dir %s" % src_path)
7447

    
7448
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
7449
    if not export_info.has_section(constants.INISECT_EXP):
7450
      raise errors.ProgrammerError("Corrupted export config",
7451
                                   errors.ECODE_ENVIRON)
7452

    
7453
    ei_version = export_info.get(constants.INISECT_EXP, "version")
7454
    if (int(ei_version) != constants.EXPORT_VERSION):
7455
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
7456
                                 (ei_version, constants.EXPORT_VERSION),
7457
                                 errors.ECODE_ENVIRON)
7458
    return export_info
7459

    
7460
  def _ReadExportParams(self, einfo):
7461
    """Use export parameters as defaults.
7462

7463
    In case the opcode doesn't specify (as in override) some instance
7464
    parameters, then try to use them from the export information, if
7465
    that declares them.
7466

7467
    """
7468
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
7469

    
7470
    if self.op.disk_template is None:
7471
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
7472
        self.op.disk_template = einfo.get(constants.INISECT_INS,
7473
                                          "disk_template")
7474
      else:
7475
        raise errors.OpPrereqError("No disk template specified and the export"
7476
                                   " is missing the disk_template information",
7477
                                   errors.ECODE_INVAL)
7478

    
7479
    if not self.op.disks:
7480
      if einfo.has_option(constants.INISECT_INS, "disk_count"):
7481
        disks = []
7482
        # TODO: import the disk iv_name too
7483
        for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
7484
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
7485
          disks.append({"size": disk_sz})
7486
        self.op.disks = disks
7487
      else:
7488
        raise errors.OpPrereqError("No disk info specified and the export"
7489
                                   " is missing the disk information",
7490
                                   errors.ECODE_INVAL)
7491

    
7492
    if (not self.op.nics and
7493
        einfo.has_option(constants.INISECT_INS, "nic_count")):
7494
      nics = []
7495
      for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
7496
        ndict = {}
7497
        for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
7498
          v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
7499
          ndict[name] = v
7500
        nics.append(ndict)
7501
      self.op.nics = nics
7502

    
7503
    if (self.op.hypervisor is None and
7504
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
7505
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
7506
    if einfo.has_section(constants.INISECT_HYP):
7507
      # use the export parameters but do not override the ones
7508
      # specified by the user
7509
      for name, value in einfo.items(constants.INISECT_HYP):
7510
        if name not in self.op.hvparams:
7511
          self.op.hvparams[name] = value
7512

    
7513
    if einfo.has_section(constants.INISECT_BEP):
7514
      # use the parameters, without overriding
7515
      for name, value in einfo.items(constants.INISECT_BEP):
7516
        if name not in self.op.beparams:
7517
          self.op.beparams[name] = value
7518
    else:
7519
      # try to read the parameters old style, from the main section
7520
      for name in constants.BES_PARAMETERS:
7521
        if (name not in self.op.beparams and
7522
            einfo.has_option(constants.INISECT_INS, name)):
7523
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
7524

    
7525
    if einfo.has_section(constants.INISECT_OSP):
7526
      # use the parameters, without overriding
7527
      for name, value in einfo.items(constants.INISECT_OSP):
7528
        if name not in self.op.osparams:
7529
          self.op.osparams[name] = value
7530

    
7531
  def _RevertToDefaults(self, cluster):
7532
    """Revert the instance parameters to the default values.
7533

7534
    """
7535
    # hvparams
7536
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
7537
    for name in self.op.hvparams.keys():
7538
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
7539
        del self.op.hvparams[name]
7540
    # beparams
7541
    be_defs = cluster.SimpleFillBE({})
7542
    for name in self.op.beparams.keys():
7543
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
7544
        del self.op.beparams[name]
7545
    # nic params
7546
    nic_defs = cluster.SimpleFillNIC({})
7547
    for nic in self.op.nics:
7548
      for name in constants.NICS_PARAMETERS:
7549
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
7550
          del nic[name]
7551
    # osparams
7552
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
7553
    for name in self.op.osparams.keys():
7554
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
7555
        del self.op.osparams[name]
7556

    
7557
  def CheckPrereq(self):
7558
    """Check prerequisites.
7559

7560
    """
7561
    if self.op.mode == constants.INSTANCE_IMPORT:
7562
      export_info = self._ReadExportInfo()
7563
      self._ReadExportParams(export_info)
7564

    
7565
    if (not self.cfg.GetVGName() and
7566
        self.op.disk_template not in constants.DTS_NOT_LVM):
7567
      raise errors.OpPrereqError("Cluster does not support lvm-based"
7568
                                 " instances", errors.ECODE_STATE)
7569

    
7570
    if self.op.hypervisor is None:
7571
      self.op.hypervisor = self.cfg.GetHypervisorType()
7572

    
7573
    cluster = self.cfg.GetClusterInfo()
7574
    enabled_hvs = cluster.enabled_hypervisors
7575
    if self.op.hypervisor not in enabled_hvs:
7576
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
7577
                                 " cluster (%s)" % (self.op.hypervisor,
7578
                                  ",".join(enabled_hvs)),
7579
                                 errors.ECODE_STATE)
7580

    
7581
    # check hypervisor parameter syntax (locally)
7582
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
7583
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
7584
                                      self.op.hvparams)
7585
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
7586
    hv_type.CheckParameterSyntax(filled_hvp)
7587
    self.hv_full = filled_hvp
7588
    # check that we don't specify global parameters on an instance
7589
    _CheckGlobalHvParams(self.op.hvparams)
7590

    
7591
    # fill and remember the beparams dict
7592
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
7593
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
7594

    
7595
    # build os parameters
7596
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
7597

    
7598
    # now that hvp/bep are in final format, let's reset to defaults,
7599
    # if told to do so
7600
    if self.op.identify_defaults:
7601
      self._RevertToDefaults(cluster)
7602

    
7603
    # NIC buildup
7604
    self.nics = []
7605
    for idx, nic in enumerate(self.op.nics):
7606
      nic_mode_req = nic.get("mode", None)
7607
      nic_mode = nic_mode_req
7608
      if nic_mode is None:
7609
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
7610

    
7611
      # in routed mode, for the first nic, the default ip is 'auto'
7612
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
7613
        default_ip_mode = constants.VALUE_AUTO
7614
      else:
7615
        default_ip_mode = constants.VALUE_NONE
7616

    
7617
      # ip validity checks
7618
      ip = nic.get("ip", default_ip_mode)
7619
      if ip is None or ip.lower() == constants.VALUE_NONE:
7620
        nic_ip = None
7621
      elif ip.lower() == constants.VALUE_AUTO:
7622
        if not self.op.name_check:
7623
          raise errors.OpPrereqError("IP address set to auto but name checks"
7624
                                     " have been skipped",
7625
                                     errors.ECODE_INVAL)
7626
        nic_ip = self.hostname1.ip
7627
      else:
7628
        if not netutils.IPAddress.IsValid(ip):
7629
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
7630
                                     errors.ECODE_INVAL)
7631
        nic_ip = ip
7632

    
7633
      # TODO: check the ip address for uniqueness
7634
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
7635
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
7636
                                   errors.ECODE_INVAL)
7637

    
7638
      # MAC address verification
7639
      mac = nic.get("mac", constants.VALUE_AUTO)
7640
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7641
        mac = utils.NormalizeAndValidateMac(mac)
7642

    
7643
        try:
7644
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
7645
        except errors.ReservationError:
7646
          raise errors.OpPrereqError("MAC address %s already in use"
7647
                                     " in cluster" % mac,
7648
                                     errors.ECODE_NOTUNIQUE)
7649

    
7650
      # bridge verification
7651
      bridge = nic.get("bridge", None)
7652
      link = nic.get("link", None)
7653
      if bridge and link:
7654
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
7655
                                   " at the same time", errors.ECODE_INVAL)
7656
      elif bridge and nic_mode == constants.NIC_MODE_ROUTED:
7657
        raise errors.OpPrereqError("Cannot pass 'bridge' on a routed nic",
7658
                                   errors.ECODE_INVAL)
7659
      elif bridge:
7660
        link = bridge
7661

    
7662
      nicparams = {}
7663
      if nic_mode_req:
7664
        nicparams[constants.NIC_MODE] = nic_mode_req
7665
      if link:
7666
        nicparams[constants.NIC_LINK] = link
7667

    
7668
      check_params = cluster.SimpleFillNIC(nicparams)
7669
      objects.NIC.CheckParameterSyntax(check_params)
7670
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
7671

    
7672
    # disk checks/pre-build
7673
    self.disks = []
7674
    for disk in self.op.disks:
7675
      mode = disk.get("mode", constants.DISK_RDWR)
7676
      if mode not in constants.DISK_ACCESS_SET:
7677
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
7678
                                   mode, errors.ECODE_INVAL)
7679
      size = disk.get("size", None)
7680
      if size is None:
7681
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
7682
      try:
7683
        size = int(size)
7684
      except (TypeError, ValueError):
7685
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
7686
                                   errors.ECODE_INVAL)
7687
      vg = disk.get("vg", self.cfg.GetVGName())
7688
      new_disk = {"size": size, "mode": mode, "vg": vg}
7689
      if "adopt" in disk:
7690
        new_disk["adopt"] = disk["adopt"]
7691
      self.disks.append(new_disk)
7692

    
7693
    if self.op.mode == constants.INSTANCE_IMPORT:
7694

    
7695
      # Check that the new instance doesn't have less disks than the export
7696
      instance_disks = len(self.disks)
7697
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
7698
      if instance_disks < export_disks:
7699
        raise errors.OpPrereqError("Not enough disks to import."
7700
                                   " (instance: %d, export: %d)" %
7701
                                   (instance_disks, export_disks),
7702
                                   errors.ECODE_INVAL)
7703

    
7704
      disk_images = []
7705
      for idx in range(export_disks):
7706
        option = 'disk%d_dump' % idx
7707
        if export_info.has_option(constants.INISECT_INS, option):
7708
          # FIXME: are the old os-es, disk sizes, etc. useful?
7709
          export_name = export_info.get(constants.INISECT_INS, option)
7710
          image = utils.PathJoin(self.op.src_path, export_name)
7711
          disk_images.append(image)
7712
        else:
7713
          disk_images.append(False)
7714

    
7715
      self.src_images = disk_images
7716

    
7717
      old_name = export_info.get(constants.INISECT_INS, 'name')
7718
      try:
7719
        exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
7720
      except (TypeError, ValueError), err:
7721
        raise errors.OpPrereqError("Invalid export file, nic_count is not"
7722
                                   " an integer: %s" % str(err),
7723
                                   errors.ECODE_STATE)
7724
      if self.op.instance_name == old_name:
7725
        for idx, nic in enumerate(self.nics):
7726
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
7727
            nic_mac_ini = 'nic%d_mac' % idx
7728
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
7729

    
7730
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
7731

    
7732
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
7733
    if self.op.ip_check:
7734
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
7735
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
7736
                                   (self.check_ip, self.op.instance_name),
7737
                                   errors.ECODE_NOTUNIQUE)
7738

    
7739
    #### mac address generation
7740
    # By generating here the mac address both the allocator and the hooks get
7741
    # the real final mac address rather than the 'auto' or 'generate' value.
7742
    # There is a race condition between the generation and the instance object
7743
    # creation, which means that we know the mac is valid now, but we're not
7744
    # sure it will be when we actually add the instance. If things go bad
7745
    # adding the instance will abort because of a duplicate mac, and the
7746
    # creation job will fail.
7747
    for nic in self.nics:
7748
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7749
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
7750

    
7751
    #### allocator run
7752

    
7753
    if self.op.iallocator is not None:
7754
      self._RunAllocator()
7755

    
7756
    #### node related checks
7757

    
7758
    # check primary node
7759
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
7760
    assert self.pnode is not None, \
7761
      "Cannot retrieve locked node %s" % self.op.pnode
7762
    if pnode.offline:
7763
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
7764
                                 pnode.name, errors.ECODE_STATE)
7765
    if pnode.drained:
7766
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
7767
                                 pnode.name, errors.ECODE_STATE)
7768
    if not pnode.vm_capable:
7769
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
7770
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
7771

    
7772
    self.secondaries = []
7773

    
7774
    # mirror node verification
7775
    if self.op.disk_template in constants.DTS_INT_MIRROR:
7776
      if self.op.snode == pnode.name:
7777
        raise errors.OpPrereqError("The secondary node cannot be the"
7778
                                   " primary node.", errors.ECODE_INVAL)
7779
      _CheckNodeOnline(self, self.op.snode)
7780
      _CheckNodeNotDrained(self, self.op.snode)
7781
      _CheckNodeVmCapable(self, self.op.snode)
7782
      self.secondaries.append(self.op.snode)
7783

    
7784
    nodenames = [pnode.name] + self.secondaries
7785

    
7786
    if not self.adopt_disks:
7787
      # Check lv size requirements, if not adopting
7788
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
7789
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
7790

    
7791
    elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
7792
      all_lvs = set([i["vg"] + "/" + i["adopt"] for i in self.disks])
7793
      if len(all_lvs) != len(self.disks):
7794
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
7795
                                   errors.ECODE_INVAL)
7796
      for lv_name in all_lvs:
7797
        try:
7798
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
7799
          # to ReserveLV uses the same syntax
7800
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
7801
        except errors.ReservationError:
7802
          raise errors.OpPrereqError("LV named %s used by another instance" %
7803
                                     lv_name, errors.ECODE_NOTUNIQUE)
7804

    
7805
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
7806
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
7807

    
7808
      node_lvs = self.rpc.call_lv_list([pnode.name],
7809
                                       vg_names.payload.keys())[pnode.name]
7810
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
7811
      node_lvs = node_lvs.payload
7812

    
7813
      delta = all_lvs.difference(node_lvs.keys())
7814
      if delta:
7815
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
7816
                                   utils.CommaJoin(delta),
7817
                                   errors.ECODE_INVAL)
7818
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
7819
      if online_lvs:
7820
        raise errors.OpPrereqError("Online logical volumes found, cannot"
7821
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
7822
                                   errors.ECODE_STATE)
7823
      # update the size of disk based on what is found
7824
      for dsk in self.disks:
7825
        dsk["size"] = int(float(node_lvs[dsk["vg"] + "/" + dsk["adopt"]][0]))
7826

    
7827
    elif self.op.disk_template == constants.DT_BLOCK:
7828
      # Normalize and de-duplicate device paths
7829
      all_disks = set([os.path.abspath(i["adopt"]) for i in self.disks])
7830
      if len(all_disks) != len(self.disks):
7831
        raise errors.OpPrereqError("Duplicate disk names given for adoption",
7832
                                   errors.ECODE_INVAL)
7833
      baddisks = [d for d in all_disks
7834
                  if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
7835
      if baddisks:
7836
        raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
7837
                                   " cannot be adopted" %
7838
                                   (", ".join(baddisks),
7839
                                    constants.ADOPTABLE_BLOCKDEV_ROOT),
7840
                                   errors.ECODE_INVAL)
7841

    
7842
      node_disks = self.rpc.call_bdev_sizes([pnode.name],
7843
                                            list(all_disks))[pnode.name]
7844
      node_disks.Raise("Cannot get block device information from node %s" %
7845
                       pnode.name)
7846
      node_disks = node_disks.payload
7847
      delta = all_disks.difference(node_disks.keys())
7848
      if delta:
7849
        raise errors.OpPrereqError("Missing block device(s): %s" %
7850
                                   utils.CommaJoin(delta),
7851
                                   errors.ECODE_INVAL)
7852
      for dsk in self.disks:
7853
        dsk["size"] = int(float(node_disks[dsk["adopt"]]))
7854

    
7855
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
7856

    
7857
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
7858
    # check OS parameters (remotely)
7859
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
7860

    
7861
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
7862

    
7863
    # memory check on primary node
7864
    if self.op.start:
7865
      _CheckNodeFreeMemory(self, self.pnode.name,
7866
                           "creating instance %s" % self.op.instance_name,
7867
                           self.be_full[constants.BE_MEMORY],
7868
                           self.op.hypervisor)
7869

    
7870
    self.dry_run_result = list(nodenames)
7871

    
7872
  def Exec(self, feedback_fn):
7873
    """Create and add the instance to the cluster.
7874

7875
    """
7876
    instance = self.op.instance_name
7877
    pnode_name = self.pnode.name
7878

    
7879
    ht_kind = self.op.hypervisor
7880
    if ht_kind in constants.HTS_REQ_PORT:
7881
      network_port = self.cfg.AllocatePort()
7882
    else:
7883
      network_port = None
7884

    
7885
    if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE:
7886
      # this is needed because os.path.join does not accept None arguments
7887
      if self.op.file_storage_dir is None:
7888
        string_file_storage_dir = ""
7889
      else:
7890
        string_file_storage_dir = self.op.file_storage_dir
7891

    
7892
      # build the full file storage dir path
7893
      if self.op.disk_template == constants.DT_SHARED_FILE:
7894
        get_fsd_fn = self.cfg.GetSharedFileStorageDir
7895
      else:
7896
        get_fsd_fn = self.cfg.GetFileStorageDir
7897

    
7898
      file_storage_dir = utils.PathJoin(get_fsd_fn(),
7899
                                        string_file_storage_dir, instance)
7900
    else:
7901
      file_storage_dir = ""
7902

    
7903
    disks = _GenerateDiskTemplate(self,
7904
                                  self.op.disk_template,
7905
                                  instance, pnode_name,
7906
                                  self.secondaries,
7907
                                  self.disks,
7908
                                  file_storage_dir,
7909
                                  self.op.file_driver,
7910
                                  0,
7911
                                  feedback_fn)
7912

    
7913
    iobj = objects.Instance(name=instance, os=self.op.os_type,
7914
                            primary_node=pnode_name,
7915
                            nics=self.nics, disks=disks,
7916
                            disk_template=self.op.disk_template,
7917
                            admin_up=False,
7918
                            network_port=network_port,
7919
                            beparams=self.op.beparams,
7920
                            hvparams=self.op.hvparams,
7921
                            hypervisor=self.op.hypervisor,
7922
                            osparams=self.op.osparams,
7923
                            )
7924

    
7925
    if self.adopt_disks:
7926
      if self.op.disk_template == constants.DT_PLAIN:
7927
        # rename LVs to the newly-generated names; we need to construct
7928
        # 'fake' LV disks with the old data, plus the new unique_id
7929
        tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
7930
        rename_to = []
7931
        for t_dsk, a_dsk in zip (tmp_disks, self.disks):
7932
          rename_to.append(t_dsk.logical_id)
7933
          t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk["adopt"])
7934
          self.cfg.SetDiskID(t_dsk, pnode_name)
7935
        result = self.rpc.call_blockdev_rename(pnode_name,
7936
                                               zip(tmp_disks, rename_to))
7937
        result.Raise("Failed to rename adoped LVs")
7938
    else:
7939
      feedback_fn("* creating instance disks...")
7940
      try:
7941
        _CreateDisks(self, iobj)
7942
      except errors.OpExecError:
7943
        self.LogWarning("Device creation failed, reverting...")
7944
        try:
7945
          _RemoveDisks(self, iobj)
7946
        finally:
7947
          self.cfg.ReleaseDRBDMinors(instance)
7948
          raise
7949

    
7950
      if self.cfg.GetClusterInfo().prealloc_wipe_disks:
7951
        feedback_fn("* wiping instance disks...")
7952
        try:
7953
          _WipeDisks(self, iobj)
7954
        except errors.OpExecError:
7955
          self.LogWarning("Device wiping failed, reverting...")
7956
          try:
7957
            _RemoveDisks(self, iobj)
7958
          finally:
7959
            self.cfg.ReleaseDRBDMinors(instance)
7960
            raise
7961

    
7962
    feedback_fn("adding instance %s to cluster config" % instance)
7963

    
7964
    self.cfg.AddInstance(iobj, self.proc.GetECId())
7965

    
7966
    # Declare that we don't want to remove the instance lock anymore, as we've
7967
    # added the instance to the config
7968
    del self.remove_locks[locking.LEVEL_INSTANCE]
7969
    # Unlock all the nodes
7970
    if self.op.mode == constants.INSTANCE_IMPORT:
7971
      nodes_keep = [self.op.src_node]
7972
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
7973
                       if node != self.op.src_node]
7974
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
7975
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
7976
    else:
7977
      self.context.glm.release(locking.LEVEL_NODE)
7978
      del self.acquired_locks[locking.LEVEL_NODE]
7979

    
7980
    if self.op.wait_for_sync:
7981
      disk_abort = not _WaitForSync(self, iobj)
7982
    elif iobj.disk_template in constants.DTS_INT_MIRROR:
7983
      # make sure the disks are not degraded (still sync-ing is ok)
7984
      time.sleep(15)
7985
      feedback_fn("* checking mirrors status")
7986
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
7987
    else:
7988
      disk_abort = False
7989

    
7990
    if disk_abort:
7991
      _RemoveDisks(self, iobj)
7992
      self.cfg.RemoveInstance(iobj.name)
7993
      # Make sure the instance lock gets removed
7994
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
7995
      raise errors.OpExecError("There are some degraded disks for"
7996
                               " this instance")
7997

    
7998
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
7999
      if self.op.mode == constants.INSTANCE_CREATE:
8000
        if not self.op.no_install:
8001
          feedback_fn("* running the instance OS create scripts...")
8002
          # FIXME: pass debug option from opcode to backend
8003
          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
8004
                                                 self.op.debug_level)
8005
          result.Raise("Could not add os for instance %s"
8006
                       " on node %s" % (instance, pnode_name))
8007

    
8008
      elif self.op.mode == constants.INSTANCE_IMPORT:
8009
        feedback_fn("* running the instance OS import scripts...")
8010

    
8011
        transfers = []
8012

    
8013
        for idx, image in enumerate(self.src_images):
8014
          if not image:
8015
            continue
8016

    
8017
          # FIXME: pass debug option from opcode to backend
8018
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
8019
                                             constants.IEIO_FILE, (image, ),
8020
                                             constants.IEIO_SCRIPT,
8021
                                             (iobj.disks[idx], idx),
8022
                                             None)
8023
          transfers.append(dt)
8024

    
8025
        import_result = \
8026
          masterd.instance.TransferInstanceData(self, feedback_fn,
8027
                                                self.op.src_node, pnode_name,
8028
                                                self.pnode.secondary_ip,
8029
                                                iobj, transfers)
8030
        if not compat.all(import_result):
8031
          self.LogWarning("Some disks for instance %s on node %s were not"
8032
                          " imported successfully" % (instance, pnode_name))
8033

    
8034
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
8035
        feedback_fn("* preparing remote import...")
8036
        # The source cluster will stop the instance before attempting to make a
8037
        # connection. In some cases stopping an instance can take a long time,
8038
        # hence the shutdown timeout is added to the connection timeout.
8039
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
8040
                           self.op.source_shutdown_timeout)
8041
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
8042

    
8043
        assert iobj.primary_node == self.pnode.name
8044
        disk_results = \
8045
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
8046
                                        self.source_x509_ca,
8047
                                        self._cds, timeouts)
8048
        if not compat.all(disk_results):
8049
          # TODO: Should the instance still be started, even if some disks
8050
          # failed to import (valid for local imports, too)?
8051
          self.LogWarning("Some disks for instance %s on node %s were not"
8052
                          " imported successfully" % (instance, pnode_name))
8053

    
8054
        # Run rename script on newly imported instance
8055
        assert iobj.name == instance
8056
        feedback_fn("Running rename script for %s" % instance)
8057
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
8058
                                                   self.source_instance_name,
8059
                                                   self.op.debug_level)
8060
        if result.fail_msg:
8061
          self.LogWarning("Failed to run rename script for %s on node"
8062
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
8063

    
8064
      else:
8065
        # also checked in the prereq part
8066
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
8067
                                     % self.op.mode)
8068

    
8069
    if self.op.start:
8070
      iobj.admin_up = True
8071
      self.cfg.Update(iobj, feedback_fn)
8072
      logging.info("Starting instance %s on node %s", instance, pnode_name)
8073
      feedback_fn("* starting instance...")
8074
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
8075
      result.Raise("Could not start instance")
8076

    
8077
    return list(iobj.all_nodes)
8078

    
8079

    
8080
class LUInstanceConsole(NoHooksLU):
8081
  """Connect to an instance's console.
8082

8083
  This is somewhat special in that it returns the command line that
8084
  you need to run on the master node in order to connect to the
8085
  console.
8086

8087
  """
8088
  REQ_BGL = False
8089

    
8090
  def ExpandNames(self):
8091
    self._ExpandAndLockInstance()
8092

    
8093
  def CheckPrereq(self):
8094
    """Check prerequisites.
8095

8096
    This checks that the instance is in the cluster.
8097

8098
    """
8099
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8100
    assert self.instance is not None, \
8101
      "Cannot retrieve locked instance %s" % self.op.instance_name
8102
    _CheckNodeOnline(self, self.instance.primary_node)
8103

    
8104
  def Exec(self, feedback_fn):
8105
    """Connect to the console of an instance
8106

8107
    """
8108
    instance = self.instance
8109
    node = instance.primary_node
8110

    
8111
    node_insts = self.rpc.call_instance_list([node],
8112
                                             [instance.hypervisor])[node]
8113
    node_insts.Raise("Can't get node information from %s" % node)
8114

    
8115
    if instance.name not in node_insts.payload:
8116
      if instance.admin_up:
8117
        state = "ERROR_down"
8118
      else:
8119
        state = "ADMIN_down"
8120
      raise errors.OpExecError("Instance %s is not running (state %s)" %
8121
                               (instance.name, state))
8122

    
8123
    logging.debug("Connecting to console of %s on %s", instance.name, node)
8124

    
8125
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
8126

    
8127

    
8128
def _GetInstanceConsole(cluster, instance):
8129
  """Returns console information for an instance.
8130

8131
  @type cluster: L{objects.Cluster}
8132
  @type instance: L{objects.Instance}
8133
  @rtype: dict
8134

8135
  """
8136
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
8137
  # beparams and hvparams are passed separately, to avoid editing the
8138
  # instance and then saving the defaults in the instance itself.
8139
  hvparams = cluster.FillHV(instance)
8140
  beparams = cluster.FillBE(instance)
8141
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
8142

    
8143
  assert console.instance == instance.name
8144
  assert console.Validate()
8145

    
8146
  return console.ToDict()
8147

    
8148

    
8149
class LUInstanceReplaceDisks(LogicalUnit):
8150
  """Replace the disks of an instance.
8151

8152
  """
8153
  HPATH = "mirrors-replace"
8154
  HTYPE = constants.HTYPE_INSTANCE
8155
  REQ_BGL = False
8156

    
8157
  def CheckArguments(self):
8158
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
8159
                                  self.op.iallocator)
8160

    
8161
  def ExpandNames(self):
8162
    self._ExpandAndLockInstance()
8163

    
8164
    if self.op.iallocator is not None:
8165
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8166

    
8167
    elif self.op.remote_node is not None:
8168
      remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8169
      self.op.remote_node = remote_node
8170

    
8171
      # Warning: do not remove the locking of the new secondary here
8172
      # unless DRBD8.AddChildren is changed to work in parallel;
8173
      # currently it doesn't since parallel invocations of
8174
      # FindUnusedMinor will conflict
8175
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
8176
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
8177

    
8178
    else:
8179
      self.needed_locks[locking.LEVEL_NODE] = []
8180
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8181

    
8182
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
8183
                                   self.op.iallocator, self.op.remote_node,
8184
                                   self.op.disks, False, self.op.early_release)
8185

    
8186
    self.tasklets = [self.replacer]
8187

    
8188
  def DeclareLocks(self, level):
8189
    # If we're not already locking all nodes in the set we have to declare the
8190
    # instance's primary/secondary nodes.
8191
    if (level == locking.LEVEL_NODE and
8192
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
8193
      self._LockInstancesNodes()
8194

    
8195
  def BuildHooksEnv(self):
8196
    """Build hooks env.
8197

8198
    This runs on the master, the primary and all the secondaries.
8199

8200
    """
8201
    instance = self.replacer.instance
8202
    env = {
8203
      "MODE": self.op.mode,
8204
      "NEW_SECONDARY": self.op.remote_node,
8205
      "OLD_SECONDARY": instance.secondary_nodes[0],
8206
      }
8207
    env.update(_BuildInstanceHookEnvByObject(self, instance))
8208
    nl = [
8209
      self.cfg.GetMasterNode(),
8210
      instance.primary_node,
8211
      ]
8212
    if self.op.remote_node is not None:
8213
      nl.append(self.op.remote_node)
8214
    return env, nl, nl
8215

    
8216

    
8217
class TLReplaceDisks(Tasklet):
8218
  """Replaces disks for an instance.
8219

8220
  Note: Locking is not within the scope of this class.
8221

8222
  """
8223
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
8224
               disks, delay_iallocator, early_release):
8225
    """Initializes this class.
8226

8227
    """
8228
    Tasklet.__init__(self, lu)
8229

    
8230
    # Parameters
8231
    self.instance_name = instance_name
8232
    self.mode = mode
8233
    self.iallocator_name = iallocator_name
8234
    self.remote_node = remote_node
8235
    self.disks = disks
8236
    self.delay_iallocator = delay_iallocator
8237
    self.early_release = early_release
8238

    
8239
    # Runtime data
8240
    self.instance = None
8241
    self.new_node = None
8242
    self.target_node = None
8243
    self.other_node = None
8244
    self.remote_node_info = None
8245
    self.node_secondary_ip = None
8246

    
8247
  @staticmethod
8248
  def CheckArguments(mode, remote_node, iallocator):
8249
    """Helper function for users of this class.
8250

8251
    """
8252
    # check for valid parameter combination
8253
    if mode == constants.REPLACE_DISK_CHG:
8254
      if remote_node is None and iallocator is None:
8255
        raise errors.OpPrereqError("When changing the secondary either an"
8256
                                   " iallocator script must be used or the"
8257
                                   " new node given", errors.ECODE_INVAL)
8258

    
8259
      if remote_node is not None and iallocator is not None:
8260
        raise errors.OpPrereqError("Give either the iallocator or the new"
8261
                                   " secondary, not both", errors.ECODE_INVAL)
8262

    
8263
    elif remote_node is not None or iallocator is not None:
8264
      # Not replacing the secondary
8265
      raise errors.OpPrereqError("The iallocator and new node options can"
8266
                                 " only be used when changing the"
8267
                                 " secondary node", errors.ECODE_INVAL)
8268

    
8269
  @staticmethod
8270
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
8271
    """Compute a new secondary node using an IAllocator.
8272

8273
    """
8274
    ial = IAllocator(lu.cfg, lu.rpc,
8275
                     mode=constants.IALLOCATOR_MODE_RELOC,
8276
                     name=instance_name,
8277
                     relocate_from=relocate_from)
8278

    
8279
    ial.Run(iallocator_name)
8280

    
8281
    if not ial.success:
8282
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
8283
                                 " %s" % (iallocator_name, ial.info),
8284
                                 errors.ECODE_NORES)
8285

    
8286
    if len(ial.result) != ial.required_nodes:
8287
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8288
                                 " of nodes (%s), required %s" %
8289
                                 (iallocator_name,
8290
                                  len(ial.result), ial.required_nodes),
8291
                                 errors.ECODE_FAULT)
8292

    
8293
    remote_node_name = ial.result[0]
8294

    
8295
    lu.LogInfo("Selected new secondary for instance '%s': %s",
8296
               instance_name, remote_node_name)
8297

    
8298
    return remote_node_name
8299

    
8300
  def _FindFaultyDisks(self, node_name):
8301
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
8302
                                    node_name, True)
8303

    
8304
  def CheckPrereq(self):
8305
    """Check prerequisites.
8306

8307
    This checks that the instance is in the cluster.
8308

8309
    """
8310
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
8311
    assert instance is not None, \
8312
      "Cannot retrieve locked instance %s" % self.instance_name
8313

    
8314
    if instance.disk_template != constants.DT_DRBD8:
8315
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
8316
                                 " instances", errors.ECODE_INVAL)
8317

    
8318
    if len(instance.secondary_nodes) != 1:
8319
      raise errors.OpPrereqError("The instance has a strange layout,"
8320
                                 " expected one secondary but found %d" %
8321
                                 len(instance.secondary_nodes),
8322
                                 errors.ECODE_FAULT)
8323

    
8324
    if not self.delay_iallocator:
8325
      self._CheckPrereq2()
8326

    
8327
  def _CheckPrereq2(self):
8328
    """Check prerequisites, second part.
8329

8330
    This function should always be part of CheckPrereq. It was separated and is
8331
    now called from Exec because during node evacuation iallocator was only
8332
    called with an unmodified cluster model, not taking planned changes into
8333
    account.
8334

8335
    """
8336
    instance = self.instance
8337
    secondary_node = instance.secondary_nodes[0]
8338

    
8339
    if self.iallocator_name is None:
8340
      remote_node = self.remote_node
8341
    else:
8342
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
8343
                                       instance.name, instance.secondary_nodes)
8344

    
8345
    if remote_node is not None:
8346
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
8347
      assert self.remote_node_info is not None, \
8348
        "Cannot retrieve locked node %s" % remote_node
8349
    else:
8350
      self.remote_node_info = None
8351

    
8352
    if remote_node == self.instance.primary_node:
8353
      raise errors.OpPrereqError("The specified node is the primary node of"
8354
                                 " the instance.", errors.ECODE_INVAL)
8355

    
8356
    if remote_node == secondary_node:
8357
      raise errors.OpPrereqError("The specified node is already the"
8358
                                 " secondary node of the instance.",
8359
                                 errors.ECODE_INVAL)
8360

    
8361
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
8362
                                    constants.REPLACE_DISK_CHG):
8363
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
8364
                                 errors.ECODE_INVAL)
8365

    
8366
    if self.mode == constants.REPLACE_DISK_AUTO:
8367
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
8368
      faulty_secondary = self._FindFaultyDisks(secondary_node)
8369

    
8370
      if faulty_primary and faulty_secondary:
8371
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
8372
                                   " one node and can not be repaired"
8373
                                   " automatically" % self.instance_name,
8374
                                   errors.ECODE_STATE)
8375

    
8376
      if faulty_primary:
8377
        self.disks = faulty_primary
8378
        self.target_node = instance.primary_node
8379
        self.other_node = secondary_node
8380
        check_nodes = [self.target_node, self.other_node]
8381
      elif faulty_secondary:
8382
        self.disks = faulty_secondary
8383
        self.target_node = secondary_node
8384
        self.other_node = instance.primary_node
8385
        check_nodes = [self.target_node, self.other_node]
8386
      else:
8387
        self.disks = []
8388
        check_nodes = []
8389

    
8390
    else:
8391
      # Non-automatic modes
8392
      if self.mode == constants.REPLACE_DISK_PRI:
8393
        self.target_node = instance.primary_node
8394
        self.other_node = secondary_node
8395
        check_nodes = [self.target_node, self.other_node]
8396

    
8397
      elif self.mode == constants.REPLACE_DISK_SEC:
8398
        self.target_node = secondary_node
8399
        self.other_node = instance.primary_node
8400
        check_nodes = [self.target_node, self.other_node]
8401

    
8402
      elif self.mode == constants.REPLACE_DISK_CHG:
8403
        self.new_node = remote_node
8404
        self.other_node = instance.primary_node
8405
        self.target_node = secondary_node
8406
        check_nodes = [self.new_node, self.other_node]
8407

    
8408
        _CheckNodeNotDrained(self.lu, remote_node)
8409
        _CheckNodeVmCapable(self.lu, remote_node)
8410

    
8411
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
8412
        assert old_node_info is not None
8413
        if old_node_info.offline and not self.early_release:
8414
          # doesn't make sense to delay the release
8415
          self.early_release = True
8416
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
8417
                          " early-release mode", secondary_node)
8418

    
8419
      else:
8420
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
8421
                                     self.mode)
8422

    
8423
      # If not specified all disks should be replaced
8424
      if not self.disks:
8425
        self.disks = range(len(self.instance.disks))
8426

    
8427
    for node in check_nodes:
8428
      _CheckNodeOnline(self.lu, node)
8429

    
8430
    # Check whether disks are valid
8431
    for disk_idx in self.disks:
8432
      instance.FindDisk(disk_idx)
8433

    
8434
    # Get secondary node IP addresses
8435
    node_2nd_ip = {}
8436

    
8437
    for node_name in [self.target_node, self.other_node, self.new_node]:
8438
      if node_name is not None:
8439
        node_2nd_ip[node_name] = self.cfg.GetNodeInfo(node_name).secondary_ip
8440

    
8441
    self.node_secondary_ip = node_2nd_ip
8442

    
8443
  def Exec(self, feedback_fn):
8444
    """Execute disk replacement.
8445

8446
    This dispatches the disk replacement to the appropriate handler.
8447

8448
    """
8449
    if self.delay_iallocator:
8450
      self._CheckPrereq2()
8451

    
8452
    if not self.disks:
8453
      feedback_fn("No disks need replacement")
8454
      return
8455

    
8456
    feedback_fn("Replacing disk(s) %s for %s" %
8457
                (utils.CommaJoin(self.disks), self.instance.name))
8458

    
8459
    activate_disks = (not self.instance.admin_up)
8460

    
8461
    # Activate the instance disks if we're replacing them on a down instance
8462
    if activate_disks:
8463
      _StartInstanceDisks(self.lu, self.instance, True)
8464

    
8465
    try:
8466
      # Should we replace the secondary node?
8467
      if self.new_node is not None:
8468
        fn = self._ExecDrbd8Secondary
8469
      else:
8470
        fn = self._ExecDrbd8DiskOnly
8471

    
8472
      return fn(feedback_fn)
8473

    
8474
    finally:
8475
      # Deactivate the instance disks if we're replacing them on a
8476
      # down instance
8477
      if activate_disks:
8478
        _SafeShutdownInstanceDisks(self.lu, self.instance)
8479

    
8480
  def _CheckVolumeGroup(self, nodes):
8481
    self.lu.LogInfo("Checking volume groups")
8482

    
8483
    vgname = self.cfg.GetVGName()
8484

    
8485
    # Make sure volume group exists on all involved nodes
8486
    results = self.rpc.call_vg_list(nodes)
8487
    if not results:
8488
      raise errors.OpExecError("Can't list volume groups on the nodes")
8489

    
8490
    for node in nodes:
8491
      res = results[node]
8492
      res.Raise("Error checking node %s" % node)
8493
      if vgname not in res.payload:
8494
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
8495
                                 (vgname, node))
8496

    
8497
  def _CheckDisksExistence(self, nodes):
8498
    # Check disk existence
8499
    for idx, dev in enumerate(self.instance.disks):
8500
      if idx not in self.disks:
8501
        continue
8502

    
8503
      for node in nodes:
8504
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
8505
        self.cfg.SetDiskID(dev, node)
8506

    
8507
        result = self.rpc.call_blockdev_find(node, dev)
8508

    
8509
        msg = result.fail_msg
8510
        if msg or not result.payload:
8511
          if not msg:
8512
            msg = "disk not found"
8513
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
8514
                                   (idx, node, msg))
8515

    
8516
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
8517
    for idx, dev in enumerate(self.instance.disks):
8518
      if idx not in self.disks:
8519
        continue
8520

    
8521
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
8522
                      (idx, node_name))
8523

    
8524
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
8525
                                   ldisk=ldisk):
8526
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
8527
                                 " replace disks for instance %s" %
8528
                                 (node_name, self.instance.name))
8529

    
8530
  def _CreateNewStorage(self, node_name):
8531
    vgname = self.cfg.GetVGName()
8532
    iv_names = {}
8533

    
8534
    for idx, dev in enumerate(self.instance.disks):
8535
      if idx not in self.disks:
8536
        continue
8537

    
8538
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
8539

    
8540
      self.cfg.SetDiskID(dev, node_name)
8541

    
8542
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
8543
      names = _GenerateUniqueNames(self.lu, lv_names)
8544

    
8545
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
8546
                             logical_id=(vgname, names[0]))
8547
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
8548
                             logical_id=(vgname, names[1]))
8549

    
8550
      new_lvs = [lv_data, lv_meta]
8551
      old_lvs = dev.children
8552
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
8553

    
8554
      # we pass force_create=True to force the LVM creation
8555
      for new_lv in new_lvs:
8556
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
8557
                        _GetInstanceInfoText(self.instance), False)
8558

    
8559
    return iv_names
8560

    
8561
  def _CheckDevices(self, node_name, iv_names):
8562
    for name, (dev, _, _) in iv_names.iteritems():
8563
      self.cfg.SetDiskID(dev, node_name)
8564

    
8565
      result = self.rpc.call_blockdev_find(node_name, dev)
8566

    
8567
      msg = result.fail_msg
8568
      if msg or not result.payload:
8569
        if not msg:
8570
          msg = "disk not found"
8571
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
8572
                                 (name, msg))
8573

    
8574
      if result.payload.is_degraded:
8575
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
8576

    
8577
  def _RemoveOldStorage(self, node_name, iv_names):
8578
    for name, (_, old_lvs, _) in iv_names.iteritems():
8579
      self.lu.LogInfo("Remove logical volumes for %s" % name)
8580

    
8581
      for lv in old_lvs:
8582
        self.cfg.SetDiskID(lv, node_name)
8583

    
8584
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
8585
        if msg:
8586
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
8587
                             hint="remove unused LVs manually")
8588

    
8589
  def _ReleaseNodeLock(self, node_name):
8590
    """Releases the lock for a given node."""
8591
    self.lu.context.glm.release(locking.LEVEL_NODE, node_name)
8592

    
8593
  def _ExecDrbd8DiskOnly(self, feedback_fn):
8594
    """Replace a disk on the primary or secondary for DRBD 8.
8595

8596
    The algorithm for replace is quite complicated:
8597

8598
      1. for each disk to be replaced:
8599

8600
        1. create new LVs on the target node with unique names
8601
        1. detach old LVs from the drbd device
8602
        1. rename old LVs to name_replaced.<time_t>
8603
        1. rename new LVs to old LVs
8604
        1. attach the new LVs (with the old names now) to the drbd device
8605

8606
      1. wait for sync across all devices
8607

8608
      1. for each modified disk:
8609

8610
        1. remove old LVs (which have the name name_replaces.<time_t>)
8611

8612
    Failures are not very well handled.
8613

8614
    """
8615
    steps_total = 6
8616

    
8617
    # Step: check device activation
8618
    self.lu.LogStep(1, steps_total, "Check device existence")
8619
    self._CheckDisksExistence([self.other_node, self.target_node])
8620
    self._CheckVolumeGroup([self.target_node, self.other_node])
8621

    
8622
    # Step: check other node consistency
8623
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8624
    self._CheckDisksConsistency(self.other_node,
8625
                                self.other_node == self.instance.primary_node,
8626
                                False)
8627

    
8628
    # Step: create new storage
8629
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8630
    iv_names = self._CreateNewStorage(self.target_node)
8631

    
8632
    # Step: for each lv, detach+rename*2+attach
8633
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8634
    for dev, old_lvs, new_lvs in iv_names.itervalues():
8635
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
8636

    
8637
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
8638
                                                     old_lvs)
8639
      result.Raise("Can't detach drbd from local storage on node"
8640
                   " %s for device %s" % (self.target_node, dev.iv_name))
8641
      #dev.children = []
8642
      #cfg.Update(instance)
8643

    
8644
      # ok, we created the new LVs, so now we know we have the needed
8645
      # storage; as such, we proceed on the target node to rename
8646
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
8647
      # using the assumption that logical_id == physical_id (which in
8648
      # turn is the unique_id on that node)
8649

    
8650
      # FIXME(iustin): use a better name for the replaced LVs
8651
      temp_suffix = int(time.time())
8652
      ren_fn = lambda d, suff: (d.physical_id[0],
8653
                                d.physical_id[1] + "_replaced-%s" % suff)
8654

    
8655
      # Build the rename list based on what LVs exist on the node
8656
      rename_old_to_new = []
8657
      for to_ren in old_lvs:
8658
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
8659
        if not result.fail_msg and result.payload:
8660
          # device exists
8661
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
8662

    
8663
      self.lu.LogInfo("Renaming the old LVs on the target node")
8664
      result = self.rpc.call_blockdev_rename(self.target_node,
8665
                                             rename_old_to_new)
8666
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
8667

    
8668
      # Now we rename the new LVs to the old LVs
8669
      self.lu.LogInfo("Renaming the new LVs on the target node")
8670
      rename_new_to_old = [(new, old.physical_id)
8671
                           for old, new in zip(old_lvs, new_lvs)]
8672
      result = self.rpc.call_blockdev_rename(self.target_node,
8673
                                             rename_new_to_old)
8674
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
8675

    
8676
      for old, new in zip(old_lvs, new_lvs):
8677
        new.logical_id = old.logical_id
8678
        self.cfg.SetDiskID(new, self.target_node)
8679

    
8680
      for disk in old_lvs:
8681
        disk.logical_id = ren_fn(disk, temp_suffix)
8682
        self.cfg.SetDiskID(disk, self.target_node)
8683

    
8684
      # Now that the new lvs have the old name, we can add them to the device
8685
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
8686
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
8687
                                                  new_lvs)
8688
      msg = result.fail_msg
8689
      if msg:
8690
        for new_lv in new_lvs:
8691
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
8692
                                               new_lv).fail_msg
8693
          if msg2:
8694
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
8695
                               hint=("cleanup manually the unused logical"
8696
                                     "volumes"))
8697
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
8698

    
8699
      dev.children = new_lvs
8700

    
8701
      self.cfg.Update(self.instance, feedback_fn)
8702

    
8703
    cstep = 5
8704
    if self.early_release:
8705
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8706
      cstep += 1
8707
      self._RemoveOldStorage(self.target_node, iv_names)
8708
      # WARNING: we release both node locks here, do not do other RPCs
8709
      # than WaitForSync to the primary node
8710
      self._ReleaseNodeLock([self.target_node, self.other_node])
8711

    
8712
    # Wait for sync
8713
    # This can fail as the old devices are degraded and _WaitForSync
8714
    # does a combined result over all disks, so we don't check its return value
8715
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8716
    cstep += 1
8717
    _WaitForSync(self.lu, self.instance)
8718

    
8719
    # Check all devices manually
8720
    self._CheckDevices(self.instance.primary_node, iv_names)
8721

    
8722
    # Step: remove old storage
8723
    if not self.early_release:
8724
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8725
      cstep += 1
8726
      self._RemoveOldStorage(self.target_node, iv_names)
8727

    
8728
  def _ExecDrbd8Secondary(self, feedback_fn):
8729
    """Replace the secondary node for DRBD 8.
8730

8731
    The algorithm for replace is quite complicated:
8732
      - for all disks of the instance:
8733
        - create new LVs on the new node with same names
8734
        - shutdown the drbd device on the old secondary
8735
        - disconnect the drbd network on the primary
8736
        - create the drbd device on the new secondary
8737
        - network attach the drbd on the primary, using an artifice:
8738
          the drbd code for Attach() will connect to the network if it
8739
          finds a device which is connected to the good local disks but
8740
          not network enabled
8741
      - wait for sync across all devices
8742
      - remove all disks from the old secondary
8743

8744
    Failures are not very well handled.
8745

8746
    """
8747
    steps_total = 6
8748

    
8749
    # Step: check device activation
8750
    self.lu.LogStep(1, steps_total, "Check device existence")
8751
    self._CheckDisksExistence([self.instance.primary_node])
8752
    self._CheckVolumeGroup([self.instance.primary_node])
8753

    
8754
    # Step: check other node consistency
8755
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8756
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
8757

    
8758
    # Step: create new storage
8759
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8760
    for idx, dev in enumerate(self.instance.disks):
8761
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
8762
                      (self.new_node, idx))
8763
      # we pass force_create=True to force LVM creation
8764
      for new_lv in dev.children:
8765
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
8766
                        _GetInstanceInfoText(self.instance), False)
8767

    
8768
    # Step 4: dbrd minors and drbd setups changes
8769
    # after this, we must manually remove the drbd minors on both the
8770
    # error and the success paths
8771
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8772
    minors = self.cfg.AllocateDRBDMinor([self.new_node
8773
                                         for dev in self.instance.disks],
8774
                                        self.instance.name)
8775
    logging.debug("Allocated minors %r", minors)
8776

    
8777
    iv_names = {}
8778
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
8779
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
8780
                      (self.new_node, idx))
8781
      # create new devices on new_node; note that we create two IDs:
8782
      # one without port, so the drbd will be activated without
8783
      # networking information on the new node at this stage, and one
8784
      # with network, for the latter activation in step 4
8785
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
8786
      if self.instance.primary_node == o_node1:
8787
        p_minor = o_minor1
8788
      else:
8789
        assert self.instance.primary_node == o_node2, "Three-node instance?"
8790
        p_minor = o_minor2
8791

    
8792
      new_alone_id = (self.instance.primary_node, self.new_node, None,
8793
                      p_minor, new_minor, o_secret)
8794
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
8795
                    p_minor, new_minor, o_secret)
8796

    
8797
      iv_names[idx] = (dev, dev.children, new_net_id)
8798
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
8799
                    new_net_id)
8800
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
8801
                              logical_id=new_alone_id,
8802
                              children=dev.children,
8803
                              size=dev.size)
8804
      try:
8805
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
8806
                              _GetInstanceInfoText(self.instance), False)
8807
      except errors.GenericError:
8808
        self.cfg.ReleaseDRBDMinors(self.instance.name)
8809
        raise
8810

    
8811
    # We have new devices, shutdown the drbd on the old secondary
8812
    for idx, dev in enumerate(self.instance.disks):
8813
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
8814
      self.cfg.SetDiskID(dev, self.target_node)
8815
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
8816
      if msg:
8817
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
8818
                           "node: %s" % (idx, msg),
8819
                           hint=("Please cleanup this device manually as"
8820
                                 " soon as possible"))
8821

    
8822
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
8823
    result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
8824
                                               self.node_secondary_ip,
8825
                                               self.instance.disks)\
8826
                                              [self.instance.primary_node]
8827

    
8828
    msg = result.fail_msg
8829
    if msg:
8830
      # detaches didn't succeed (unlikely)
8831
      self.cfg.ReleaseDRBDMinors(self.instance.name)
8832
      raise errors.OpExecError("Can't detach the disks from the network on"
8833
                               " old node: %s" % (msg,))
8834

    
8835
    # if we managed to detach at least one, we update all the disks of
8836
    # the instance to point to the new secondary
8837
    self.lu.LogInfo("Updating instance configuration")
8838
    for dev, _, new_logical_id in iv_names.itervalues():
8839
      dev.logical_id = new_logical_id
8840
      self.cfg.SetDiskID(dev, self.instance.primary_node)
8841

    
8842
    self.cfg.Update(self.instance, feedback_fn)
8843

    
8844
    # and now perform the drbd attach
8845
    self.lu.LogInfo("Attaching primary drbds to new secondary"
8846
                    " (standalone => connected)")
8847
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
8848
                                            self.new_node],
8849
                                           self.node_secondary_ip,
8850
                                           self.instance.disks,
8851
                                           self.instance.name,
8852
                                           False)
8853
    for to_node, to_result in result.items():
8854
      msg = to_result.fail_msg
8855
      if msg:
8856
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
8857
                           to_node, msg,
8858
                           hint=("please do a gnt-instance info to see the"
8859
                                 " status of disks"))
8860
    cstep = 5
8861
    if self.early_release:
8862
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8863
      cstep += 1
8864
      self._RemoveOldStorage(self.target_node, iv_names)
8865
      # WARNING: we release all node locks here, do not do other RPCs
8866
      # than WaitForSync to the primary node
8867
      self._ReleaseNodeLock([self.instance.primary_node,
8868
                             self.target_node,
8869
                             self.new_node])
8870

    
8871
    # Wait for sync
8872
    # This can fail as the old devices are degraded and _WaitForSync
8873
    # does a combined result over all disks, so we don't check its return value
8874
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8875
    cstep += 1
8876
    _WaitForSync(self.lu, self.instance)
8877

    
8878
    # Check all devices manually
8879
    self._CheckDevices(self.instance.primary_node, iv_names)
8880

    
8881
    # Step: remove old storage
8882
    if not self.early_release:
8883
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8884
      self._RemoveOldStorage(self.target_node, iv_names)
8885

    
8886

    
8887
class LURepairNodeStorage(NoHooksLU):
8888
  """Repairs the volume group on a node.
8889

8890
  """
8891
  REQ_BGL = False
8892

    
8893
  def CheckArguments(self):
8894
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
8895

    
8896
    storage_type = self.op.storage_type
8897

    
8898
    if (constants.SO_FIX_CONSISTENCY not in
8899
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
8900
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
8901
                                 " repaired" % storage_type,
8902
                                 errors.ECODE_INVAL)
8903

    
8904
  def ExpandNames(self):
8905
    self.needed_locks = {
8906
      locking.LEVEL_NODE: [self.op.node_name],
8907
      }
8908

    
8909
  def _CheckFaultyDisks(self, instance, node_name):
8910
    """Ensure faulty disks abort the opcode or at least warn."""
8911
    try:
8912
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
8913
                                  node_name, True):
8914
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
8915
                                   " node '%s'" % (instance.name, node_name),
8916
                                   errors.ECODE_STATE)
8917
    except errors.OpPrereqError, err:
8918
      if self.op.ignore_consistency:
8919
        self.proc.LogWarning(str(err.args[0]))
8920
      else:
8921
        raise
8922

    
8923
  def CheckPrereq(self):
8924
    """Check prerequisites.
8925

8926
    """
8927
    # Check whether any instance on this node has faulty disks
8928
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
8929
      if not inst.admin_up:
8930
        continue
8931
      check_nodes = set(inst.all_nodes)
8932
      check_nodes.discard(self.op.node_name)
8933
      for inst_node_name in check_nodes:
8934
        self._CheckFaultyDisks(inst, inst_node_name)
8935

    
8936
  def Exec(self, feedback_fn):
8937
    feedback_fn("Repairing storage unit '%s' on %s ..." %
8938
                (self.op.name, self.op.node_name))
8939

    
8940
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
8941
    result = self.rpc.call_storage_execute(self.op.node_name,
8942
                                           self.op.storage_type, st_args,
8943
                                           self.op.name,
8944
                                           constants.SO_FIX_CONSISTENCY)
8945
    result.Raise("Failed to repair storage unit '%s' on %s" %
8946
                 (self.op.name, self.op.node_name))
8947

    
8948

    
8949
class LUNodeEvacStrategy(NoHooksLU):
8950
  """Computes the node evacuation strategy.
8951

8952
  """
8953
  REQ_BGL = False
8954

    
8955
  def CheckArguments(self):
8956
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
8957

    
8958
  def ExpandNames(self):
8959
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
8960
    self.needed_locks = locks = {}
8961
    if self.op.remote_node is None:
8962
      locks[locking.LEVEL_NODE] = locking.ALL_SET
8963
    else:
8964
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8965
      locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
8966

    
8967
  def Exec(self, feedback_fn):
8968
    if self.op.remote_node is not None:
8969
      instances = []
8970
      for node in self.op.nodes:
8971
        instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
8972
      result = []
8973
      for i in instances:
8974
        if i.primary_node == self.op.remote_node:
8975
          raise errors.OpPrereqError("Node %s is the primary node of"
8976
                                     " instance %s, cannot use it as"
8977
                                     " secondary" %
8978
                                     (self.op.remote_node, i.name),
8979
                                     errors.ECODE_INVAL)
8980
        result.append([i.name, self.op.remote_node])
8981
    else:
8982
      ial = IAllocator(self.cfg, self.rpc,
8983
                       mode=constants.IALLOCATOR_MODE_MEVAC,
8984
                       evac_nodes=self.op.nodes)
8985
      ial.Run(self.op.iallocator, validate=True)
8986
      if not ial.success:
8987
        raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
8988
                                 errors.ECODE_NORES)
8989
      result = ial.result
8990
    return result
8991

    
8992

    
8993
class LUInstanceGrowDisk(LogicalUnit):
8994
  """Grow a disk of an instance.
8995

8996
  """
8997
  HPATH = "disk-grow"
8998
  HTYPE = constants.HTYPE_INSTANCE
8999
  REQ_BGL = False
9000

    
9001
  def ExpandNames(self):
9002
    self._ExpandAndLockInstance()
9003
    self.needed_locks[locking.LEVEL_NODE] = []
9004
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9005

    
9006
  def DeclareLocks(self, level):
9007
    if level == locking.LEVEL_NODE:
9008
      self._LockInstancesNodes()
9009

    
9010
  def BuildHooksEnv(self):
9011
    """Build hooks env.
9012

9013
    This runs on the master, the primary and all the secondaries.
9014

9015
    """
9016
    env = {
9017
      "DISK": self.op.disk,
9018
      "AMOUNT": self.op.amount,
9019
      }
9020
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
9021
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
9022
    return env, nl, nl
9023

    
9024
  def CheckPrereq(self):
9025
    """Check prerequisites.
9026

9027
    This checks that the instance is in the cluster.
9028

9029
    """
9030
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9031
    assert instance is not None, \
9032
      "Cannot retrieve locked instance %s" % self.op.instance_name
9033
    nodenames = list(instance.all_nodes)
9034
    for node in nodenames:
9035
      _CheckNodeOnline(self, node)
9036

    
9037
    self.instance = instance
9038

    
9039
    if instance.disk_template not in constants.DTS_GROWABLE:
9040
      raise errors.OpPrereqError("Instance's disk layout does not support"
9041
                                 " growing.", errors.ECODE_INVAL)
9042

    
9043
    self.disk = instance.FindDisk(self.op.disk)
9044

    
9045
    if instance.disk_template not in (constants.DT_FILE,
9046
                                      constants.DT_SHARED_FILE):
9047
      # TODO: check the free disk space for file, when that feature will be
9048
      # supported
9049
      _CheckNodesFreeDiskPerVG(self, nodenames,
9050
                               self.disk.ComputeGrowth(self.op.amount))
9051

    
9052
  def Exec(self, feedback_fn):
9053
    """Execute disk grow.
9054

9055
    """
9056
    instance = self.instance
9057
    disk = self.disk
9058

    
9059
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
9060
    if not disks_ok:
9061
      raise errors.OpExecError("Cannot activate block device to grow")
9062

    
9063
    for node in instance.all_nodes:
9064
      self.cfg.SetDiskID(disk, node)
9065
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
9066
      result.Raise("Grow request failed to node %s" % node)
9067

    
9068
      # TODO: Rewrite code to work properly
9069
      # DRBD goes into sync mode for a short amount of time after executing the
9070
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
9071
      # calling "resize" in sync mode fails. Sleeping for a short amount of
9072
      # time is a work-around.
9073
      time.sleep(5)
9074

    
9075
    disk.RecordGrow(self.op.amount)
9076
    self.cfg.Update(instance, feedback_fn)
9077
    if self.op.wait_for_sync:
9078
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
9079
      if disk_abort:
9080
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
9081
                             " status.\nPlease check the instance.")
9082
      if not instance.admin_up:
9083
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
9084
    elif not instance.admin_up:
9085
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
9086
                           " not supposed to be running because no wait for"
9087
                           " sync mode was requested.")
9088

    
9089

    
9090
class LUInstanceQueryData(NoHooksLU):
9091
  """Query runtime instance data.
9092

9093
  """
9094
  REQ_BGL = False
9095

    
9096
  def ExpandNames(self):
9097
    self.needed_locks = {}
9098
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
9099

    
9100
    if self.op.instances:
9101
      self.wanted_names = []
9102
      for name in self.op.instances:
9103
        full_name = _ExpandInstanceName(self.cfg, name)
9104
        self.wanted_names.append(full_name)
9105
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
9106
    else:
9107
      self.wanted_names = None
9108
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
9109

    
9110
    self.needed_locks[locking.LEVEL_NODE] = []
9111
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9112

    
9113
  def DeclareLocks(self, level):
9114
    if level == locking.LEVEL_NODE:
9115
      self._LockInstancesNodes()
9116

    
9117
  def CheckPrereq(self):
9118
    """Check prerequisites.
9119

9120
    This only checks the optional instance list against the existing names.
9121

9122
    """
9123
    if self.wanted_names is None:
9124
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
9125

    
9126
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
9127
                             in self.wanted_names]
9128

    
9129
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
9130
    """Returns the status of a block device
9131

9132
    """
9133
    if self.op.static or not node:
9134
      return None
9135

    
9136
    self.cfg.SetDiskID(dev, node)
9137

    
9138
    result = self.rpc.call_blockdev_find(node, dev)
9139
    if result.offline:
9140
      return None
9141

    
9142
    result.Raise("Can't compute disk status for %s" % instance_name)
9143

    
9144
    status = result.payload
9145
    if status is None:
9146
      return None
9147

    
9148
    return (status.dev_path, status.major, status.minor,
9149
            status.sync_percent, status.estimated_time,
9150
            status.is_degraded, status.ldisk_status)
9151

    
9152
  def _ComputeDiskStatus(self, instance, snode, dev):
9153
    """Compute block device status.
9154

9155
    """
9156
    if dev.dev_type in constants.LDS_DRBD:
9157
      # we change the snode then (otherwise we use the one passed in)
9158
      if dev.logical_id[0] == instance.primary_node:
9159
        snode = dev.logical_id[1]
9160
      else:
9161
        snode = dev.logical_id[0]
9162

    
9163
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
9164
                                              instance.name, dev)
9165
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
9166

    
9167
    if dev.children:
9168
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
9169
                      for child in dev.children]
9170
    else:
9171
      dev_children = []
9172

    
9173
    data = {
9174
      "iv_name": dev.iv_name,
9175
      "dev_type": dev.dev_type,
9176
      "logical_id": dev.logical_id,
9177
      "physical_id": dev.physical_id,
9178
      "pstatus": dev_pstatus,
9179
      "sstatus": dev_sstatus,
9180
      "children": dev_children,
9181
      "mode": dev.mode,
9182
      "size": dev.size,
9183
      }
9184

    
9185
    return data
9186

    
9187
  def Exec(self, feedback_fn):
9188
    """Gather and return data"""
9189
    result = {}
9190

    
9191
    cluster = self.cfg.GetClusterInfo()
9192

    
9193
    for instance in self.wanted_instances:
9194
      if not self.op.static:
9195
        remote_info = self.rpc.call_instance_info(instance.primary_node,
9196
                                                  instance.name,
9197
                                                  instance.hypervisor)
9198
        remote_info.Raise("Error checking node %s" % instance.primary_node)
9199
        remote_info = remote_info.payload
9200
        if remote_info and "state" in remote_info:
9201
          remote_state = "up"
9202
        else:
9203
          remote_state = "down"
9204
      else:
9205
        remote_state = None
9206
      if instance.admin_up:
9207
        config_state = "up"
9208
      else:
9209
        config_state = "down"
9210

    
9211
      disks = [self._ComputeDiskStatus(instance, None, device)
9212
               for device in instance.disks]
9213

    
9214
      idict = {
9215
        "name": instance.name,
9216
        "config_state": config_state,
9217
        "run_state": remote_state,
9218
        "pnode": instance.primary_node,
9219
        "snodes": instance.secondary_nodes,
9220
        "os": instance.os,
9221
        # this happens to be the same format used for hooks
9222
        "nics": _NICListToTuple(self, instance.nics),
9223
        "disk_template": instance.disk_template,
9224
        "disks": disks,
9225
        "hypervisor": instance.hypervisor,
9226
        "network_port": instance.network_port,
9227
        "hv_instance": instance.hvparams,
9228
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
9229
        "be_instance": instance.beparams,
9230
        "be_actual": cluster.FillBE(instance),
9231
        "os_instance": instance.osparams,
9232
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
9233
        "serial_no": instance.serial_no,
9234
        "mtime": instance.mtime,
9235
        "ctime": instance.ctime,
9236
        "uuid": instance.uuid,
9237
        }
9238

    
9239
      result[instance.name] = idict
9240

    
9241
    return result
9242

    
9243

    
9244
class LUInstanceSetParams(LogicalUnit):
9245
  """Modifies an instances's parameters.
9246

9247
  """
9248
  HPATH = "instance-modify"
9249
  HTYPE = constants.HTYPE_INSTANCE
9250
  REQ_BGL = False
9251

    
9252
  def CheckArguments(self):
9253
    if not (self.op.nics or self.op.disks or self.op.disk_template or
9254
            self.op.hvparams or self.op.beparams or self.op.os_name):
9255
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
9256

    
9257
    if self.op.hvparams:
9258
      _CheckGlobalHvParams(self.op.hvparams)
9259

    
9260
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9261

    
9262
    # Disk validation
9263
    disk_addremove = 0
9264
    for disk_op, disk_dict in self.op.disks:
9265
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
9266
      if disk_op == constants.DDM_REMOVE:
9267
        disk_addremove += 1
9268
        continue
9269
      elif disk_op == constants.DDM_ADD:
9270
        disk_addremove += 1
9271
      else:
9272
        if not isinstance(disk_op, int):
9273
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
9274
        if not isinstance(disk_dict, dict):
9275
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
9276
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9277

    
9278
      if disk_op == constants.DDM_ADD:
9279
        if "adopt" in disk_dict:
9280
          has_adopt = True
9281
        else:
9282
          has_adopt = False
9283

    
9284
        if has_adopt:
9285
          if instance.disk_template not in constants.DTS_MAY_ADOPT:
9286
            raise errors.OpPrereqError("Disk adoption is not supported for the"
9287
                                       " '%s' disk template" %
9288
                                       instance.disk_template,
9289
                                       errors.ECODE_INVAL)
9290
        self.adopt_disks = has_adopt
9291

    
9292
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
9293
        if mode not in constants.DISK_ACCESS_SET:
9294
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
9295
                                     errors.ECODE_INVAL)
9296
        if not has_adopt:
9297
          size = disk_dict.get('size', None)
9298
          if size is None:
9299
            raise errors.OpPrereqError("Required disk parameter size missing",
9300
                                       errors.ECODE_INVAL)
9301
          try:
9302
            size = int(size)
9303
          except (TypeError, ValueError), err:
9304
            raise errors.OpPrereqError("Invalid disk size parameter: %s" %
9305
                                       str(err), errors.ECODE_INVAL)
9306
          disk_dict['size'] = size
9307
      else:
9308
        # modification of disk
9309
        if 'size' in disk_dict:
9310
          raise errors.OpPrereqError("Disk size change not possible, use"
9311
                                     " grow-disk", errors.ECODE_INVAL)
9312

    
9313
    if disk_addremove > 1:
9314
      raise errors.OpPrereqError("Only one disk add or remove operation"
9315
                                 " supported at a time", errors.ECODE_INVAL)
9316

    
9317
    if self.op.disks and self.op.disk_template is not None:
9318
      raise errors.OpPrereqError("Disk template conversion and other disk"
9319
                                 " changes not supported at the same time",
9320
                                 errors.ECODE_INVAL)
9321

    
9322
    if (self.op.disk_template and
9323
        self.op.disk_template in constants.DTS_INT_MIRROR and
9324
        self.op.remote_node is None):
9325
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
9326
                                 " one requires specifying a secondary node",
9327
                                 errors.ECODE_INVAL)
9328

    
9329
    # NIC validation
9330
    nic_addremove = 0
9331
    for nic_op, nic_dict in self.op.nics:
9332
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
9333
      if nic_op == constants.DDM_REMOVE:
9334
        nic_addremove += 1
9335
        continue
9336
      elif nic_op == constants.DDM_ADD:
9337
        nic_addremove += 1
9338
      else:
9339
        if not isinstance(nic_op, int):
9340
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
9341
        if not isinstance(nic_dict, dict):
9342
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
9343
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9344

    
9345
      # nic_dict should be a dict
9346
      nic_ip = nic_dict.get('ip', None)
9347
      if nic_ip is not None:
9348
        if nic_ip.lower() == constants.VALUE_NONE:
9349
          nic_dict['ip'] = None
9350
        else:
9351
          if not netutils.IPAddress.IsValid(nic_ip):
9352
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
9353
                                       errors.ECODE_INVAL)
9354

    
9355
      nic_bridge = nic_dict.get('bridge', None)
9356
      nic_link = nic_dict.get('link', None)
9357
      if nic_bridge and nic_link:
9358
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
9359
                                   " at the same time", errors.ECODE_INVAL)
9360
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
9361
        nic_dict['bridge'] = None
9362
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
9363
        nic_dict['link'] = None
9364

    
9365
      if nic_op == constants.DDM_ADD:
9366
        nic_mac = nic_dict.get('mac', None)
9367
        if nic_mac is None:
9368
          nic_dict['mac'] = constants.VALUE_AUTO
9369

    
9370
      if 'mac' in nic_dict:
9371
        nic_mac = nic_dict['mac']
9372
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9373
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
9374

    
9375
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
9376
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
9377
                                     " modifying an existing nic",
9378
                                     errors.ECODE_INVAL)
9379

    
9380
    if nic_addremove > 1:
9381
      raise errors.OpPrereqError("Only one NIC add or remove operation"
9382
                                 " supported at a time", errors.ECODE_INVAL)
9383

    
9384
  def ExpandNames(self):
9385
    self._ExpandAndLockInstance()
9386
    self.needed_locks[locking.LEVEL_NODE] = []
9387
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9388

    
9389
  def DeclareLocks(self, level):
9390
    if level == locking.LEVEL_NODE:
9391
      self._LockInstancesNodes()
9392
      if self.op.disk_template and self.op.remote_node:
9393
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9394
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
9395

    
9396
  def BuildHooksEnv(self):
9397
    """Build hooks env.
9398

9399
    This runs on the master, primary and secondaries.
9400

9401
    """
9402
    args = dict()
9403
    if constants.BE_MEMORY in self.be_new:
9404
      args['memory'] = self.be_new[constants.BE_MEMORY]
9405
    if constants.BE_VCPUS in self.be_new:
9406
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
9407
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
9408
    # information at all.
9409
    if self.op.nics:
9410
      args['nics'] = []
9411
      nic_override = dict(self.op.nics)
9412
      for idx, nic in enumerate(self.instance.nics):
9413
        if idx in nic_override:
9414
          this_nic_override = nic_override[idx]
9415
        else:
9416
          this_nic_override = {}
9417
        if 'ip' in this_nic_override:
9418
          ip = this_nic_override['ip']
9419
        else:
9420
          ip = nic.ip
9421
        if 'mac' in this_nic_override:
9422
          mac = this_nic_override['mac']
9423
        else:
9424
          mac = nic.mac
9425
        if idx in self.nic_pnew:
9426
          nicparams = self.nic_pnew[idx]
9427
        else:
9428
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
9429
        mode = nicparams[constants.NIC_MODE]
9430
        link = nicparams[constants.NIC_LINK]
9431
        args['nics'].append((ip, mac, mode, link))
9432
      if constants.DDM_ADD in nic_override:
9433
        ip = nic_override[constants.DDM_ADD].get('ip', None)
9434
        mac = nic_override[constants.DDM_ADD]['mac']
9435
        nicparams = self.nic_pnew[constants.DDM_ADD]
9436
        mode = nicparams[constants.NIC_MODE]
9437
        link = nicparams[constants.NIC_LINK]
9438
        args['nics'].append((ip, mac, mode, link))
9439
      elif constants.DDM_REMOVE in nic_override:
9440
        del args['nics'][-1]
9441

    
9442
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
9443
    if self.op.disk_template:
9444
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
9445
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
9446
    return env, nl, nl
9447

    
9448
  def CheckPrereq(self):
9449
    """Check prerequisites.
9450

9451
    This only checks the instance list against the existing names.
9452

9453
    """
9454
    # checking the new params on the primary/secondary nodes
9455

    
9456
    instance = self.instance
9457
    cluster = self.cluster = self.cfg.GetClusterInfo()
9458
    assert self.instance is not None, \
9459
      "Cannot retrieve locked instance %s" % self.op.instance_name
9460
    pnode = instance.primary_node
9461
    nodelist = list(instance.all_nodes)
9462

    
9463
    # OS change
9464
    if self.op.os_name and not self.op.force:
9465
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
9466
                      self.op.force_variant)
9467
      instance_os = self.op.os_name
9468
    else:
9469
      instance_os = instance.os
9470

    
9471
    if self.op.disk_template:
9472
      if instance.disk_template == self.op.disk_template:
9473
        raise errors.OpPrereqError("Instance already has disk template %s" %
9474
                                   instance.disk_template, errors.ECODE_INVAL)
9475

    
9476
      if (instance.disk_template,
9477
          self.op.disk_template) not in self._DISK_CONVERSIONS:
9478
        raise errors.OpPrereqError("Unsupported disk template conversion from"
9479
                                   " %s to %s" % (instance.disk_template,
9480
                                                  self.op.disk_template),
9481
                                   errors.ECODE_INVAL)
9482
      _CheckInstanceDown(self, instance, "cannot change disk template")
9483
      if self.op.disk_template in constants.DTS_INT_MIRROR:
9484
        if self.op.remote_node == pnode:
9485
          raise errors.OpPrereqError("Given new secondary node %s is the same"
9486
                                     " as the primary node of the instance" %
9487
                                     self.op.remote_node, errors.ECODE_STATE)
9488
        _CheckNodeOnline(self, self.op.remote_node)
9489
        _CheckNodeNotDrained(self, self.op.remote_node)
9490
        # FIXME: here we assume that the old instance type is DT_PLAIN
9491
        assert instance.disk_template == constants.DT_PLAIN
9492
        disks = [{"size": d.size, "vg": d.logical_id[0]}
9493
                 for d in instance.disks]
9494
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
9495
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
9496

    
9497
    # hvparams processing
9498
    if self.op.hvparams:
9499
      hv_type = instance.hypervisor
9500
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
9501
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
9502
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
9503

    
9504
      # local check
9505
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
9506
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
9507
      self.hv_new = hv_new # the new actual values
9508
      self.hv_inst = i_hvdict # the new dict (without defaults)
9509
    else:
9510
      self.hv_new = self.hv_inst = {}
9511

    
9512
    # beparams processing
9513
    if self.op.beparams:
9514
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
9515
                                   use_none=True)
9516
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
9517
      be_new = cluster.SimpleFillBE(i_bedict)
9518
      self.be_new = be_new # the new actual values
9519
      self.be_inst = i_bedict # the new dict (without defaults)
9520
    else:
9521
      self.be_new = self.be_inst = {}
9522

    
9523
    # osparams processing
9524
    if self.op.osparams:
9525
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
9526
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
9527
      self.os_inst = i_osdict # the new dict (without defaults)
9528
    else:
9529
      self.os_inst = {}
9530

    
9531
    self.warn = []
9532

    
9533
    if constants.BE_MEMORY in self.op.beparams and not self.op.force:
9534
      mem_check_list = [pnode]
9535
      if be_new[constants.BE_AUTO_BALANCE]:
9536
        # either we changed auto_balance to yes or it was from before
9537
        mem_check_list.extend(instance.secondary_nodes)
9538
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
9539
                                                  instance.hypervisor)
9540
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
9541
                                         instance.hypervisor)
9542
      pninfo = nodeinfo[pnode]
9543
      msg = pninfo.fail_msg
9544
      if msg:
9545
        # Assume the primary node is unreachable and go ahead
9546
        self.warn.append("Can't get info from primary node %s: %s" %
9547
                         (pnode,  msg))
9548
      elif not isinstance(pninfo.payload.get('memory_free', None), int):
9549
        self.warn.append("Node data from primary node %s doesn't contain"
9550
                         " free memory information" % pnode)
9551
      elif instance_info.fail_msg:
9552
        self.warn.append("Can't get instance runtime information: %s" %
9553
                        instance_info.fail_msg)
9554
      else:
9555
        if instance_info.payload:
9556
          current_mem = int(instance_info.payload['memory'])
9557
        else:
9558
          # Assume instance not running
9559
          # (there is a slight race condition here, but it's not very probable,
9560
          # and we have no other way to check)
9561
          current_mem = 0
9562
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
9563
                    pninfo.payload['memory_free'])
9564
        if miss_mem > 0:
9565
          raise errors.OpPrereqError("This change will prevent the instance"
9566
                                     " from starting, due to %d MB of memory"
9567
                                     " missing on its primary node" % miss_mem,
9568
                                     errors.ECODE_NORES)
9569

    
9570
      if be_new[constants.BE_AUTO_BALANCE]:
9571
        for node, nres in nodeinfo.items():
9572
          if node not in instance.secondary_nodes:
9573
            continue
9574
          msg = nres.fail_msg
9575
          if msg:
9576
            self.warn.append("Can't get info from secondary node %s: %s" %
9577
                             (node, msg))
9578
          elif not isinstance(nres.payload.get('memory_free', None), int):
9579
            self.warn.append("Secondary node %s didn't return free"
9580
                             " memory information" % node)
9581
          elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
9582
            self.warn.append("Not enough memory to failover instance to"
9583
                             " secondary node %s" % node)
9584

    
9585
    # NIC processing
9586
    self.nic_pnew = {}
9587
    self.nic_pinst = {}
9588
    for nic_op, nic_dict in self.op.nics:
9589
      if nic_op == constants.DDM_REMOVE:
9590
        if not instance.nics:
9591
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
9592
                                     errors.ECODE_INVAL)
9593
        continue
9594
      if nic_op != constants.DDM_ADD:
9595
        # an existing nic
9596
        if not instance.nics:
9597
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
9598
                                     " no NICs" % nic_op,
9599
                                     errors.ECODE_INVAL)
9600
        if nic_op < 0 or nic_op >= len(instance.nics):
9601
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
9602
                                     " are 0 to %d" %
9603
                                     (nic_op, len(instance.nics) - 1),
9604
                                     errors.ECODE_INVAL)
9605
        old_nic_params = instance.nics[nic_op].nicparams
9606
        old_nic_ip = instance.nics[nic_op].ip
9607
      else:
9608
        old_nic_params = {}
9609
        old_nic_ip = None
9610

    
9611
      update_params_dict = dict([(key, nic_dict[key])
9612
                                 for key in constants.NICS_PARAMETERS
9613
                                 if key in nic_dict])
9614

    
9615
      if 'bridge' in nic_dict:
9616
        update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
9617

    
9618
      new_nic_params = _GetUpdatedParams(old_nic_params,
9619
                                         update_params_dict)
9620
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
9621
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
9622
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
9623
      self.nic_pinst[nic_op] = new_nic_params
9624
      self.nic_pnew[nic_op] = new_filled_nic_params
9625
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
9626

    
9627
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
9628
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
9629
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
9630
        if msg:
9631
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
9632
          if self.op.force:
9633
            self.warn.append(msg)
9634
          else:
9635
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
9636
      if new_nic_mode == constants.NIC_MODE_ROUTED:
9637
        if 'ip' in nic_dict:
9638
          nic_ip = nic_dict['ip']
9639
        else:
9640
          nic_ip = old_nic_ip
9641
        if nic_ip is None:
9642
          raise errors.OpPrereqError('Cannot set the nic ip to None'
9643
                                     ' on a routed nic', errors.ECODE_INVAL)
9644
      if 'mac' in nic_dict:
9645
        nic_mac = nic_dict['mac']
9646
        if nic_mac is None:
9647
          raise errors.OpPrereqError('Cannot set the nic mac to None',
9648
                                     errors.ECODE_INVAL)
9649
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9650
          # otherwise generate the mac
9651
          nic_dict['mac'] = self.cfg.GenerateMAC(self.proc.GetECId())
9652
        else:
9653
          # or validate/reserve the current one
9654
          try:
9655
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
9656
          except errors.ReservationError:
9657
            raise errors.OpPrereqError("MAC address %s already in use"
9658
                                       " in cluster" % nic_mac,
9659
                                       errors.ECODE_NOTUNIQUE)
9660

    
9661
    # DISK processing
9662
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
9663
      raise errors.OpPrereqError("Disk operations not supported for"
9664
                                 " diskless instances",
9665
                                 errors.ECODE_INVAL)
9666
    for disk_op, disk_dict in self.op.disks:
9667
      if disk_op == constants.DDM_REMOVE:
9668
        if len(instance.disks) == 1:
9669
          raise errors.OpPrereqError("Cannot remove the last disk of"
9670
                                     " an instance", errors.ECODE_INVAL)
9671
        _CheckInstanceDown(self, instance, "cannot remove disks")
9672

    
9673
      if (disk_op == constants.DDM_ADD and
9674
          len(instance.disks) >= constants.MAX_DISKS):
9675
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
9676
                                   " add more" % constants.MAX_DISKS,
9677
                                   errors.ECODE_STATE)
9678
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
9679
        # an existing disk
9680
        if disk_op < 0 or disk_op >= len(instance.disks):
9681
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
9682
                                     " are 0 to %d" %
9683
                                     (disk_op, len(instance.disks)),
9684
                                     errors.ECODE_INVAL)
9685

    
9686
      if disk_op == constants.DDM_ADD and self.adopt_disks:
9687
        if instance.disk_template == constants.DT_PLAIN:
9688
          # Check the adoption data
9689
          vg = disk_dict.get("vg", self.cfg.GetVGName())
9690
          lv_name = vg + "/" + disk_dict["adopt"]
9691
          try:
9692
            # FIXME: lv_name here is "vg/lv" need to ensure that other calls
9693
            # to ReserveLV uses the same syntax
9694
            self.cfg.ReserveLV(lv_name, self.proc.GetECId())
9695
          except errors.ReservationError:
9696
            raise errors.OpPrereqError("LV named %s used by another instance" %
9697
                                       lv_name, errors.ECODE_NOTUNIQUE)
9698

    
9699
          vg_names = self.rpc.call_vg_list([instance.primary_node])
9700
          vg_names = vg_names[instance.primary_node]
9701
          vg_names.Raise("Cannot get VG information from node %s" %
9702
                         instance.primary_node)
9703

    
9704
          node_lvs = self.rpc.call_lv_list([instance.primary_node],
9705
                                           vg_names.payload.keys()
9706
                                          )[instance.primary_node]
9707
          node_lvs.Raise("Cannot get LV information from node %s" %
9708
                         instance.primary_node)
9709
          node_lvs = node_lvs.payload
9710

    
9711
          if lv_name not in node_lvs:
9712
            raise errors.OpPrereqError("Logical volume: %s not present "
9713
                                       " on node %s" %
9714
                                       (lv_name, instance.primary_node),
9715
                                       errors.ECODE_INVAL)
9716
          if node_lvs[lv_name][2]:
9717
            raise errors.OpPrereqError("Logical volume %s is online, cannot"
9718
                                       " adopt." % lv_name,
9719
                                       errors.ECODE_STATE)
9720
          # update the size of disk based on what is found
9721
          disk_dict["size"] = int(float(node_lvs[lv_name][0]))
9722

    
9723
        elif instance.disk_template == constants.DT_BLOCK:
9724
          disk = disk_dict["adopt"]
9725
          node_disks = self.rpc.call_bdev_sizes([instance.primary_node],
9726
                                                [disk])[instance.primary_node]
9727
          node_disks = node_disks.payload
9728
          if disk not in node_disks:
9729
            raise errors.OpPrereqError("Missing block device %s" % disk,
9730
                                       errors.ECODE_INVAL)
9731
          disk_dict["size"] = int(float(node_disks[disk]))
9732

    
9733
    return
9734

    
9735
  def _ConvertPlainToDrbd(self, feedback_fn):
9736
    """Converts an instance from plain to drbd.
9737

9738
    """
9739
    feedback_fn("Converting template to drbd")
9740
    instance = self.instance
9741
    pnode = instance.primary_node
9742
    snode = self.op.remote_node
9743

    
9744
    # create a fake disk info for _GenerateDiskTemplate
9745
    disk_info = [{"size": d.size, "mode": d.mode} for d in instance.disks]
9746
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
9747
                                      instance.name, pnode, [snode],
9748
                                      disk_info, None, None, 0, feedback_fn)
9749
    info = _GetInstanceInfoText(instance)
9750
    feedback_fn("Creating aditional volumes...")
9751
    # first, create the missing data and meta devices
9752
    for disk in new_disks:
9753
      # unfortunately this is... not too nice
9754
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
9755
                            info, True)
9756
      for child in disk.children:
9757
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
9758
    # at this stage, all new LVs have been created, we can rename the
9759
    # old ones
9760
    feedback_fn("Renaming original volumes...")
9761
    rename_list = [(o, n.children[0].logical_id)
9762
                   for (o, n) in zip(instance.disks, new_disks)]
9763
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
9764
    result.Raise("Failed to rename original LVs")
9765

    
9766
    feedback_fn("Initializing DRBD devices...")
9767
    # all child devices are in place, we can now create the DRBD devices
9768
    for disk in new_disks:
9769
      for node in [pnode, snode]:
9770
        f_create = node == pnode
9771
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
9772

    
9773
    # at this point, the instance has been modified
9774
    instance.disk_template = constants.DT_DRBD8
9775
    instance.disks = new_disks
9776
    self.cfg.Update(instance, feedback_fn)
9777

    
9778
    # disks are created, waiting for sync
9779
    disk_abort = not _WaitForSync(self, instance)
9780
    if disk_abort:
9781
      raise errors.OpExecError("There are some degraded disks for"
9782
                               " this instance, please cleanup manually")
9783

    
9784
  def _ConvertDrbdToPlain(self, feedback_fn):
9785
    """Converts an instance from drbd to plain.
9786

9787
    """
9788
    instance = self.instance
9789
    assert len(instance.secondary_nodes) == 1
9790
    pnode = instance.primary_node
9791
    snode = instance.secondary_nodes[0]
9792
    feedback_fn("Converting template to plain")
9793

    
9794
    old_disks = instance.disks
9795
    new_disks = [d.children[0] for d in old_disks]
9796

    
9797
    # copy over size and mode
9798
    for parent, child in zip(old_disks, new_disks):
9799
      child.size = parent.size
9800
      child.mode = parent.mode
9801

    
9802
    # update instance structure
9803
    instance.disks = new_disks
9804
    instance.disk_template = constants.DT_PLAIN
9805
    self.cfg.Update(instance, feedback_fn)
9806

    
9807
    feedback_fn("Removing volumes on the secondary node...")
9808
    for disk in old_disks:
9809
      self.cfg.SetDiskID(disk, snode)
9810
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
9811
      if msg:
9812
        self.LogWarning("Could not remove block device %s on node %s,"
9813
                        " continuing anyway: %s", disk.iv_name, snode, msg)
9814

    
9815
    feedback_fn("Removing unneeded volumes on the primary node...")
9816
    for idx, disk in enumerate(old_disks):
9817
      meta = disk.children[1]
9818
      self.cfg.SetDiskID(meta, pnode)
9819
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
9820
      if msg:
9821
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
9822
                        " continuing anyway: %s", idx, pnode, msg)
9823

    
9824
  def Exec(self, feedback_fn):
9825
    """Modifies an instance.
9826

9827
    All parameters take effect only at the next restart of the instance.
9828

9829
    """
9830
    # Process here the warnings from CheckPrereq, as we don't have a
9831
    # feedback_fn there.
9832
    for warn in self.warn:
9833
      feedback_fn("WARNING: %s" % warn)
9834

    
9835
    result = []
9836
    instance = self.instance
9837
    # disk changes
9838
    for disk_op, disk_dict in self.op.disks:
9839
      if disk_op == constants.DDM_REMOVE:
9840
        # remove the last disk
9841
        device = instance.disks.pop()
9842
        device_idx = len(instance.disks)
9843
        for node, disk in device.ComputeNodeTree(instance.primary_node):
9844
          self.cfg.SetDiskID(disk, node)
9845
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
9846
          if msg:
9847
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
9848
                            " continuing anyway", device_idx, node, msg)
9849
        result.append(("disk/%d" % device_idx, "remove"))
9850
      elif disk_op == constants.DDM_ADD:
9851
        # add a new disk
9852
        if instance.disk_template in (constants.DT_FILE,
9853
                                        constants.DT_SHARED_FILE):
9854
          file_driver, file_path = instance.disks[0].logical_id
9855
          file_path = os.path.dirname(file_path)
9856
        else:
9857
          file_driver = file_path = None
9858
        disk_idx_base = len(instance.disks)
9859
        new_disk = _GenerateDiskTemplate(self,
9860
                                         instance.disk_template,
9861
                                         instance.name, instance.primary_node,
9862
                                         instance.secondary_nodes,
9863
                                         [disk_dict],
9864
                                         file_path,
9865
                                         file_driver,
9866
                                         disk_idx_base, feedback_fn)[0]
9867
        instance.disks.append(new_disk)
9868
        info = _GetInstanceInfoText(instance)
9869

    
9870
        if self.adopt_disks:
9871
          if instance.disk_template == constants.DT_PLAIN:
9872
            # rename LV to the a newly-generated name; we need to construct
9873
            # 'fake' LV disk with the old data, plus the new unique_id
9874
            tmp_disk = objects.Disk.FromDict(disk_dict)
9875
            rename_to = tmp_disk.logical_id
9876
            tmp_disk.logical_id = (tmp_disk.logical_id[0], disk_dict["adopt"])
9877
            self.cfg.SetDiskID(tmp_disk, instance.primary_node)
9878
            result = self.rpc.call_blockdev_rename(instance.primary_node,
9879
                                                   [(tmp_disk, rename_to)])
9880
            result.Raise("Failed to rename adopted LV")
9881
          result.append(("disk/%d" % disk_idx_base, "add:adopt=%s,mode=%s" %
9882
                         (disk_dict["adopt"], new_disk.mode)))
9883
        else:
9884
          logging.info("Creating volume %s for instance %s",
9885
                       new_disk.iv_name, instance.name)
9886
          # Note: this needs to be kept in sync with _CreateDisks
9887
          #HARDCODE
9888
          for node in instance.all_nodes:
9889
            f_create = node == instance.primary_node
9890
            try:
9891
              _CreateBlockDev(self, node, instance, new_disk,
9892
                              f_create, info, f_create)
9893
            except errors.OpExecError, err:
9894
              self.LogWarning("Failed to create volume %s (%s) on"
9895
                              " node %s: %s",
9896
                              new_disk.iv_name, new_disk, node, err)
9897
          result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
9898
                         (new_disk.size, new_disk.mode)))
9899
      else:
9900
        # change a given disk
9901
        instance.disks[disk_op].mode = disk_dict['mode']
9902
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
9903

    
9904
    if self.op.disk_template:
9905
      r_shut = _ShutdownInstanceDisks(self, instance)
9906
      if not r_shut:
9907
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
9908
                                 " proceed with disk template conversion")
9909
      mode = (instance.disk_template, self.op.disk_template)
9910
      try:
9911
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
9912
      except:
9913
        self.cfg.ReleaseDRBDMinors(instance.name)
9914
        raise
9915
      result.append(("disk_template", self.op.disk_template))
9916

    
9917
    # NIC changes
9918
    for nic_op, nic_dict in self.op.nics:
9919
      if nic_op == constants.DDM_REMOVE:
9920
        # remove the last nic
9921
        del instance.nics[-1]
9922
        result.append(("nic.%d" % len(instance.nics), "remove"))
9923
      elif nic_op == constants.DDM_ADD:
9924
        # mac and bridge should be set, by now
9925
        mac = nic_dict['mac']
9926
        ip = nic_dict.get('ip', None)
9927
        nicparams = self.nic_pinst[constants.DDM_ADD]
9928
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
9929
        instance.nics.append(new_nic)
9930
        result.append(("nic.%d" % (len(instance.nics) - 1),
9931
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
9932
                       (new_nic.mac, new_nic.ip,
9933
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
9934
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
9935
                       )))
9936
      else:
9937
        for key in 'mac', 'ip':
9938
          if key in nic_dict:
9939
            setattr(instance.nics[nic_op], key, nic_dict[key])
9940
        if nic_op in self.nic_pinst:
9941
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
9942
        for key, val in nic_dict.iteritems():
9943
          result.append(("nic.%s/%d" % (key, nic_op), val))
9944

    
9945
    # hvparams changes
9946
    if self.op.hvparams:
9947
      instance.hvparams = self.hv_inst
9948
      for key, val in self.op.hvparams.iteritems():
9949
        result.append(("hv/%s" % key, val))
9950

    
9951
    # beparams changes
9952
    if self.op.beparams:
9953
      instance.beparams = self.be_inst
9954
      for key, val in self.op.beparams.iteritems():
9955
        result.append(("be/%s" % key, val))
9956

    
9957
    # OS change
9958
    if self.op.os_name:
9959
      instance.os = self.op.os_name
9960

    
9961
    # osparams changes
9962
    if self.op.osparams:
9963
      instance.osparams = self.os_inst
9964
      for key, val in self.op.osparams.iteritems():
9965
        result.append(("os/%s" % key, val))
9966

    
9967
    self.cfg.Update(instance, feedback_fn)
9968

    
9969
    return result
9970

    
9971
  _DISK_CONVERSIONS = {
9972
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
9973
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
9974
    }
9975

    
9976

    
9977
class LUBackupQuery(NoHooksLU):
9978
  """Query the exports list
9979

9980
  """
9981
  REQ_BGL = False
9982

    
9983
  def ExpandNames(self):
9984
    self.needed_locks = {}
9985
    self.share_locks[locking.LEVEL_NODE] = 1
9986
    if not self.op.nodes:
9987
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9988
    else:
9989
      self.needed_locks[locking.LEVEL_NODE] = \
9990
        _GetWantedNodes(self, self.op.nodes)
9991

    
9992
  def Exec(self, feedback_fn):
9993
    """Compute the list of all the exported system images.
9994

9995
    @rtype: dict
9996
    @return: a dictionary with the structure node->(export-list)
9997
        where export-list is a list of the instances exported on
9998
        that node.
9999

10000
    """
10001
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
10002
    rpcresult = self.rpc.call_export_list(self.nodes)
10003
    result = {}
10004
    for node in rpcresult:
10005
      if rpcresult[node].fail_msg:
10006
        result[node] = False
10007
      else:
10008
        result[node] = rpcresult[node].payload
10009

    
10010
    return result
10011

    
10012

    
10013
class LUBackupPrepare(NoHooksLU):
10014
  """Prepares an instance for an export and returns useful information.
10015

10016
  """
10017
  REQ_BGL = False
10018

    
10019
  def ExpandNames(self):
10020
    self._ExpandAndLockInstance()
10021

    
10022
  def CheckPrereq(self):
10023
    """Check prerequisites.
10024

10025
    """
10026
    instance_name = self.op.instance_name
10027

    
10028
    self.instance = self.cfg.GetInstanceInfo(instance_name)
10029
    assert self.instance is not None, \
10030
          "Cannot retrieve locked instance %s" % self.op.instance_name
10031
    _CheckNodeOnline(self, self.instance.primary_node)
10032

    
10033
    self._cds = _GetClusterDomainSecret()
10034

    
10035
  def Exec(self, feedback_fn):
10036
    """Prepares an instance for an export.
10037

10038
    """
10039
    instance = self.instance
10040

    
10041
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
10042
      salt = utils.GenerateSecret(8)
10043

    
10044
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
10045
      result = self.rpc.call_x509_cert_create(instance.primary_node,
10046
                                              constants.RIE_CERT_VALIDITY)
10047
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
10048

    
10049
      (name, cert_pem) = result.payload
10050

    
10051
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
10052
                                             cert_pem)
10053

    
10054
      return {
10055
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
10056
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
10057
                          salt),
10058
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
10059
        }
10060

    
10061
    return None
10062

    
10063

    
10064
class LUBackupExport(LogicalUnit):
10065
  """Export an instance to an image in the cluster.
10066

10067
  """
10068
  HPATH = "instance-export"
10069
  HTYPE = constants.HTYPE_INSTANCE
10070
  REQ_BGL = False
10071

    
10072
  def CheckArguments(self):
10073
    """Check the arguments.
10074

10075
    """
10076
    self.x509_key_name = self.op.x509_key_name
10077
    self.dest_x509_ca_pem = self.op.destination_x509_ca
10078

    
10079
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
10080
      if not self.x509_key_name:
10081
        raise errors.OpPrereqError("Missing X509 key name for encryption",
10082
                                   errors.ECODE_INVAL)
10083

    
10084
      if not self.dest_x509_ca_pem:
10085
        raise errors.OpPrereqError("Missing destination X509 CA",
10086
                                   errors.ECODE_INVAL)
10087

    
10088
  def ExpandNames(self):
10089
    self._ExpandAndLockInstance()
10090

    
10091
    # Lock all nodes for local exports
10092
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10093
      # FIXME: lock only instance primary and destination node
10094
      #
10095
      # Sad but true, for now we have do lock all nodes, as we don't know where
10096
      # the previous export might be, and in this LU we search for it and
10097
      # remove it from its current node. In the future we could fix this by:
10098
      #  - making a tasklet to search (share-lock all), then create the
10099
      #    new one, then one to remove, after
10100
      #  - removing the removal operation altogether
10101
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10102

    
10103
  def DeclareLocks(self, level):
10104
    """Last minute lock declaration."""
10105
    # All nodes are locked anyway, so nothing to do here.
10106

    
10107
  def BuildHooksEnv(self):
10108
    """Build hooks env.
10109

10110
    This will run on the master, primary node and target node.
10111

10112
    """
10113
    env = {
10114
      "EXPORT_MODE": self.op.mode,
10115
      "EXPORT_NODE": self.op.target_node,
10116
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
10117
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
10118
      # TODO: Generic function for boolean env variables
10119
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
10120
      }
10121

    
10122
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
10123

    
10124
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
10125

    
10126
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10127
      nl.append(self.op.target_node)
10128

    
10129
    return env, nl, nl
10130

    
10131
  def CheckPrereq(self):
10132
    """Check prerequisites.
10133

10134
    This checks that the instance and node names are valid.
10135

10136
    """
10137
    instance_name = self.op.instance_name
10138

    
10139
    self.instance = self.cfg.GetInstanceInfo(instance_name)
10140
    assert self.instance is not None, \
10141
          "Cannot retrieve locked instance %s" % self.op.instance_name
10142
    _CheckNodeOnline(self, self.instance.primary_node)
10143

    
10144
    if (self.op.remove_instance and self.instance.admin_up and
10145
        not self.op.shutdown):
10146
      raise errors.OpPrereqError("Can not remove instance without shutting it"
10147
                                 " down before")
10148

    
10149
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10150
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
10151
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
10152
      assert self.dst_node is not None
10153

    
10154
      _CheckNodeOnline(self, self.dst_node.name)
10155
      _CheckNodeNotDrained(self, self.dst_node.name)
10156

    
10157
      self._cds = None
10158
      self.dest_disk_info = None
10159
      self.dest_x509_ca = None
10160

    
10161
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10162
      self.dst_node = None
10163

    
10164
      if len(self.op.target_node) != len(self.instance.disks):
10165
        raise errors.OpPrereqError(("Received destination information for %s"
10166
                                    " disks, but instance %s has %s disks") %
10167
                                   (len(self.op.target_node), instance_name,
10168
                                    len(self.instance.disks)),
10169
                                   errors.ECODE_INVAL)
10170

    
10171
      cds = _GetClusterDomainSecret()
10172

    
10173
      # Check X509 key name
10174
      try:
10175
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
10176
      except (TypeError, ValueError), err:
10177
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
10178

    
10179
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
10180
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
10181
                                   errors.ECODE_INVAL)
10182

    
10183
      # Load and verify CA
10184
      try:
10185
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
10186
      except OpenSSL.crypto.Error, err:
10187
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
10188
                                   (err, ), errors.ECODE_INVAL)
10189

    
10190
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
10191
      if errcode is not None:
10192
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
10193
                                   (msg, ), errors.ECODE_INVAL)
10194

    
10195
      self.dest_x509_ca = cert
10196

    
10197
      # Verify target information
10198
      disk_info = []
10199
      for idx, disk_data in enumerate(self.op.target_node):
10200
        try:
10201
          (host, port, magic) = \
10202
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
10203
        except errors.GenericError, err:
10204
          raise errors.OpPrereqError("Target info for disk %s: %s" %
10205
                                     (idx, err), errors.ECODE_INVAL)
10206

    
10207
        disk_info.append((host, port, magic))
10208

    
10209
      assert len(disk_info) == len(self.op.target_node)
10210
      self.dest_disk_info = disk_info
10211

    
10212
    else:
10213
      raise errors.ProgrammerError("Unhandled export mode %r" %
10214
                                   self.op.mode)
10215

    
10216
    # instance disk type verification
10217
    # TODO: Implement export support for file-based disks
10218
    for disk in self.instance.disks:
10219
      if disk.dev_type == constants.LD_FILE:
10220
        raise errors.OpPrereqError("Export not supported for instances with"
10221
                                   " file-based disks", errors.ECODE_INVAL)
10222

    
10223
  def _CleanupExports(self, feedback_fn):
10224
    """Removes exports of current instance from all other nodes.
10225

10226
    If an instance in a cluster with nodes A..D was exported to node C, its
10227
    exports will be removed from the nodes A, B and D.
10228

10229
    """
10230
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
10231

    
10232
    nodelist = self.cfg.GetNodeList()
10233
    nodelist.remove(self.dst_node.name)
10234

    
10235
    # on one-node clusters nodelist will be empty after the removal
10236
    # if we proceed the backup would be removed because OpBackupQuery
10237
    # substitutes an empty list with the full cluster node list.
10238
    iname = self.instance.name
10239
    if nodelist:
10240
      feedback_fn("Removing old exports for instance %s" % iname)
10241
      exportlist = self.rpc.call_export_list(nodelist)
10242
      for node in exportlist:
10243
        if exportlist[node].fail_msg:
10244
          continue
10245
        if iname in exportlist[node].payload:
10246
          msg = self.rpc.call_export_remove(node, iname).fail_msg
10247
          if msg:
10248
            self.LogWarning("Could not remove older export for instance %s"
10249
                            " on node %s: %s", iname, node, msg)
10250

    
10251
  def Exec(self, feedback_fn):
10252
    """Export an instance to an image in the cluster.
10253

10254
    """
10255
    assert self.op.mode in constants.EXPORT_MODES
10256

    
10257
    instance = self.instance
10258
    src_node = instance.primary_node
10259

    
10260
    if self.op.shutdown:
10261
      # shutdown the instance, but not the disks
10262
      feedback_fn("Shutting down instance %s" % instance.name)
10263
      result = self.rpc.call_instance_shutdown(src_node, instance,
10264
                                               self.op.shutdown_timeout)
10265
      # TODO: Maybe ignore failures if ignore_remove_failures is set
10266
      result.Raise("Could not shutdown instance %s on"
10267
                   " node %s" % (instance.name, src_node))
10268

    
10269
    # set the disks ID correctly since call_instance_start needs the
10270
    # correct drbd minor to create the symlinks
10271
    for disk in instance.disks:
10272
      self.cfg.SetDiskID(disk, src_node)
10273

    
10274
    activate_disks = (not instance.admin_up)
10275

    
10276
    if activate_disks:
10277
      # Activate the instance disks if we'exporting a stopped instance
10278
      feedback_fn("Activating disks for %s" % instance.name)
10279
      _StartInstanceDisks(self, instance, None)
10280

    
10281
    try:
10282
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
10283
                                                     instance)
10284

    
10285
      helper.CreateSnapshots()
10286
      try:
10287
        if (self.op.shutdown and instance.admin_up and
10288
            not self.op.remove_instance):
10289
          assert not activate_disks
10290
          feedback_fn("Starting instance %s" % instance.name)
10291
          result = self.rpc.call_instance_start(src_node, instance, None, None)
10292
          msg = result.fail_msg
10293
          if msg:
10294
            feedback_fn("Failed to start instance: %s" % msg)
10295
            _ShutdownInstanceDisks(self, instance)
10296
            raise errors.OpExecError("Could not start instance: %s" % msg)
10297

    
10298
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
10299
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
10300
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10301
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
10302
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
10303

    
10304
          (key_name, _, _) = self.x509_key_name
10305

    
10306
          dest_ca_pem = \
10307
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
10308
                                            self.dest_x509_ca)
10309

    
10310
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
10311
                                                     key_name, dest_ca_pem,
10312
                                                     timeouts)
10313
      finally:
10314
        helper.Cleanup()
10315

    
10316
      # Check for backwards compatibility
10317
      assert len(dresults) == len(instance.disks)
10318
      assert compat.all(isinstance(i, bool) for i in dresults), \
10319
             "Not all results are boolean: %r" % dresults
10320

    
10321
    finally:
10322
      if activate_disks:
10323
        feedback_fn("Deactivating disks for %s" % instance.name)
10324
        _ShutdownInstanceDisks(self, instance)
10325

    
10326
    if not (compat.all(dresults) and fin_resu):
10327
      failures = []
10328
      if not fin_resu:
10329
        failures.append("export finalization")
10330
      if not compat.all(dresults):
10331
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
10332
                               if not dsk)
10333
        failures.append("disk export: disk(s) %s" % fdsk)
10334

    
10335
      raise errors.OpExecError("Export failed, errors in %s" %
10336
                               utils.CommaJoin(failures))
10337

    
10338
    # At this point, the export was successful, we can cleanup/finish
10339

    
10340
    # Remove instance if requested
10341
    if self.op.remove_instance:
10342
      feedback_fn("Removing instance %s" % instance.name)
10343
      _RemoveInstance(self, feedback_fn, instance,
10344
                      self.op.ignore_remove_failures)
10345

    
10346
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10347
      self._CleanupExports(feedback_fn)
10348

    
10349
    return fin_resu, dresults
10350

    
10351

    
10352
class LUBackupRemove(NoHooksLU):
10353
  """Remove exports related to the named instance.
10354

10355
  """
10356
  REQ_BGL = False
10357

    
10358
  def ExpandNames(self):
10359
    self.needed_locks = {}
10360
    # We need all nodes to be locked in order for RemoveExport to work, but we
10361
    # don't need to lock the instance itself, as nothing will happen to it (and
10362
    # we can remove exports also for a removed instance)
10363
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10364

    
10365
  def Exec(self, feedback_fn):
10366
    """Remove any export.
10367

10368
    """
10369
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
10370
    # If the instance was not found we'll try with the name that was passed in.
10371
    # This will only work if it was an FQDN, though.
10372
    fqdn_warn = False
10373
    if not instance_name:
10374
      fqdn_warn = True
10375
      instance_name = self.op.instance_name
10376

    
10377
    locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
10378
    exportlist = self.rpc.call_export_list(locked_nodes)
10379
    found = False
10380
    for node in exportlist:
10381
      msg = exportlist[node].fail_msg
10382
      if msg:
10383
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
10384
        continue
10385
      if instance_name in exportlist[node].payload:
10386
        found = True
10387
        result = self.rpc.call_export_remove(node, instance_name)
10388
        msg = result.fail_msg
10389
        if msg:
10390
          logging.error("Could not remove export for instance %s"
10391
                        " on node %s: %s", instance_name, node, msg)
10392

    
10393
    if fqdn_warn and not found:
10394
      feedback_fn("Export not found. If trying to remove an export belonging"
10395
                  " to a deleted instance please use its Fully Qualified"
10396
                  " Domain Name.")
10397

    
10398

    
10399
class LUGroupAdd(LogicalUnit):
10400
  """Logical unit for creating node groups.
10401

10402
  """
10403
  HPATH = "group-add"
10404
  HTYPE = constants.HTYPE_GROUP
10405
  REQ_BGL = False
10406

    
10407
  def ExpandNames(self):
10408
    # We need the new group's UUID here so that we can create and acquire the
10409
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
10410
    # that it should not check whether the UUID exists in the configuration.
10411
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
10412
    self.needed_locks = {}
10413
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10414

    
10415
  def CheckPrereq(self):
10416
    """Check prerequisites.
10417

10418
    This checks that the given group name is not an existing node group
10419
    already.
10420

10421
    """
10422
    try:
10423
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10424
    except errors.OpPrereqError:
10425
      pass
10426
    else:
10427
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
10428
                                 " node group (UUID: %s)" %
10429
                                 (self.op.group_name, existing_uuid),
10430
                                 errors.ECODE_EXISTS)
10431

    
10432
    if self.op.ndparams:
10433
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10434

    
10435
  def BuildHooksEnv(self):
10436
    """Build hooks env.
10437

10438
    """
10439
    env = {
10440
      "GROUP_NAME": self.op.group_name,
10441
      }
10442
    mn = self.cfg.GetMasterNode()
10443
    return env, [mn], [mn]
10444

    
10445
  def Exec(self, feedback_fn):
10446
    """Add the node group to the cluster.
10447

10448
    """
10449
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
10450
                                  uuid=self.group_uuid,
10451
                                  alloc_policy=self.op.alloc_policy,
10452
                                  ndparams=self.op.ndparams)
10453

    
10454
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
10455
    del self.remove_locks[locking.LEVEL_NODEGROUP]
10456

    
10457

    
10458
class LUGroupAssignNodes(NoHooksLU):
10459
  """Logical unit for assigning nodes to groups.
10460

10461
  """
10462
  REQ_BGL = False
10463

    
10464
  def ExpandNames(self):
10465
    # These raise errors.OpPrereqError on their own:
10466
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10467
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
10468

    
10469
    # We want to lock all the affected nodes and groups. We have readily
10470
    # available the list of nodes, and the *destination* group. To gather the
10471
    # list of "source" groups, we need to fetch node information.
10472
    self.node_data = self.cfg.GetAllNodesInfo()
10473
    affected_groups = set(self.node_data[node].group for node in self.op.nodes)
10474
    affected_groups.add(self.group_uuid)
10475

    
10476
    self.needed_locks = {
10477
      locking.LEVEL_NODEGROUP: list(affected_groups),
10478
      locking.LEVEL_NODE: self.op.nodes,
10479
      }
10480

    
10481
  def CheckPrereq(self):
10482
    """Check prerequisites.
10483

10484
    """
10485
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10486
    instance_data = self.cfg.GetAllInstancesInfo()
10487

    
10488
    if self.group is None:
10489
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10490
                               (self.op.group_name, self.group_uuid))
10491

    
10492
    (new_splits, previous_splits) = \
10493
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
10494
                                             for node in self.op.nodes],
10495
                                            self.node_data, instance_data)
10496

    
10497
    if new_splits:
10498
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
10499

    
10500
      if not self.op.force:
10501
        raise errors.OpExecError("The following instances get split by this"
10502
                                 " change and --force was not given: %s" %
10503
                                 fmt_new_splits)
10504
      else:
10505
        self.LogWarning("This operation will split the following instances: %s",
10506
                        fmt_new_splits)
10507

    
10508
        if previous_splits:
10509
          self.LogWarning("In addition, these already-split instances continue"
10510
                          " to be spit across groups: %s",
10511
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
10512

    
10513
  def Exec(self, feedback_fn):
10514
    """Assign nodes to a new group.
10515

10516
    """
10517
    for node in self.op.nodes:
10518
      self.node_data[node].group = self.group_uuid
10519

    
10520
    self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
10521

    
10522
  @staticmethod
10523
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
10524
    """Check for split instances after a node assignment.
10525

10526
    This method considers a series of node assignments as an atomic operation,
10527
    and returns information about split instances after applying the set of
10528
    changes.
10529

10530
    In particular, it returns information about newly split instances, and
10531
    instances that were already split, and remain so after the change.
10532

10533
    Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
10534
    considered.
10535

10536
    @type changes: list of (node_name, new_group_uuid) pairs.
10537
    @param changes: list of node assignments to consider.
10538
    @param node_data: a dict with data for all nodes
10539
    @param instance_data: a dict with all instances to consider
10540
    @rtype: a two-tuple
10541
    @return: a list of instances that were previously okay and result split as a
10542
      consequence of this change, and a list of instances that were previously
10543
      split and this change does not fix.
10544

10545
    """
10546
    changed_nodes = dict((node, group) for node, group in changes
10547
                         if node_data[node].group != group)
10548

    
10549
    all_split_instances = set()
10550
    previously_split_instances = set()
10551

    
10552
    def InstanceNodes(instance):
10553
      return [instance.primary_node] + list(instance.secondary_nodes)
10554

    
10555
    for inst in instance_data.values():
10556
      if inst.disk_template not in constants.DTS_INT_MIRROR:
10557
        continue
10558

    
10559
      instance_nodes = InstanceNodes(inst)
10560

    
10561
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
10562
        previously_split_instances.add(inst.name)
10563

    
10564
      if len(set(changed_nodes.get(node, node_data[node].group)
10565
                 for node in instance_nodes)) > 1:
10566
        all_split_instances.add(inst.name)
10567

    
10568
    return (list(all_split_instances - previously_split_instances),
10569
            list(previously_split_instances & all_split_instances))
10570

    
10571

    
10572
class _GroupQuery(_QueryBase):
10573

    
10574
  FIELDS = query.GROUP_FIELDS
10575

    
10576
  def ExpandNames(self, lu):
10577
    lu.needed_locks = {}
10578

    
10579
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
10580
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
10581

    
10582
    if not self.names:
10583
      self.wanted = [name_to_uuid[name]
10584
                     for name in utils.NiceSort(name_to_uuid.keys())]
10585
    else:
10586
      # Accept names to be either names or UUIDs.
10587
      missing = []
10588
      self.wanted = []
10589
      all_uuid = frozenset(self._all_groups.keys())
10590

    
10591
      for name in self.names:
10592
        if name in all_uuid:
10593
          self.wanted.append(name)
10594
        elif name in name_to_uuid:
10595
          self.wanted.append(name_to_uuid[name])
10596
        else:
10597
          missing.append(name)
10598

    
10599
      if missing:
10600
        raise errors.OpPrereqError("Some groups do not exist: %s" % missing,
10601
                                   errors.ECODE_NOENT)
10602

    
10603
  def DeclareLocks(self, lu, level):
10604
    pass
10605

    
10606
  def _GetQueryData(self, lu):
10607
    """Computes the list of node groups and their attributes.
10608

10609
    """
10610
    do_nodes = query.GQ_NODE in self.requested_data
10611
    do_instances = query.GQ_INST in self.requested_data
10612

    
10613
    group_to_nodes = None
10614
    group_to_instances = None
10615

    
10616
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
10617
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
10618
    # latter GetAllInstancesInfo() is not enough, for we have to go through
10619
    # instance->node. Hence, we will need to process nodes even if we only need
10620
    # instance information.
10621
    if do_nodes or do_instances:
10622
      all_nodes = lu.cfg.GetAllNodesInfo()
10623
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
10624
      node_to_group = {}
10625

    
10626
      for node in all_nodes.values():
10627
        if node.group in group_to_nodes:
10628
          group_to_nodes[node.group].append(node.name)
10629
          node_to_group[node.name] = node.group
10630

    
10631
      if do_instances:
10632
        all_instances = lu.cfg.GetAllInstancesInfo()
10633
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
10634

    
10635
        for instance in all_instances.values():
10636
          node = instance.primary_node
10637
          if node in node_to_group:
10638
            group_to_instances[node_to_group[node]].append(instance.name)
10639

    
10640
        if not do_nodes:
10641
          # Do not pass on node information if it was not requested.
10642
          group_to_nodes = None
10643

    
10644
    return query.GroupQueryData([self._all_groups[uuid]
10645
                                 for uuid in self.wanted],
10646
                                group_to_nodes, group_to_instances)
10647

    
10648

    
10649
class LUGroupQuery(NoHooksLU):
10650
  """Logical unit for querying node groups.
10651

10652
  """
10653
  REQ_BGL = False
10654

    
10655
  def CheckArguments(self):
10656
    self.gq = _GroupQuery(self.op.names, self.op.output_fields, False)
10657

    
10658
  def ExpandNames(self):
10659
    self.gq.ExpandNames(self)
10660

    
10661
  def Exec(self, feedback_fn):
10662
    return self.gq.OldStyleQuery(self)
10663

    
10664

    
10665
class LUGroupSetParams(LogicalUnit):
10666
  """Modifies the parameters of a node group.
10667

10668
  """
10669
  HPATH = "group-modify"
10670
  HTYPE = constants.HTYPE_GROUP
10671
  REQ_BGL = False
10672

    
10673
  def CheckArguments(self):
10674
    all_changes = [
10675
      self.op.ndparams,
10676
      self.op.alloc_policy,
10677
      ]
10678

    
10679
    if all_changes.count(None) == len(all_changes):
10680
      raise errors.OpPrereqError("Please pass at least one modification",
10681
                                 errors.ECODE_INVAL)
10682

    
10683
  def ExpandNames(self):
10684
    # This raises errors.OpPrereqError on its own:
10685
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10686

    
10687
    self.needed_locks = {
10688
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10689
      }
10690

    
10691
  def CheckPrereq(self):
10692
    """Check prerequisites.
10693

10694
    """
10695
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10696

    
10697
    if self.group is None:
10698
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10699
                               (self.op.group_name, self.group_uuid))
10700

    
10701
    if self.op.ndparams:
10702
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
10703
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10704
      self.new_ndparams = new_ndparams
10705

    
10706
  def BuildHooksEnv(self):
10707
    """Build hooks env.
10708

10709
    """
10710
    env = {
10711
      "GROUP_NAME": self.op.group_name,
10712
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
10713
      }
10714
    mn = self.cfg.GetMasterNode()
10715
    return env, [mn], [mn]
10716

    
10717
  def Exec(self, feedback_fn):
10718
    """Modifies the node group.
10719

10720
    """
10721
    result = []
10722

    
10723
    if self.op.ndparams:
10724
      self.group.ndparams = self.new_ndparams
10725
      result.append(("ndparams", str(self.group.ndparams)))
10726

    
10727
    if self.op.alloc_policy:
10728
      self.group.alloc_policy = self.op.alloc_policy
10729

    
10730
    self.cfg.Update(self.group, feedback_fn)
10731
    return result
10732

    
10733

    
10734

    
10735
class LUGroupRemove(LogicalUnit):
10736
  HPATH = "group-remove"
10737
  HTYPE = constants.HTYPE_GROUP
10738
  REQ_BGL = False
10739

    
10740
  def ExpandNames(self):
10741
    # This will raises errors.OpPrereqError on its own:
10742
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10743
    self.needed_locks = {
10744
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10745
      }
10746

    
10747
  def CheckPrereq(self):
10748
    """Check prerequisites.
10749

10750
    This checks that the given group name exists as a node group, that is
10751
    empty (i.e., contains no nodes), and that is not the last group of the
10752
    cluster.
10753

10754
    """
10755
    # Verify that the group is empty.
10756
    group_nodes = [node.name
10757
                   for node in self.cfg.GetAllNodesInfo().values()
10758
                   if node.group == self.group_uuid]
10759

    
10760
    if group_nodes:
10761
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
10762
                                 " nodes: %s" %
10763
                                 (self.op.group_name,
10764
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
10765
                                 errors.ECODE_STATE)
10766

    
10767
    # Verify the cluster would not be left group-less.
10768
    if len(self.cfg.GetNodeGroupList()) == 1:
10769
      raise errors.OpPrereqError("Group '%s' is the only group,"
10770
                                 " cannot be removed" %
10771
                                 self.op.group_name,
10772
                                 errors.ECODE_STATE)
10773

    
10774
  def BuildHooksEnv(self):
10775
    """Build hooks env.
10776

10777
    """
10778
    env = {
10779
      "GROUP_NAME": self.op.group_name,
10780
      }
10781
    mn = self.cfg.GetMasterNode()
10782
    return env, [mn], [mn]
10783

    
10784
  def Exec(self, feedback_fn):
10785
    """Remove the node group.
10786

10787
    """
10788
    try:
10789
      self.cfg.RemoveNodeGroup(self.group_uuid)
10790
    except errors.ConfigurationError:
10791
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
10792
                               (self.op.group_name, self.group_uuid))
10793

    
10794
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10795

    
10796

    
10797
class LUGroupRename(LogicalUnit):
10798
  HPATH = "group-rename"
10799
  HTYPE = constants.HTYPE_GROUP
10800
  REQ_BGL = False
10801

    
10802
  def ExpandNames(self):
10803
    # This raises errors.OpPrereqError on its own:
10804
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.old_name)
10805

    
10806
    self.needed_locks = {
10807
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10808
      }
10809

    
10810
  def CheckPrereq(self):
10811
    """Check prerequisites.
10812

10813
    This checks that the given old_name exists as a node group, and that
10814
    new_name doesn't.
10815

10816
    """
10817
    try:
10818
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
10819
    except errors.OpPrereqError:
10820
      pass
10821
    else:
10822
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
10823
                                 " node group (UUID: %s)" %
10824
                                 (self.op.new_name, new_name_uuid),
10825
                                 errors.ECODE_EXISTS)
10826

    
10827
  def BuildHooksEnv(self):
10828
    """Build hooks env.
10829

10830
    """
10831
    env = {
10832
      "OLD_NAME": self.op.old_name,
10833
      "NEW_NAME": self.op.new_name,
10834
      }
10835

    
10836
    mn = self.cfg.GetMasterNode()
10837
    all_nodes = self.cfg.GetAllNodesInfo()
10838
    run_nodes = [mn]
10839
    all_nodes.pop(mn, None)
10840

    
10841
    for node in all_nodes.values():
10842
      if node.group == self.group_uuid:
10843
        run_nodes.append(node.name)
10844

    
10845
    return env, run_nodes, run_nodes
10846

    
10847
  def Exec(self, feedback_fn):
10848
    """Rename the node group.
10849

10850
    """
10851
    group = self.cfg.GetNodeGroup(self.group_uuid)
10852

    
10853
    if group is None:
10854
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10855
                               (self.op.old_name, self.group_uuid))
10856

    
10857
    group.name = self.op.new_name
10858
    self.cfg.Update(group, feedback_fn)
10859

    
10860
    return self.op.new_name
10861

    
10862

    
10863
class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
10864
  """Generic tags LU.
10865

10866
  This is an abstract class which is the parent of all the other tags LUs.
10867

10868
  """
10869

    
10870
  def ExpandNames(self):
10871
    self.needed_locks = {}
10872
    if self.op.kind == constants.TAG_NODE:
10873
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
10874
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
10875
    elif self.op.kind == constants.TAG_INSTANCE:
10876
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
10877
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
10878

    
10879
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
10880
    # not possible to acquire the BGL based on opcode parameters)
10881

    
10882
  def CheckPrereq(self):
10883
    """Check prerequisites.
10884

10885
    """
10886
    if self.op.kind == constants.TAG_CLUSTER:
10887
      self.target = self.cfg.GetClusterInfo()
10888
    elif self.op.kind == constants.TAG_NODE:
10889
      self.target = self.cfg.GetNodeInfo(self.op.name)
10890
    elif self.op.kind == constants.TAG_INSTANCE:
10891
      self.target = self.cfg.GetInstanceInfo(self.op.name)
10892
    else:
10893
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
10894
                                 str(self.op.kind), errors.ECODE_INVAL)
10895

    
10896

    
10897
class LUTagsGet(TagsLU):
10898
  """Returns the tags of a given object.
10899

10900
  """
10901
  REQ_BGL = False
10902

    
10903
  def ExpandNames(self):
10904
    TagsLU.ExpandNames(self)
10905

    
10906
    # Share locks as this is only a read operation
10907
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
10908

    
10909
  def Exec(self, feedback_fn):
10910
    """Returns the tag list.
10911

10912
    """
10913
    return list(self.target.GetTags())
10914

    
10915

    
10916
class LUTagsSearch(NoHooksLU):
10917
  """Searches the tags for a given pattern.
10918

10919
  """
10920
  REQ_BGL = False
10921

    
10922
  def ExpandNames(self):
10923
    self.needed_locks = {}
10924

    
10925
  def CheckPrereq(self):
10926
    """Check prerequisites.
10927

10928
    This checks the pattern passed for validity by compiling it.
10929

10930
    """
10931
    try:
10932
      self.re = re.compile(self.op.pattern)
10933
    except re.error, err:
10934
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
10935
                                 (self.op.pattern, err), errors.ECODE_INVAL)
10936

    
10937
  def Exec(self, feedback_fn):
10938
    """Returns the tag list.
10939

10940
    """
10941
    cfg = self.cfg
10942
    tgts = [("/cluster", cfg.GetClusterInfo())]
10943
    ilist = cfg.GetAllInstancesInfo().values()
10944
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
10945
    nlist = cfg.GetAllNodesInfo().values()
10946
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
10947
    results = []
10948
    for path, target in tgts:
10949
      for tag in target.GetTags():
10950
        if self.re.search(tag):
10951
          results.append((path, tag))
10952
    return results
10953

    
10954

    
10955
class LUTagsSet(TagsLU):
10956
  """Sets a tag on a given object.
10957

10958
  """
10959
  REQ_BGL = False
10960

    
10961
  def CheckPrereq(self):
10962
    """Check prerequisites.
10963

10964
    This checks the type and length of the tag name and value.
10965

10966
    """
10967
    TagsLU.CheckPrereq(self)
10968
    for tag in self.op.tags:
10969
      objects.TaggableObject.ValidateTag(tag)
10970

    
10971
  def Exec(self, feedback_fn):
10972
    """Sets the tag.
10973

10974
    """
10975
    try:
10976
      for tag in self.op.tags:
10977
        self.target.AddTag(tag)
10978
    except errors.TagError, err:
10979
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
10980
    self.cfg.Update(self.target, feedback_fn)
10981

    
10982

    
10983
class LUTagsDel(TagsLU):
10984
  """Delete a list of tags from a given object.
10985

10986
  """
10987
  REQ_BGL = False
10988

    
10989
  def CheckPrereq(self):
10990
    """Check prerequisites.
10991

10992
    This checks that we have the given tag.
10993

10994
    """
10995
    TagsLU.CheckPrereq(self)
10996
    for tag in self.op.tags:
10997
      objects.TaggableObject.ValidateTag(tag)
10998
    del_tags = frozenset(self.op.tags)
10999
    cur_tags = self.target.GetTags()
11000

    
11001
    diff_tags = del_tags - cur_tags
11002
    if diff_tags:
11003
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
11004
      raise errors.OpPrereqError("Tag(s) %s not found" %
11005
                                 (utils.CommaJoin(diff_names), ),
11006
                                 errors.ECODE_NOENT)
11007

    
11008
  def Exec(self, feedback_fn):
11009
    """Remove the tag from the object.
11010

11011
    """
11012
    for tag in self.op.tags:
11013
      self.target.RemoveTag(tag)
11014
    self.cfg.Update(self.target, feedback_fn)
11015

    
11016

    
11017
class LUTestDelay(NoHooksLU):
11018
  """Sleep for a specified amount of time.
11019

11020
  This LU sleeps on the master and/or nodes for a specified amount of
11021
  time.
11022

11023
  """
11024
  REQ_BGL = False
11025

    
11026
  def ExpandNames(self):
11027
    """Expand names and set required locks.
11028

11029
    This expands the node list, if any.
11030

11031
    """
11032
    self.needed_locks = {}
11033
    if self.op.on_nodes:
11034
      # _GetWantedNodes can be used here, but is not always appropriate to use
11035
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
11036
      # more information.
11037
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
11038
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
11039

    
11040
  def _TestDelay(self):
11041
    """Do the actual sleep.
11042

11043
    """
11044
    if self.op.on_master:
11045
      if not utils.TestDelay(self.op.duration):
11046
        raise errors.OpExecError("Error during master delay test")
11047
    if self.op.on_nodes:
11048
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
11049
      for node, node_result in result.items():
11050
        node_result.Raise("Failure during rpc call to node %s" % node)
11051

    
11052
  def Exec(self, feedback_fn):
11053
    """Execute the test delay opcode, with the wanted repetitions.
11054

11055
    """
11056
    if self.op.repeat == 0:
11057
      self._TestDelay()
11058
    else:
11059
      top_value = self.op.repeat - 1
11060
      for i in range(self.op.repeat):
11061
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
11062
        self._TestDelay()
11063

    
11064

    
11065
class LUTestJqueue(NoHooksLU):
11066
  """Utility LU to test some aspects of the job queue.
11067

11068
  """
11069
  REQ_BGL = False
11070

    
11071
  # Must be lower than default timeout for WaitForJobChange to see whether it
11072
  # notices changed jobs
11073
  _CLIENT_CONNECT_TIMEOUT = 20.0
11074
  _CLIENT_CONFIRM_TIMEOUT = 60.0
11075

    
11076
  @classmethod
11077
  def _NotifyUsingSocket(cls, cb, errcls):
11078
    """Opens a Unix socket and waits for another program to connect.
11079

11080
    @type cb: callable
11081
    @param cb: Callback to send socket name to client
11082
    @type errcls: class
11083
    @param errcls: Exception class to use for errors
11084

11085
    """
11086
    # Using a temporary directory as there's no easy way to create temporary
11087
    # sockets without writing a custom loop around tempfile.mktemp and
11088
    # socket.bind
11089
    tmpdir = tempfile.mkdtemp()
11090
    try:
11091
      tmpsock = utils.PathJoin(tmpdir, "sock")
11092

    
11093
      logging.debug("Creating temporary socket at %s", tmpsock)
11094
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
11095
      try:
11096
        sock.bind(tmpsock)
11097
        sock.listen(1)
11098

    
11099
        # Send details to client
11100
        cb(tmpsock)
11101

    
11102
        # Wait for client to connect before continuing
11103
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
11104
        try:
11105
          (conn, _) = sock.accept()
11106
        except socket.error, err:
11107
          raise errcls("Client didn't connect in time (%s)" % err)
11108
      finally:
11109
        sock.close()
11110
    finally:
11111
      # Remove as soon as client is connected
11112
      shutil.rmtree(tmpdir)
11113

    
11114
    # Wait for client to close
11115
    try:
11116
      try:
11117
        # pylint: disable-msg=E1101
11118
        # Instance of '_socketobject' has no ... member
11119
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
11120
        conn.recv(1)
11121
      except socket.error, err:
11122
        raise errcls("Client failed to confirm notification (%s)" % err)
11123
    finally:
11124
      conn.close()
11125

    
11126
  def _SendNotification(self, test, arg, sockname):
11127
    """Sends a notification to the client.
11128

11129
    @type test: string
11130
    @param test: Test name
11131
    @param arg: Test argument (depends on test)
11132
    @type sockname: string
11133
    @param sockname: Socket path
11134

11135
    """
11136
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
11137

    
11138
  def _Notify(self, prereq, test, arg):
11139
    """Notifies the client of a test.
11140

11141
    @type prereq: bool
11142
    @param prereq: Whether this is a prereq-phase test
11143
    @type test: string
11144
    @param test: Test name
11145
    @param arg: Test argument (depends on test)
11146

11147
    """
11148
    if prereq:
11149
      errcls = errors.OpPrereqError
11150
    else:
11151
      errcls = errors.OpExecError
11152

    
11153
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
11154
                                                  test, arg),
11155
                                   errcls)
11156

    
11157
  def CheckArguments(self):
11158
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
11159
    self.expandnames_calls = 0
11160

    
11161
  def ExpandNames(self):
11162
    checkargs_calls = getattr(self, "checkargs_calls", 0)
11163
    if checkargs_calls < 1:
11164
      raise errors.ProgrammerError("CheckArguments was not called")
11165

    
11166
    self.expandnames_calls += 1
11167

    
11168
    if self.op.notify_waitlock:
11169
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
11170

    
11171
    self.LogInfo("Expanding names")
11172

    
11173
    # Get lock on master node (just to get a lock, not for a particular reason)
11174
    self.needed_locks = {
11175
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
11176
      }
11177

    
11178
  def Exec(self, feedback_fn):
11179
    if self.expandnames_calls < 1:
11180
      raise errors.ProgrammerError("ExpandNames was not called")
11181

    
11182
    if self.op.notify_exec:
11183
      self._Notify(False, constants.JQT_EXEC, None)
11184

    
11185
    self.LogInfo("Executing")
11186

    
11187
    if self.op.log_messages:
11188
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
11189
      for idx, msg in enumerate(self.op.log_messages):
11190
        self.LogInfo("Sending log message %s", idx + 1)
11191
        feedback_fn(constants.JQT_MSGPREFIX + msg)
11192
        # Report how many test messages have been sent
11193
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
11194

    
11195
    if self.op.fail:
11196
      raise errors.OpExecError("Opcode failure was requested")
11197

    
11198
    return True
11199

    
11200

    
11201
class IAllocator(object):
11202
  """IAllocator framework.
11203

11204
  An IAllocator instance has three sets of attributes:
11205
    - cfg that is needed to query the cluster
11206
    - input data (all members of the _KEYS class attribute are required)
11207
    - four buffer attributes (in|out_data|text), that represent the
11208
      input (to the external script) in text and data structure format,
11209
      and the output from it, again in two formats
11210
    - the result variables from the script (success, info, nodes) for
11211
      easy usage
11212

11213
  """
11214
  # pylint: disable-msg=R0902
11215
  # lots of instance attributes
11216
  _ALLO_KEYS = [
11217
    "name", "mem_size", "disks", "disk_template",
11218
    "os", "tags", "nics", "vcpus", "hypervisor",
11219
    ]
11220
  _RELO_KEYS = [
11221
    "name", "relocate_from",
11222
    ]
11223
  _EVAC_KEYS = [
11224
    "evac_nodes",
11225
    ]
11226

    
11227
  def __init__(self, cfg, rpc, mode, **kwargs):
11228
    self.cfg = cfg
11229
    self.rpc = rpc
11230
    # init buffer variables
11231
    self.in_text = self.out_text = self.in_data = self.out_data = None
11232
    # init all input fields so that pylint is happy
11233
    self.mode = mode
11234
    self.mem_size = self.disks = self.disk_template = None
11235
    self.os = self.tags = self.nics = self.vcpus = None
11236
    self.hypervisor = None
11237
    self.relocate_from = None
11238
    self.name = None
11239
    self.evac_nodes = None
11240
    # computed fields
11241
    self.required_nodes = None
11242
    # init result fields
11243
    self.success = self.info = self.result = None
11244
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11245
      keyset = self._ALLO_KEYS
11246
      fn = self._AddNewInstance
11247
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11248
      keyset = self._RELO_KEYS
11249
      fn = self._AddRelocateInstance
11250
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11251
      keyset = self._EVAC_KEYS
11252
      fn = self._AddEvacuateNodes
11253
    else:
11254
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
11255
                                   " IAllocator" % self.mode)
11256
    for key in kwargs:
11257
      if key not in keyset:
11258
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
11259
                                     " IAllocator" % key)
11260
      setattr(self, key, kwargs[key])
11261

    
11262
    for key in keyset:
11263
      if key not in kwargs:
11264
        raise errors.ProgrammerError("Missing input parameter '%s' to"
11265
                                     " IAllocator" % key)
11266
    self._BuildInputData(fn)
11267

    
11268
  def _ComputeClusterData(self):
11269
    """Compute the generic allocator input data.
11270

11271
    This is the data that is independent of the actual operation.
11272

11273
    """
11274
    cfg = self.cfg
11275
    cluster_info = cfg.GetClusterInfo()
11276
    # cluster data
11277
    data = {
11278
      "version": constants.IALLOCATOR_VERSION,
11279
      "cluster_name": cfg.GetClusterName(),
11280
      "cluster_tags": list(cluster_info.GetTags()),
11281
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
11282
      # we don't have job IDs
11283
      }
11284
    ninfo = cfg.GetAllNodesInfo()
11285
    iinfo = cfg.GetAllInstancesInfo().values()
11286
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
11287

    
11288
    # node data
11289
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
11290

    
11291
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11292
      hypervisor_name = self.hypervisor
11293
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11294
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
11295
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11296
      hypervisor_name = cluster_info.enabled_hypervisors[0]
11297

    
11298
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
11299
                                        hypervisor_name)
11300
    node_iinfo = \
11301
      self.rpc.call_all_instances_info(node_list,
11302
                                       cluster_info.enabled_hypervisors)
11303

    
11304
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
11305

    
11306
    config_ndata = self._ComputeBasicNodeData(ninfo)
11307
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
11308
                                                 i_list, config_ndata)
11309
    assert len(data["nodes"]) == len(ninfo), \
11310
        "Incomplete node data computed"
11311

    
11312
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
11313

    
11314
    self.in_data = data
11315

    
11316
  @staticmethod
11317
  def _ComputeNodeGroupData(cfg):
11318
    """Compute node groups data.
11319

11320
    """
11321
    ng = {}
11322
    for guuid, gdata in cfg.GetAllNodeGroupsInfo().items():
11323
      ng[guuid] = {
11324
        "name": gdata.name,
11325
        "alloc_policy": gdata.alloc_policy,
11326
        }
11327
    return ng
11328

    
11329
  @staticmethod
11330
  def _ComputeBasicNodeData(node_cfg):
11331
    """Compute global node data.
11332

11333
    @rtype: dict
11334
    @returns: a dict of name: (node dict, node config)
11335

11336
    """
11337
    node_results = {}
11338
    for ninfo in node_cfg.values():
11339
      # fill in static (config-based) values
11340
      pnr = {
11341
        "tags": list(ninfo.GetTags()),
11342
        "primary_ip": ninfo.primary_ip,
11343
        "secondary_ip": ninfo.secondary_ip,
11344
        "offline": ninfo.offline,
11345
        "drained": ninfo.drained,
11346
        "master_candidate": ninfo.master_candidate,
11347
        "group": ninfo.group,
11348
        "master_capable": ninfo.master_capable,
11349
        "vm_capable": ninfo.vm_capable,
11350
        }
11351

    
11352
      node_results[ninfo.name] = pnr
11353

    
11354
    return node_results
11355

    
11356
  @staticmethod
11357
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
11358
                              node_results):
11359
    """Compute global node data.
11360

11361
    @param node_results: the basic node structures as filled from the config
11362

11363
    """
11364
    # make a copy of the current dict
11365
    node_results = dict(node_results)
11366
    for nname, nresult in node_data.items():
11367
      assert nname in node_results, "Missing basic data for node %s" % nname
11368
      ninfo = node_cfg[nname]
11369

    
11370
      if not (ninfo.offline or ninfo.drained):
11371
        nresult.Raise("Can't get data for node %s" % nname)
11372
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
11373
                                nname)
11374
        remote_info = nresult.payload
11375

    
11376
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
11377
                     'vg_size', 'vg_free', 'cpu_total']:
11378
          if attr not in remote_info:
11379
            raise errors.OpExecError("Node '%s' didn't return attribute"
11380
                                     " '%s'" % (nname, attr))
11381
          if not isinstance(remote_info[attr], int):
11382
            raise errors.OpExecError("Node '%s' returned invalid value"
11383
                                     " for '%s': %s" %
11384
                                     (nname, attr, remote_info[attr]))
11385
        # compute memory used by primary instances
11386
        i_p_mem = i_p_up_mem = 0
11387
        for iinfo, beinfo in i_list:
11388
          if iinfo.primary_node == nname:
11389
            i_p_mem += beinfo[constants.BE_MEMORY]
11390
            if iinfo.name not in node_iinfo[nname].payload:
11391
              i_used_mem = 0
11392
            else:
11393
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
11394
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
11395
            remote_info['memory_free'] -= max(0, i_mem_diff)
11396

    
11397
            if iinfo.admin_up:
11398
              i_p_up_mem += beinfo[constants.BE_MEMORY]
11399

    
11400
        # compute memory used by instances
11401
        pnr_dyn = {
11402
          "total_memory": remote_info['memory_total'],
11403
          "reserved_memory": remote_info['memory_dom0'],
11404
          "free_memory": remote_info['memory_free'],
11405
          "total_disk": remote_info['vg_size'],
11406
          "free_disk": remote_info['vg_free'],
11407
          "total_cpus": remote_info['cpu_total'],
11408
          "i_pri_memory": i_p_mem,
11409
          "i_pri_up_memory": i_p_up_mem,
11410
          }
11411
        pnr_dyn.update(node_results[nname])
11412
        node_results[nname] = pnr_dyn
11413

    
11414
    return node_results
11415

    
11416
  @staticmethod
11417
  def _ComputeInstanceData(cluster_info, i_list):
11418
    """Compute global instance data.
11419

11420
    """
11421
    instance_data = {}
11422
    for iinfo, beinfo in i_list:
11423
      nic_data = []
11424
      for nic in iinfo.nics:
11425
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
11426
        nic_dict = {"mac": nic.mac,
11427
                    "ip": nic.ip,
11428
                    "mode": filled_params[constants.NIC_MODE],
11429
                    "link": filled_params[constants.NIC_LINK],
11430
                   }
11431
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
11432
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
11433
        nic_data.append(nic_dict)
11434
      pir = {
11435
        "tags": list(iinfo.GetTags()),
11436
        "admin_up": iinfo.admin_up,
11437
        "vcpus": beinfo[constants.BE_VCPUS],
11438
        "memory": beinfo[constants.BE_MEMORY],
11439
        "os": iinfo.os,
11440
        "nics": nic_data,
11441
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
11442
        "disk_template": iinfo.disk_template,
11443
        "hypervisor": iinfo.hypervisor,
11444
        }
11445
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
11446
                                                 pir["disks"])
11447
      # hail's relocation mode does not work without secondaries,
11448
      # as it exclusively tries replace-secondary moves. So, let's trick hail
11449
      # by specifying our primary and secondary node to be the same.
11450
      if iinfo.disk_template in constants.DTS_EXT_MIRROR:
11451
        pir["nodes"] = [iinfo.primary_node, iinfo.primary_node]
11452
      else:
11453
        pir["nodes"] = [iinfo.primary_node] + list(iinfo.secondary_nodes)
11454
      instance_data[iinfo.name] = pir
11455

    
11456
    return instance_data
11457

    
11458
  def _AddNewInstance(self):
11459
    """Add new instance data to allocator structure.
11460

11461
    This in combination with _AllocatorGetClusterData will create the
11462
    correct structure needed as input for the allocator.
11463

11464
    The checks for the completeness of the opcode must have already been
11465
    done.
11466

11467
    """
11468
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
11469

    
11470
    if self.disk_template in constants.DTS_INT_MIRROR:
11471
      self.required_nodes = 2
11472
    else:
11473
      self.required_nodes = 1
11474
    request = {
11475
      "name": self.name,
11476
      "disk_template": self.disk_template,
11477
      "tags": self.tags,
11478
      "os": self.os,
11479
      "vcpus": self.vcpus,
11480
      "memory": self.mem_size,
11481
      "disks": self.disks,
11482
      "disk_space_total": disk_space,
11483
      "nics": self.nics,
11484
      "required_nodes": self.required_nodes,
11485
      }
11486
    return request
11487

    
11488
  def _AddRelocateInstance(self):
11489
    """Add relocate instance data to allocator structure.
11490

11491
    This in combination with _IAllocatorGetClusterData will create the
11492
    correct structure needed as input for the allocator.
11493

11494
    The checks for the completeness of the opcode must have already been
11495
    done.
11496

11497
    """
11498
    instance = self.cfg.GetInstanceInfo(self.name)
11499
    if instance is None:
11500
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
11501
                                   " IAllocator" % self.name)
11502

    
11503
    if instance.disk_template not in constants.DTS_MIRRORED:
11504
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
11505
                                 errors.ECODE_INVAL)
11506

    
11507
    if instance.disk_template in constants.DTS_INT_MIRROR and \
11508
        len(instance.secondary_nodes) != 1:
11509
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
11510
                                 errors.ECODE_STATE)
11511

    
11512
    self.required_nodes = 1
11513
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
11514
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
11515

    
11516
    request = {
11517
      "name": self.name,
11518
      "disk_space_total": disk_space,
11519
      "required_nodes": self.required_nodes,
11520
      "relocate_from": self.relocate_from,
11521
      }
11522
    return request
11523

    
11524
  def _AddEvacuateNodes(self):
11525
    """Add evacuate nodes data to allocator structure.
11526

11527
    """
11528
    request = {
11529
      "evac_nodes": self.evac_nodes
11530
      }
11531
    return request
11532

    
11533
  def _BuildInputData(self, fn):
11534
    """Build input data structures.
11535

11536
    """
11537
    self._ComputeClusterData()
11538

    
11539
    request = fn()
11540
    request["type"] = self.mode
11541
    self.in_data["request"] = request
11542

    
11543
    self.in_text = serializer.Dump(self.in_data)
11544

    
11545
  def Run(self, name, validate=True, call_fn=None):
11546
    """Run an instance allocator and return the results.
11547

11548
    """
11549
    if call_fn is None:
11550
      call_fn = self.rpc.call_iallocator_runner
11551

    
11552
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
11553
    result.Raise("Failure while running the iallocator script")
11554

    
11555
    self.out_text = result.payload
11556
    if validate:
11557
      self._ValidateResult()
11558

    
11559
  def _ValidateResult(self):
11560
    """Process the allocator results.
11561

11562
    This will process and if successful save the result in
11563
    self.out_data and the other parameters.
11564

11565
    """
11566
    try:
11567
      rdict = serializer.Load(self.out_text)
11568
    except Exception, err:
11569
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
11570

    
11571
    if not isinstance(rdict, dict):
11572
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
11573

    
11574
    # TODO: remove backwards compatiblity in later versions
11575
    if "nodes" in rdict and "result" not in rdict:
11576
      rdict["result"] = rdict["nodes"]
11577
      del rdict["nodes"]
11578

    
11579
    for key in "success", "info", "result":
11580
      if key not in rdict:
11581
        raise errors.OpExecError("Can't parse iallocator results:"
11582
                                 " missing key '%s'" % key)
11583
      setattr(self, key, rdict[key])
11584

    
11585
    if not isinstance(rdict["result"], list):
11586
      raise errors.OpExecError("Can't parse iallocator results: 'result' key"
11587
                               " is not a list")
11588
    self.out_data = rdict
11589

    
11590

    
11591
class LUTestAllocator(NoHooksLU):
11592
  """Run allocator tests.
11593

11594
  This LU runs the allocator tests
11595

11596
  """
11597
  def CheckPrereq(self):
11598
    """Check prerequisites.
11599

11600
    This checks the opcode parameters depending on the director and mode test.
11601

11602
    """
11603
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11604
      for attr in ["mem_size", "disks", "disk_template",
11605
                   "os", "tags", "nics", "vcpus"]:
11606
        if not hasattr(self.op, attr):
11607
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
11608
                                     attr, errors.ECODE_INVAL)
11609
      iname = self.cfg.ExpandInstanceName(self.op.name)
11610
      if iname is not None:
11611
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
11612
                                   iname, errors.ECODE_EXISTS)
11613
      if not isinstance(self.op.nics, list):
11614
        raise errors.OpPrereqError("Invalid parameter 'nics'",
11615
                                   errors.ECODE_INVAL)
11616
      if not isinstance(self.op.disks, list):
11617
        raise errors.OpPrereqError("Invalid parameter 'disks'",
11618
                                   errors.ECODE_INVAL)
11619
      for row in self.op.disks:
11620
        if (not isinstance(row, dict) or
11621
            "size" not in row or
11622
            not isinstance(row["size"], int) or
11623
            "mode" not in row or
11624
            row["mode"] not in ['r', 'w']):
11625
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
11626
                                     " parameter", errors.ECODE_INVAL)
11627
      if self.op.hypervisor is None:
11628
        self.op.hypervisor = self.cfg.GetHypervisorType()
11629
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11630
      fname = _ExpandInstanceName(self.cfg, self.op.name)
11631
      self.op.name = fname
11632
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
11633
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11634
      if not hasattr(self.op, "evac_nodes"):
11635
        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
11636
                                   " opcode input", errors.ECODE_INVAL)
11637
    else:
11638
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
11639
                                 self.op.mode, errors.ECODE_INVAL)
11640

    
11641
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
11642
      if self.op.allocator is None:
11643
        raise errors.OpPrereqError("Missing allocator name",
11644
                                   errors.ECODE_INVAL)
11645
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
11646
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
11647
                                 self.op.direction, errors.ECODE_INVAL)
11648

    
11649
  def Exec(self, feedback_fn):
11650
    """Run the allocator test.
11651

11652
    """
11653
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11654
      ial = IAllocator(self.cfg, self.rpc,
11655
                       mode=self.op.mode,
11656
                       name=self.op.name,
11657
                       mem_size=self.op.mem_size,
11658
                       disks=self.op.disks,
11659
                       disk_template=self.op.disk_template,
11660
                       os=self.op.os,
11661
                       tags=self.op.tags,
11662
                       nics=self.op.nics,
11663
                       vcpus=self.op.vcpus,
11664
                       hypervisor=self.op.hypervisor,
11665
                       )
11666
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11667
      ial = IAllocator(self.cfg, self.rpc,
11668
                       mode=self.op.mode,
11669
                       name=self.op.name,
11670
                       relocate_from=list(self.relocate_from),
11671
                       )
11672
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11673
      ial = IAllocator(self.cfg, self.rpc,
11674
                       mode=self.op.mode,
11675
                       evac_nodes=self.op.evac_nodes)
11676
    else:
11677
      raise errors.ProgrammerError("Uncatched mode %s in"
11678
                                   " LUTestAllocator.Exec", self.op.mode)
11679

    
11680
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
11681
      result = ial.in_text
11682
    else:
11683
      ial.Run(self.op.allocator, validate=False)
11684
      result = ial.out_text
11685
    return result
11686

    
11687

    
11688
#: Query type implementations
11689
_QUERY_IMPL = {
11690
  constants.QR_INSTANCE: _InstanceQuery,
11691
  constants.QR_NODE: _NodeQuery,
11692
  constants.QR_GROUP: _GroupQuery,
11693
  }
11694

    
11695

    
11696
def _GetQueryImplementation(name):
11697
  """Returns the implemtnation for a query type.
11698

11699
  @param name: Query type, must be one of L{constants.QR_OP_QUERY}
11700

11701
  """
11702
  try:
11703
    return _QUERY_IMPL[name]
11704
  except KeyError:
11705
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
11706
                               errors.ECODE_INVAL)