Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 9d5d703e

History | View | Annotate | Download (413.4 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

    
6239
      if len(self.lu.tasklets) == 1:
6240
        # It is safe to remove locks only when we're the only tasklet in the LU
6241
        nodes_keep = [instance.primary_node, target_node]
6242
        nodes_rel = [node for node in self.lu.acquired_locks[locking.LEVEL_NODE]
6243
                     if node not in nodes_keep]
6244
        self.lu.context.glm.release(locking.LEVEL_NODE, nodes_rel)
6245
        self.lu.acquired_locks[locking.LEVEL_NODE] = nodes_keep
6246

    
6247
    else:
6248
      secondary_nodes = instance.secondary_nodes
6249
      if not secondary_nodes:
6250
        raise errors.ConfigurationError("No secondary node but using"
6251
                                        " %s disk template" %
6252
                                        instance.disk_template)
6253
      target_node = secondary_nodes[0]
6254
      if self.lu.op.iallocator or (self.lu.op.target_node and
6255
                                   self.lu.op.target_node != target_node):
6256
        raise errors.OpPrereqError("Instances with disk template %s cannot"
6257
                                   " be migrated over to arbitrary nodes"
6258
                                   " (neither an iallocator nor a target"
6259
                                   " node can be passed)" %
6260
                                   instance.disk_template, errors.ECODE_INVAL)
6261

    
6262
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
6263

    
6264
    # check memory requirements on the secondary node
6265
    _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
6266
                         instance.name, i_be[constants.BE_MEMORY],
6267
                         instance.hypervisor)
6268

    
6269
    # check bridge existance
6270
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
6271

    
6272
    if not self.cleanup:
6273
      _CheckNodeNotDrained(self.lu, target_node)
6274
      result = self.rpc.call_instance_migratable(instance.primary_node,
6275
                                                 instance)
6276
      result.Raise("Can't migrate, please use failover",
6277
                   prereq=True, ecode=errors.ECODE_STATE)
6278

    
6279

    
6280
    if self.lu.op.live is not None and self.lu.op.mode is not None:
6281
      raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
6282
                                 " parameters are accepted",
6283
                                 errors.ECODE_INVAL)
6284
    if self.lu.op.live is not None:
6285
      if self.lu.op.live:
6286
        self.lu.op.mode = constants.HT_MIGRATION_LIVE
6287
      else:
6288
        self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
6289
      # reset the 'live' parameter to None so that repeated
6290
      # invocations of CheckPrereq do not raise an exception
6291
      self.lu.op.live = None
6292
    elif self.lu.op.mode is None:
6293
      # read the default value from the hypervisor
6294
      i_hv = self.cfg.GetClusterInfo().FillHV(self.instance, skip_globals=False)
6295
      self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
6296

    
6297
    self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
6298

    
6299
  def _RunAllocator(self):
6300
    """Run the allocator based on input opcode.
6301

6302
    """
6303
    ial = IAllocator(self.cfg, self.rpc,
6304
                     mode=constants.IALLOCATOR_MODE_RELOC,
6305
                     name=self.instance_name,
6306
                     # TODO See why hail breaks with a single node below
6307
                     relocate_from=[self.instance.primary_node,
6308
                                    self.instance.primary_node],
6309
                     )
6310

    
6311
    ial.Run(self.lu.op.iallocator)
6312

    
6313
    if not ial.success:
6314
      raise errors.OpPrereqError("Can't compute nodes using"
6315
                                 " iallocator '%s': %s" %
6316
                                 (self.lu.op.iallocator, ial.info),
6317
                                 errors.ECODE_NORES)
6318
    if len(ial.result) != ial.required_nodes:
6319
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
6320
                                 " of nodes (%s), required %s" %
6321
                                 (self.lu.op.iallocator, len(ial.result),
6322
                                  ial.required_nodes), errors.ECODE_FAULT)
6323
    self.target_node = ial.result[0]
6324
    self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
6325
                 self.instance_name, self.lu.op.iallocator,
6326
                 utils.CommaJoin(ial.result))
6327

    
6328
  def _WaitUntilSync(self):
6329
    """Poll with custom rpc for disk sync.
6330

6331
    This uses our own step-based rpc call.
6332

6333
    """
6334
    self.feedback_fn("* wait until resync is done")
6335
    all_done = False
6336
    while not all_done:
6337
      all_done = True
6338
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
6339
                                            self.nodes_ip,
6340
                                            self.instance.disks)
6341
      min_percent = 100
6342
      for node, nres in result.items():
6343
        nres.Raise("Cannot resync disks on node %s" % node)
6344
        node_done, node_percent = nres.payload
6345
        all_done = all_done and node_done
6346
        if node_percent is not None:
6347
          min_percent = min(min_percent, node_percent)
6348
      if not all_done:
6349
        if min_percent < 100:
6350
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
6351
        time.sleep(2)
6352

    
6353
  def _EnsureSecondary(self, node):
6354
    """Demote a node to secondary.
6355

6356
    """
6357
    self.feedback_fn("* switching node %s to secondary mode" % node)
6358

    
6359
    for dev in self.instance.disks:
6360
      self.cfg.SetDiskID(dev, node)
6361

    
6362
    result = self.rpc.call_blockdev_close(node, self.instance.name,
6363
                                          self.instance.disks)
6364
    result.Raise("Cannot change disk to secondary on node %s" % node)
6365

    
6366
  def _GoStandalone(self):
6367
    """Disconnect from the network.
6368

6369
    """
6370
    self.feedback_fn("* changing into standalone mode")
6371
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
6372
                                               self.instance.disks)
6373
    for node, nres in result.items():
6374
      nres.Raise("Cannot disconnect disks node %s" % node)
6375

    
6376
  def _GoReconnect(self, multimaster):
6377
    """Reconnect to the network.
6378

6379
    """
6380
    if multimaster:
6381
      msg = "dual-master"
6382
    else:
6383
      msg = "single-master"
6384
    self.feedback_fn("* changing disks into %s mode" % msg)
6385
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
6386
                                           self.instance.disks,
6387
                                           self.instance.name, multimaster)
6388
    for node, nres in result.items():
6389
      nres.Raise("Cannot change disks config on node %s" % node)
6390

    
6391
  def _ExecCleanup(self):
6392
    """Try to cleanup after a failed migration.
6393

6394
    The cleanup is done by:
6395
      - check that the instance is running only on one node
6396
        (and update the config if needed)
6397
      - change disks on its secondary node to secondary
6398
      - wait until disks are fully synchronized
6399
      - disconnect from the network
6400
      - change disks into single-master mode
6401
      - wait again until disks are fully synchronized
6402

6403
    """
6404
    instance = self.instance
6405
    target_node = self.target_node
6406
    source_node = self.source_node
6407

    
6408
    # check running on only one node
6409
    self.feedback_fn("* checking where the instance actually runs"
6410
                     " (if this hangs, the hypervisor might be in"
6411
                     " a bad state)")
6412
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
6413
    for node, result in ins_l.items():
6414
      result.Raise("Can't contact node %s" % node)
6415

    
6416
    runningon_source = instance.name in ins_l[source_node].payload
6417
    runningon_target = instance.name in ins_l[target_node].payload
6418

    
6419
    if runningon_source and runningon_target:
6420
      raise errors.OpExecError("Instance seems to be running on two nodes,"
6421
                               " or the hypervisor is confused. You will have"
6422
                               " to ensure manually that it runs only on one"
6423
                               " and restart this operation.")
6424

    
6425
    if not (runningon_source or runningon_target):
6426
      raise errors.OpExecError("Instance does not seem to be running at all."
6427
                               " In this case, it's safer to repair by"
6428
                               " running 'gnt-instance stop' to ensure disk"
6429
                               " shutdown, and then restarting it.")
6430

    
6431
    if runningon_target:
6432
      # the migration has actually succeeded, we need to update the config
6433
      self.feedback_fn("* instance running on secondary node (%s),"
6434
                       " updating config" % target_node)
6435
      instance.primary_node = target_node
6436
      self.cfg.Update(instance, self.feedback_fn)
6437
      demoted_node = source_node
6438
    else:
6439
      self.feedback_fn("* instance confirmed to be running on its"
6440
                       " primary node (%s)" % source_node)
6441
      demoted_node = target_node
6442

    
6443
    if instance.disk_template in constants.DTS_INT_MIRROR:
6444
      self._EnsureSecondary(demoted_node)
6445
      try:
6446
        self._WaitUntilSync()
6447
      except errors.OpExecError:
6448
        # we ignore here errors, since if the device is standalone, it
6449
        # won't be able to sync
6450
        pass
6451
      self._GoStandalone()
6452
      self._GoReconnect(False)
6453
      self._WaitUntilSync()
6454

    
6455
    self.feedback_fn("* done")
6456

    
6457
  def _RevertDiskStatus(self):
6458
    """Try to revert the disk status after a failed migration.
6459

6460
    """
6461
    target_node = self.target_node
6462
    if self.instance.disk_template in constants.DTS_EXT_MIRROR:
6463
      return
6464

    
6465
    try:
6466
      self._EnsureSecondary(target_node)
6467
      self._GoStandalone()
6468
      self._GoReconnect(False)
6469
      self._WaitUntilSync()
6470
    except errors.OpExecError, err:
6471
      self.lu.LogWarning("Migration failed and I can't reconnect the"
6472
                         " drives: error '%s'\n"
6473
                         "Please look and recover the instance status" %
6474
                         str(err))
6475

    
6476
  def _AbortMigration(self):
6477
    """Call the hypervisor code to abort a started migration.
6478

6479
    """
6480
    instance = self.instance
6481
    target_node = self.target_node
6482
    migration_info = self.migration_info
6483

    
6484
    abort_result = self.rpc.call_finalize_migration(target_node,
6485
                                                    instance,
6486
                                                    migration_info,
6487
                                                    False)
6488
    abort_msg = abort_result.fail_msg
6489
    if abort_msg:
6490
      logging.error("Aborting migration failed on target node %s: %s",
6491
                    target_node, abort_msg)
6492
      # Don't raise an exception here, as we stil have to try to revert the
6493
      # disk status, even if this step failed.
6494

    
6495
  def _ExecMigration(self):
6496
    """Migrate an instance.
6497

6498
    The migrate is done by:
6499
      - change the disks into dual-master mode
6500
      - wait until disks are fully synchronized again
6501
      - migrate the instance
6502
      - change disks on the new secondary node (the old primary) to secondary
6503
      - wait until disks are fully synchronized
6504
      - change disks into single-master mode
6505

6506
    """
6507
    instance = self.instance
6508
    target_node = self.target_node
6509
    source_node = self.source_node
6510

    
6511
    self.feedback_fn("* checking disk consistency between source and target")
6512
    for dev in instance.disks:
6513
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
6514
        raise errors.OpExecError("Disk %s is degraded or not fully"
6515
                                 " synchronized on target node,"
6516
                                 " aborting migrate." % dev.iv_name)
6517

    
6518
    # First get the migration information from the remote node
6519
    result = self.rpc.call_migration_info(source_node, instance)
6520
    msg = result.fail_msg
6521
    if msg:
6522
      log_err = ("Failed fetching source migration information from %s: %s" %
6523
                 (source_node, msg))
6524
      logging.error(log_err)
6525
      raise errors.OpExecError(log_err)
6526

    
6527
    self.migration_info = migration_info = result.payload
6528

    
6529
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
6530
      # Then switch the disks to master/master mode
6531
      self._EnsureSecondary(target_node)
6532
      self._GoStandalone()
6533
      self._GoReconnect(True)
6534
      self._WaitUntilSync()
6535

    
6536
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
6537
    result = self.rpc.call_accept_instance(target_node,
6538
                                           instance,
6539
                                           migration_info,
6540
                                           self.nodes_ip[target_node])
6541

    
6542
    msg = result.fail_msg
6543
    if msg:
6544
      logging.error("Instance pre-migration failed, trying to revert"
6545
                    " disk status: %s", msg)
6546
      self.feedback_fn("Pre-migration failed, aborting")
6547
      self._AbortMigration()
6548
      self._RevertDiskStatus()
6549
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
6550
                               (instance.name, msg))
6551

    
6552
    self.feedback_fn("* migrating instance to %s" % target_node)
6553
    result = self.rpc.call_instance_migrate(source_node, instance,
6554
                                            self.nodes_ip[target_node],
6555
                                            self.live)
6556
    msg = result.fail_msg
6557
    if msg:
6558
      logging.error("Instance migration failed, trying to revert"
6559
                    " disk status: %s", msg)
6560
      self.feedback_fn("Migration failed, aborting")
6561
      self._AbortMigration()
6562
      self._RevertDiskStatus()
6563
      raise errors.OpExecError("Could not migrate instance %s: %s" %
6564
                               (instance.name, msg))
6565

    
6566
    instance.primary_node = target_node
6567
    # distribute new instance config to the other nodes
6568
    self.cfg.Update(instance, self.feedback_fn)
6569

    
6570
    result = self.rpc.call_finalize_migration(target_node,
6571
                                              instance,
6572
                                              migration_info,
6573
                                              True)
6574
    msg = result.fail_msg
6575
    if msg:
6576
      logging.error("Instance migration succeeded, but finalization failed:"
6577
                    " %s", msg)
6578
      raise errors.OpExecError("Could not finalize instance migration: %s" %
6579
                               msg)
6580

    
6581
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
6582
      self._EnsureSecondary(source_node)
6583
      self._WaitUntilSync()
6584
      self._GoStandalone()
6585
      self._GoReconnect(False)
6586
      self._WaitUntilSync()
6587

    
6588
    self.feedback_fn("* done")
6589

    
6590
  def Exec(self, feedback_fn):
6591
    """Perform the migration.
6592

6593
    """
6594
    feedback_fn("Migrating instance %s" % self.instance.name)
6595

    
6596
    self.feedback_fn = feedback_fn
6597

    
6598
    self.source_node = self.instance.primary_node
6599

    
6600
    # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
6601
    if self.instance.disk_template in constants.DTS_INT_MIRROR:
6602
      self.target_node = self.instance.secondary_nodes[0]
6603
      # Otherwise self.target_node has been populated either
6604
      # directly, or through an iallocator.
6605

    
6606
    self.all_nodes = [self.source_node, self.target_node]
6607
    self.nodes_ip = {
6608
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
6609
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
6610
      }
6611

    
6612
    if self.cleanup:
6613
      return self._ExecCleanup()
6614
    else:
6615
      return self._ExecMigration()
6616

    
6617

    
6618
def _CreateBlockDev(lu, node, instance, device, force_create,
6619
                    info, force_open):
6620
  """Create a tree of block devices on a given node.
6621

6622
  If this device type has to be created on secondaries, create it and
6623
  all its children.
6624

6625
  If not, just recurse to children keeping the same 'force' value.
6626

6627
  @param lu: the lu on whose behalf we execute
6628
  @param node: the node on which to create the device
6629
  @type instance: L{objects.Instance}
6630
  @param instance: the instance which owns the device
6631
  @type device: L{objects.Disk}
6632
  @param device: the device to create
6633
  @type force_create: boolean
6634
  @param force_create: whether to force creation of this device; this
6635
      will be change to True whenever we find a device which has
6636
      CreateOnSecondary() attribute
6637
  @param info: the extra 'metadata' we should attach to the device
6638
      (this will be represented as a LVM tag)
6639
  @type force_open: boolean
6640
  @param force_open: this parameter will be passes to the
6641
      L{backend.BlockdevCreate} function where it specifies
6642
      whether we run on primary or not, and it affects both
6643
      the child assembly and the device own Open() execution
6644

6645
  """
6646
  if device.CreateOnSecondary():
6647
    force_create = True
6648

    
6649
  if device.children:
6650
    for child in device.children:
6651
      _CreateBlockDev(lu, node, instance, child, force_create,
6652
                      info, force_open)
6653

    
6654
  if not force_create:
6655
    return
6656

    
6657
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
6658

    
6659

    
6660
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
6661
  """Create a single block device on a given node.
6662

6663
  This will not recurse over children of the device, so they must be
6664
  created in advance.
6665

6666
  @param lu: the lu on whose behalf we execute
6667
  @param node: the node on which to create the device
6668
  @type instance: L{objects.Instance}
6669
  @param instance: the instance which owns the device
6670
  @type device: L{objects.Disk}
6671
  @param device: the device to create
6672
  @param info: the extra 'metadata' we should attach to the device
6673
      (this will be represented as a LVM tag)
6674
  @type force_open: boolean
6675
  @param force_open: this parameter will be passes to the
6676
      L{backend.BlockdevCreate} function where it specifies
6677
      whether we run on primary or not, and it affects both
6678
      the child assembly and the device own Open() execution
6679

6680
  """
6681
  lu.cfg.SetDiskID(device, node)
6682
  result = lu.rpc.call_blockdev_create(node, device, device.size,
6683
                                       instance.name, force_open, info)
6684
  result.Raise("Can't create block device %s on"
6685
               " node %s for instance %s" % (device, node, instance.name))
6686
  if device.physical_id is None:
6687
    device.physical_id = result.payload
6688

    
6689

    
6690
def _GenerateUniqueNames(lu, exts):
6691
  """Generate a suitable LV name.
6692

6693
  This will generate a logical volume name for the given instance.
6694

6695
  """
6696
  results = []
6697
  for val in exts:
6698
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
6699
    results.append("%s%s" % (new_id, val))
6700
  return results
6701

    
6702

    
6703
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgname, names, iv_name,
6704
                         p_minor, s_minor):
6705
  """Generate a drbd8 device complete with its children.
6706

6707
  """
6708
  port = lu.cfg.AllocatePort()
6709
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
6710
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
6711
                          logical_id=(vgname, names[0]))
6712
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
6713
                          logical_id=(vgname, names[1]))
6714
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
6715
                          logical_id=(primary, secondary, port,
6716
                                      p_minor, s_minor,
6717
                                      shared_secret),
6718
                          children=[dev_data, dev_meta],
6719
                          iv_name=iv_name)
6720
  return drbd_dev
6721

    
6722

    
6723
def _GenerateDiskTemplate(lu, template_name,
6724
                          instance_name, primary_node,
6725
                          secondary_nodes, disk_info,
6726
                          file_storage_dir, file_driver,
6727
                          base_index, feedback_fn):
6728
  """Generate the entire disk layout for a given template type.
6729

6730
  """
6731
  #TODO: compute space requirements
6732

    
6733
  vgname = lu.cfg.GetVGName()
6734
  disk_count = len(disk_info)
6735
  disks = []
6736
  if template_name == constants.DT_DISKLESS:
6737
    pass
6738
  elif template_name == constants.DT_PLAIN:
6739
    if len(secondary_nodes) != 0:
6740
      raise errors.ProgrammerError("Wrong template configuration")
6741

    
6742
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6743
                                      for i in range(disk_count)])
6744
    for idx, disk in enumerate(disk_info):
6745
      disk_index = idx + base_index
6746
      vg = disk.get("vg", vgname)
6747
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
6748
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
6749
                              logical_id=(vg, names[idx]),
6750
                              iv_name="disk/%d" % disk_index,
6751
                              mode=disk["mode"])
6752
      disks.append(disk_dev)
6753
  elif template_name == constants.DT_DRBD8:
6754
    if len(secondary_nodes) != 1:
6755
      raise errors.ProgrammerError("Wrong template configuration")
6756
    remote_node = secondary_nodes[0]
6757
    minors = lu.cfg.AllocateDRBDMinor(
6758
      [primary_node, remote_node] * len(disk_info), instance_name)
6759

    
6760
    names = []
6761
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6762
                                               for i in range(disk_count)]):
6763
      names.append(lv_prefix + "_data")
6764
      names.append(lv_prefix + "_meta")
6765
    for idx, disk in enumerate(disk_info):
6766
      disk_index = idx + base_index
6767
      vg = disk.get("vg", vgname)
6768
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
6769
                                      disk["size"], vg, names[idx*2:idx*2+2],
6770
                                      "disk/%d" % disk_index,
6771
                                      minors[idx*2], minors[idx*2+1])
6772
      disk_dev.mode = disk["mode"]
6773
      disks.append(disk_dev)
6774
  elif template_name == constants.DT_FILE:
6775
    if len(secondary_nodes) != 0:
6776
      raise errors.ProgrammerError("Wrong template configuration")
6777

    
6778
    opcodes.RequireFileStorage()
6779

    
6780
    for idx, disk in enumerate(disk_info):
6781
      disk_index = idx + base_index
6782
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6783
                              iv_name="disk/%d" % disk_index,
6784
                              logical_id=(file_driver,
6785
                                          "%s/disk%d" % (file_storage_dir,
6786
                                                         disk_index)),
6787
                              mode=disk["mode"])
6788
      disks.append(disk_dev)
6789
  elif template_name == constants.DT_SHARED_FILE:
6790
    if len(secondary_nodes) != 0:
6791
      raise errors.ProgrammerError("Wrong template configuration")
6792

    
6793
    opcodes.RequireSharedFileStorage()
6794

    
6795
    for idx, disk in enumerate(disk_info):
6796
      disk_index = idx + base_index
6797
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6798
                              iv_name="disk/%d" % disk_index,
6799
                              logical_id=(file_driver,
6800
                                          "%s/disk%d" % (file_storage_dir,
6801
                                                         disk_index)),
6802
                              mode=disk["mode"])
6803
      disks.append(disk_dev)
6804
  elif template_name == constants.DT_BLOCK:
6805
    if len(secondary_nodes) != 0:
6806
      raise errors.ProgrammerError("Wrong template configuration")
6807

    
6808
    for idx, disk in enumerate(disk_info):
6809
      disk_index = idx + base_index
6810
      disk_dev = objects.Disk(dev_type=constants.LD_BLOCKDEV, size=disk["size"],
6811
                              logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
6812
                                          disk["adopt"]),
6813
                              iv_name="disk/%d" % disk_index,
6814
                              mode=disk["mode"])
6815
      disks.append(disk_dev)
6816

    
6817
  else:
6818
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
6819
  return disks
6820

    
6821

    
6822
def _GetInstanceInfoText(instance):
6823
  """Compute that text that should be added to the disk's metadata.
6824

6825
  """
6826
  return "originstname+%s" % instance.name
6827

    
6828

    
6829
def _CalcEta(time_taken, written, total_size):
6830
  """Calculates the ETA based on size written and total size.
6831

6832
  @param time_taken: The time taken so far
6833
  @param written: amount written so far
6834
  @param total_size: The total size of data to be written
6835
  @return: The remaining time in seconds
6836

6837
  """
6838
  avg_time = time_taken / float(written)
6839
  return (total_size - written) * avg_time
6840

    
6841

    
6842
def _WipeDisks(lu, instance):
6843
  """Wipes instance disks.
6844

6845
  @type lu: L{LogicalUnit}
6846
  @param lu: the logical unit on whose behalf we execute
6847
  @type instance: L{objects.Instance}
6848
  @param instance: the instance whose disks we should create
6849
  @return: the success of the wipe
6850

6851
  """
6852
  node = instance.primary_node
6853

    
6854
  for device in instance.disks:
6855
    lu.cfg.SetDiskID(device, node)
6856

    
6857
  logging.info("Pause sync of instance %s disks", instance.name)
6858
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
6859

    
6860
  for idx, success in enumerate(result.payload):
6861
    if not success:
6862
      logging.warn("pause-sync of instance %s for disks %d failed",
6863
                   instance.name, idx)
6864

    
6865
  try:
6866
    for idx, device in enumerate(instance.disks):
6867
      lu.LogInfo("* Wiping disk %d", idx)
6868
      logging.info("Wiping disk %d for instance %s, node %s",
6869
                   idx, instance.name, node)
6870

    
6871
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
6872
      # MAX_WIPE_CHUNK at max
6873
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
6874
                            constants.MIN_WIPE_CHUNK_PERCENT)
6875

    
6876
      offset = 0
6877
      size = device.size
6878
      last_output = 0
6879
      start_time = time.time()
6880

    
6881
      while offset < size:
6882
        wipe_size = min(wipe_chunk_size, size - offset)
6883
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
6884
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
6885
                     (idx, offset, wipe_size))
6886
        now = time.time()
6887
        offset += wipe_size
6888
        if now - last_output >= 60:
6889
          eta = _CalcEta(now - start_time, offset, size)
6890
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
6891
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
6892
          last_output = now
6893
  finally:
6894
    logging.info("Resume sync of instance %s disks", instance.name)
6895

    
6896
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
6897

    
6898
    for idx, success in enumerate(result.payload):
6899
      if not success:
6900
        lu.LogWarning("Warning: Resume sync of disk %d failed. Please have a"
6901
                      " look at the status and troubleshoot the issue.", idx)
6902
        logging.warn("resume-sync of instance %s for disks %d failed",
6903
                     instance.name, idx)
6904

    
6905

    
6906
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
6907
  """Create all disks for an instance.
6908

6909
  This abstracts away some work from AddInstance.
6910

6911
  @type lu: L{LogicalUnit}
6912
  @param lu: the logical unit on whose behalf we execute
6913
  @type instance: L{objects.Instance}
6914
  @param instance: the instance whose disks we should create
6915
  @type to_skip: list
6916
  @param to_skip: list of indices to skip
6917
  @type target_node: string
6918
  @param target_node: if passed, overrides the target node for creation
6919
  @rtype: boolean
6920
  @return: the success of the creation
6921

6922
  """
6923
  info = _GetInstanceInfoText(instance)
6924
  if target_node is None:
6925
    pnode = instance.primary_node
6926
    all_nodes = instance.all_nodes
6927
  else:
6928
    pnode = target_node
6929
    all_nodes = [pnode]
6930

    
6931
  if instance.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE):
6932
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6933
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
6934

    
6935
    result.Raise("Failed to create directory '%s' on"
6936
                 " node %s" % (file_storage_dir, pnode))
6937

    
6938
  # Note: this needs to be kept in sync with adding of disks in
6939
  # LUInstanceSetParams
6940
  for idx, device in enumerate(instance.disks):
6941
    if to_skip and idx in to_skip:
6942
      continue
6943
    logging.info("Creating volume %s for instance %s",
6944
                 device.iv_name, instance.name)
6945
    #HARDCODE
6946
    for node in all_nodes:
6947
      f_create = node == pnode
6948
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
6949

    
6950

    
6951
def _RemoveDisks(lu, instance, target_node=None):
6952
  """Remove all disks for an instance.
6953

6954
  This abstracts away some work from `AddInstance()` and
6955
  `RemoveInstance()`. Note that in case some of the devices couldn't
6956
  be removed, the removal will continue with the other ones (compare
6957
  with `_CreateDisks()`).
6958

6959
  @type lu: L{LogicalUnit}
6960
  @param lu: the logical unit on whose behalf we execute
6961
  @type instance: L{objects.Instance}
6962
  @param instance: the instance whose disks we should remove
6963
  @type target_node: string
6964
  @param target_node: used to override the node on which to remove the disks
6965
  @rtype: boolean
6966
  @return: the success of the removal
6967

6968
  """
6969
  logging.info("Removing block devices for instance %s", instance.name)
6970

    
6971
  all_result = True
6972
  for device in instance.disks:
6973
    if target_node:
6974
      edata = [(target_node, device)]
6975
    else:
6976
      edata = device.ComputeNodeTree(instance.primary_node)
6977
    for node, disk in edata:
6978
      lu.cfg.SetDiskID(disk, node)
6979
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
6980
      if msg:
6981
        lu.LogWarning("Could not remove block device %s on node %s,"
6982
                      " continuing anyway: %s", device.iv_name, node, msg)
6983
        all_result = False
6984

    
6985
  if instance.disk_template == constants.DT_FILE:
6986
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6987
    if target_node:
6988
      tgt = target_node
6989
    else:
6990
      tgt = instance.primary_node
6991
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
6992
    if result.fail_msg:
6993
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
6994
                    file_storage_dir, instance.primary_node, result.fail_msg)
6995
      all_result = False
6996

    
6997
  return all_result
6998

    
6999

    
7000
def _ComputeDiskSizePerVG(disk_template, disks):
7001
  """Compute disk size requirements in the volume group
7002

7003
  """
7004
  def _compute(disks, payload):
7005
    """Universal algorithm
7006

7007
    """
7008
    vgs = {}
7009
    for disk in disks:
7010
      vgs[disk["vg"]] = vgs.get("vg", 0) + disk["size"] + payload
7011

    
7012
    return vgs
7013

    
7014
  # Required free disk space as a function of disk and swap space
7015
  req_size_dict = {
7016
    constants.DT_DISKLESS: {},
7017
    constants.DT_PLAIN: _compute(disks, 0),
7018
    # 128 MB are added for drbd metadata for each disk
7019
    constants.DT_DRBD8: _compute(disks, 128),
7020
    constants.DT_FILE: {},
7021
    constants.DT_SHARED_FILE: {},
7022
  }
7023

    
7024
  if disk_template not in req_size_dict:
7025
    raise errors.ProgrammerError("Disk template '%s' size requirement"
7026
                                 " is unknown" %  disk_template)
7027

    
7028
  return req_size_dict[disk_template]
7029

    
7030

    
7031
def _ComputeDiskSize(disk_template, disks):
7032
  """Compute disk size requirements in the volume group
7033

7034
  """
7035
  # Required free disk space as a function of disk and swap space
7036
  req_size_dict = {
7037
    constants.DT_DISKLESS: None,
7038
    constants.DT_PLAIN: sum(d["size"] for d in disks),
7039
    # 128 MB are added for drbd metadata for each disk
7040
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
7041
    constants.DT_FILE: None,
7042
    constants.DT_SHARED_FILE: 0,
7043
    constants.DT_BLOCK: 0,
7044
  }
7045

    
7046
  if disk_template not in req_size_dict:
7047
    raise errors.ProgrammerError("Disk template '%s' size requirement"
7048
                                 " is unknown" %  disk_template)
7049

    
7050
  return req_size_dict[disk_template]
7051

    
7052

    
7053
def _FilterVmNodes(lu, nodenames):
7054
  """Filters out non-vm_capable nodes from a list.
7055

7056
  @type lu: L{LogicalUnit}
7057
  @param lu: the logical unit for which we check
7058
  @type nodenames: list
7059
  @param nodenames: the list of nodes on which we should check
7060
  @rtype: list
7061
  @return: the list of vm-capable nodes
7062

7063
  """
7064
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
7065
  return [name for name in nodenames if name not in vm_nodes]
7066

    
7067

    
7068
def _CheckHVParams(lu, nodenames, hvname, hvparams):
7069
  """Hypervisor parameter validation.
7070

7071
  This function abstract the hypervisor parameter validation to be
7072
  used in both instance create and instance modify.
7073

7074
  @type lu: L{LogicalUnit}
7075
  @param lu: the logical unit for which we check
7076
  @type nodenames: list
7077
  @param nodenames: the list of nodes on which we should check
7078
  @type hvname: string
7079
  @param hvname: the name of the hypervisor we should use
7080
  @type hvparams: dict
7081
  @param hvparams: the parameters which we need to check
7082
  @raise errors.OpPrereqError: if the parameters are not valid
7083

7084
  """
7085
  nodenames = _FilterVmNodes(lu, nodenames)
7086
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
7087
                                                  hvname,
7088
                                                  hvparams)
7089
  for node in nodenames:
7090
    info = hvinfo[node]
7091
    if info.offline:
7092
      continue
7093
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
7094

    
7095

    
7096
def _CheckOSParams(lu, required, nodenames, osname, osparams):
7097
  """OS parameters validation.
7098

7099
  @type lu: L{LogicalUnit}
7100
  @param lu: the logical unit for which we check
7101
  @type required: boolean
7102
  @param required: whether the validation should fail if the OS is not
7103
      found
7104
  @type nodenames: list
7105
  @param nodenames: the list of nodes on which we should check
7106
  @type osname: string
7107
  @param osname: the name of the hypervisor we should use
7108
  @type osparams: dict
7109
  @param osparams: the parameters which we need to check
7110
  @raise errors.OpPrereqError: if the parameters are not valid
7111

7112
  """
7113
  nodenames = _FilterVmNodes(lu, nodenames)
7114
  result = lu.rpc.call_os_validate(required, nodenames, osname,
7115
                                   [constants.OS_VALIDATE_PARAMETERS],
7116
                                   osparams)
7117
  for node, nres in result.items():
7118
    # we don't check for offline cases since this should be run only
7119
    # against the master node and/or an instance's nodes
7120
    nres.Raise("OS Parameters validation failed on node %s" % node)
7121
    if not nres.payload:
7122
      lu.LogInfo("OS %s not found on node %s, validation skipped",
7123
                 osname, node)
7124

    
7125

    
7126
class LUInstanceCreate(LogicalUnit):
7127
  """Create an instance.
7128

7129
  """
7130
  HPATH = "instance-add"
7131
  HTYPE = constants.HTYPE_INSTANCE
7132
  REQ_BGL = False
7133

    
7134
  def CheckArguments(self):
7135
    """Check arguments.
7136

7137
    """
7138
    # do not require name_check to ease forward/backward compatibility
7139
    # for tools
7140
    if self.op.no_install and self.op.start:
7141
      self.LogInfo("No-installation mode selected, disabling startup")
7142
      self.op.start = False
7143
    # validate/normalize the instance name
7144
    self.op.instance_name = \
7145
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
7146

    
7147
    if self.op.ip_check and not self.op.name_check:
7148
      # TODO: make the ip check more flexible and not depend on the name check
7149
      raise errors.OpPrereqError("Cannot do ip check without a name check",
7150
                                 errors.ECODE_INVAL)
7151

    
7152
    # check nics' parameter names
7153
    for nic in self.op.nics:
7154
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
7155

    
7156
    # check disks. parameter names and consistent adopt/no-adopt strategy
7157
    has_adopt = has_no_adopt = False
7158
    for disk in self.op.disks:
7159
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
7160
      if "adopt" in disk:
7161
        has_adopt = True
7162
      else:
7163
        has_no_adopt = True
7164
    if has_adopt and has_no_adopt:
7165
      raise errors.OpPrereqError("Either all disks are adopted or none is",
7166
                                 errors.ECODE_INVAL)
7167
    if has_adopt:
7168
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
7169
        raise errors.OpPrereqError("Disk adoption is not supported for the"
7170
                                   " '%s' disk template" %
7171
                                   self.op.disk_template,
7172
                                   errors.ECODE_INVAL)
7173
      if self.op.iallocator is not None:
7174
        raise errors.OpPrereqError("Disk adoption not allowed with an"
7175
                                   " iallocator script", errors.ECODE_INVAL)
7176
      if self.op.mode == constants.INSTANCE_IMPORT:
7177
        raise errors.OpPrereqError("Disk adoption not allowed for"
7178
                                   " instance import", errors.ECODE_INVAL)
7179
    else:
7180
      if self.op.disk_template in constants.DTS_MUST_ADOPT:
7181
        raise errors.OpPrereqError("Disk template %s requires disk adoption,"
7182
                                   " but no 'adopt' parameter given" %
7183
                                   self.op.disk_template,
7184
                                   errors.ECODE_INVAL)
7185

    
7186
    self.adopt_disks = has_adopt
7187

    
7188
    # instance name verification
7189
    if self.op.name_check:
7190
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
7191
      self.op.instance_name = self.hostname1.name
7192
      # used in CheckPrereq for ip ping check
7193
      self.check_ip = self.hostname1.ip
7194
    else:
7195
      self.check_ip = None
7196

    
7197
    # file storage checks
7198
    if (self.op.file_driver and
7199
        not self.op.file_driver in constants.FILE_DRIVER):
7200
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
7201
                                 self.op.file_driver, errors.ECODE_INVAL)
7202

    
7203
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
7204
      raise errors.OpPrereqError("File storage directory path not absolute",
7205
                                 errors.ECODE_INVAL)
7206

    
7207
    ### Node/iallocator related checks
7208
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
7209

    
7210
    if self.op.pnode is not None:
7211
      if self.op.disk_template in constants.DTS_INT_MIRROR:
7212
        if self.op.snode is None:
7213
          raise errors.OpPrereqError("The networked disk templates need"
7214
                                     " a mirror node", errors.ECODE_INVAL)
7215
      elif self.op.snode:
7216
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
7217
                        " template")
7218
        self.op.snode = None
7219

    
7220
    self._cds = _GetClusterDomainSecret()
7221

    
7222
    if self.op.mode == constants.INSTANCE_IMPORT:
7223
      # On import force_variant must be True, because if we forced it at
7224
      # initial install, our only chance when importing it back is that it
7225
      # works again!
7226
      self.op.force_variant = True
7227

    
7228
      if self.op.no_install:
7229
        self.LogInfo("No-installation mode has no effect during import")
7230

    
7231
    elif self.op.mode == constants.INSTANCE_CREATE:
7232
      if self.op.os_type is None:
7233
        raise errors.OpPrereqError("No guest OS specified",
7234
                                   errors.ECODE_INVAL)
7235
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
7236
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
7237
                                   " installation" % self.op.os_type,
7238
                                   errors.ECODE_STATE)
7239
      if self.op.disk_template is None:
7240
        raise errors.OpPrereqError("No disk template specified",
7241
                                   errors.ECODE_INVAL)
7242

    
7243
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7244
      # Check handshake to ensure both clusters have the same domain secret
7245
      src_handshake = self.op.source_handshake
7246
      if not src_handshake:
7247
        raise errors.OpPrereqError("Missing source handshake",
7248
                                   errors.ECODE_INVAL)
7249

    
7250
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
7251
                                                           src_handshake)
7252
      if errmsg:
7253
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
7254
                                   errors.ECODE_INVAL)
7255

    
7256
      # Load and check source CA
7257
      self.source_x509_ca_pem = self.op.source_x509_ca
7258
      if not self.source_x509_ca_pem:
7259
        raise errors.OpPrereqError("Missing source X509 CA",
7260
                                   errors.ECODE_INVAL)
7261

    
7262
      try:
7263
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
7264
                                                    self._cds)
7265
      except OpenSSL.crypto.Error, err:
7266
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
7267
                                   (err, ), errors.ECODE_INVAL)
7268

    
7269
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
7270
      if errcode is not None:
7271
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
7272
                                   errors.ECODE_INVAL)
7273

    
7274
      self.source_x509_ca = cert
7275

    
7276
      src_instance_name = self.op.source_instance_name
7277
      if not src_instance_name:
7278
        raise errors.OpPrereqError("Missing source instance name",
7279
                                   errors.ECODE_INVAL)
7280

    
7281
      self.source_instance_name = \
7282
          netutils.GetHostname(name=src_instance_name).name
7283

    
7284
    else:
7285
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
7286
                                 self.op.mode, errors.ECODE_INVAL)
7287

    
7288
  def ExpandNames(self):
7289
    """ExpandNames for CreateInstance.
7290

7291
    Figure out the right locks for instance creation.
7292

7293
    """
7294
    self.needed_locks = {}
7295

    
7296
    instance_name = self.op.instance_name
7297
    # this is just a preventive check, but someone might still add this
7298
    # instance in the meantime, and creation will fail at lock-add time
7299
    if instance_name in self.cfg.GetInstanceList():
7300
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7301
                                 instance_name, errors.ECODE_EXISTS)
7302

    
7303
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
7304

    
7305
    if self.op.iallocator:
7306
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7307
    else:
7308
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
7309
      nodelist = [self.op.pnode]
7310
      if self.op.snode is not None:
7311
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
7312
        nodelist.append(self.op.snode)
7313
      self.needed_locks[locking.LEVEL_NODE] = nodelist
7314

    
7315
    # in case of import lock the source node too
7316
    if self.op.mode == constants.INSTANCE_IMPORT:
7317
      src_node = self.op.src_node
7318
      src_path = self.op.src_path
7319

    
7320
      if src_path is None:
7321
        self.op.src_path = src_path = self.op.instance_name
7322

    
7323
      if src_node is None:
7324
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7325
        self.op.src_node = None
7326
        if os.path.isabs(src_path):
7327
          raise errors.OpPrereqError("Importing an instance from an absolute"
7328
                                     " path requires a source node option.",
7329
                                     errors.ECODE_INVAL)
7330
      else:
7331
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
7332
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
7333
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
7334
        if not os.path.isabs(src_path):
7335
          self.op.src_path = src_path = \
7336
            utils.PathJoin(constants.EXPORT_DIR, src_path)
7337

    
7338
  def _RunAllocator(self):
7339
    """Run the allocator based on input opcode.
7340

7341
    """
7342
    nics = [n.ToDict() for n in self.nics]
7343
    ial = IAllocator(self.cfg, self.rpc,
7344
                     mode=constants.IALLOCATOR_MODE_ALLOC,
7345
                     name=self.op.instance_name,
7346
                     disk_template=self.op.disk_template,
7347
                     tags=[],
7348
                     os=self.op.os_type,
7349
                     vcpus=self.be_full[constants.BE_VCPUS],
7350
                     mem_size=self.be_full[constants.BE_MEMORY],
7351
                     disks=self.disks,
7352
                     nics=nics,
7353
                     hypervisor=self.op.hypervisor,
7354
                     )
7355

    
7356
    ial.Run(self.op.iallocator)
7357

    
7358
    if not ial.success:
7359
      raise errors.OpPrereqError("Can't compute nodes using"
7360
                                 " iallocator '%s': %s" %
7361
                                 (self.op.iallocator, ial.info),
7362
                                 errors.ECODE_NORES)
7363
    if len(ial.result) != ial.required_nodes:
7364
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
7365
                                 " of nodes (%s), required %s" %
7366
                                 (self.op.iallocator, len(ial.result),
7367
                                  ial.required_nodes), errors.ECODE_FAULT)
7368
    self.op.pnode = ial.result[0]
7369
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
7370
                 self.op.instance_name, self.op.iallocator,
7371
                 utils.CommaJoin(ial.result))
7372
    if ial.required_nodes == 2:
7373
      self.op.snode = ial.result[1]
7374

    
7375
  def BuildHooksEnv(self):
7376
    """Build hooks env.
7377

7378
    This runs on master, primary and secondary nodes of the instance.
7379

7380
    """
7381
    env = {
7382
      "ADD_MODE": self.op.mode,
7383
      }
7384
    if self.op.mode == constants.INSTANCE_IMPORT:
7385
      env["SRC_NODE"] = self.op.src_node
7386
      env["SRC_PATH"] = self.op.src_path
7387
      env["SRC_IMAGES"] = self.src_images
7388

    
7389
    env.update(_BuildInstanceHookEnv(
7390
      name=self.op.instance_name,
7391
      primary_node=self.op.pnode,
7392
      secondary_nodes=self.secondaries,
7393
      status=self.op.start,
7394
      os_type=self.op.os_type,
7395
      memory=self.be_full[constants.BE_MEMORY],
7396
      vcpus=self.be_full[constants.BE_VCPUS],
7397
      nics=_NICListToTuple(self, self.nics),
7398
      disk_template=self.op.disk_template,
7399
      disks=[(d["size"], d["mode"]) for d in self.disks],
7400
      bep=self.be_full,
7401
      hvp=self.hv_full,
7402
      hypervisor_name=self.op.hypervisor,
7403
    ))
7404

    
7405
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
7406
          self.secondaries)
7407
    return env, nl, nl
7408

    
7409
  def _ReadExportInfo(self):
7410
    """Reads the export information from disk.
7411

7412
    It will override the opcode source node and path with the actual
7413
    information, if these two were not specified before.
7414

7415
    @return: the export information
7416

7417
    """
7418
    assert self.op.mode == constants.INSTANCE_IMPORT
7419

    
7420
    src_node = self.op.src_node
7421
    src_path = self.op.src_path
7422

    
7423
    if src_node is None:
7424
      locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
7425
      exp_list = self.rpc.call_export_list(locked_nodes)
7426
      found = False
7427
      for node in exp_list:
7428
        if exp_list[node].fail_msg:
7429
          continue
7430
        if src_path in exp_list[node].payload:
7431
          found = True
7432
          self.op.src_node = src_node = node
7433
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
7434
                                                       src_path)
7435
          break
7436
      if not found:
7437
        raise errors.OpPrereqError("No export found for relative path %s" %
7438
                                    src_path, errors.ECODE_INVAL)
7439

    
7440
    _CheckNodeOnline(self, src_node)
7441
    result = self.rpc.call_export_info(src_node, src_path)
7442
    result.Raise("No export or invalid export found in dir %s" % src_path)
7443

    
7444
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
7445
    if not export_info.has_section(constants.INISECT_EXP):
7446
      raise errors.ProgrammerError("Corrupted export config",
7447
                                   errors.ECODE_ENVIRON)
7448

    
7449
    ei_version = export_info.get(constants.INISECT_EXP, "version")
7450
    if (int(ei_version) != constants.EXPORT_VERSION):
7451
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
7452
                                 (ei_version, constants.EXPORT_VERSION),
7453
                                 errors.ECODE_ENVIRON)
7454
    return export_info
7455

    
7456
  def _ReadExportParams(self, einfo):
7457
    """Use export parameters as defaults.
7458

7459
    In case the opcode doesn't specify (as in override) some instance
7460
    parameters, then try to use them from the export information, if
7461
    that declares them.
7462

7463
    """
7464
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
7465

    
7466
    if self.op.disk_template is None:
7467
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
7468
        self.op.disk_template = einfo.get(constants.INISECT_INS,
7469
                                          "disk_template")
7470
      else:
7471
        raise errors.OpPrereqError("No disk template specified and the export"
7472
                                   " is missing the disk_template information",
7473
                                   errors.ECODE_INVAL)
7474

    
7475
    if not self.op.disks:
7476
      if einfo.has_option(constants.INISECT_INS, "disk_count"):
7477
        disks = []
7478
        # TODO: import the disk iv_name too
7479
        for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
7480
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
7481
          disks.append({"size": disk_sz})
7482
        self.op.disks = disks
7483
      else:
7484
        raise errors.OpPrereqError("No disk info specified and the export"
7485
                                   " is missing the disk information",
7486
                                   errors.ECODE_INVAL)
7487

    
7488
    if (not self.op.nics and
7489
        einfo.has_option(constants.INISECT_INS, "nic_count")):
7490
      nics = []
7491
      for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
7492
        ndict = {}
7493
        for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
7494
          v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
7495
          ndict[name] = v
7496
        nics.append(ndict)
7497
      self.op.nics = nics
7498

    
7499
    if (self.op.hypervisor is None and
7500
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
7501
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
7502
    if einfo.has_section(constants.INISECT_HYP):
7503
      # use the export parameters but do not override the ones
7504
      # specified by the user
7505
      for name, value in einfo.items(constants.INISECT_HYP):
7506
        if name not in self.op.hvparams:
7507
          self.op.hvparams[name] = value
7508

    
7509
    if einfo.has_section(constants.INISECT_BEP):
7510
      # use the parameters, without overriding
7511
      for name, value in einfo.items(constants.INISECT_BEP):
7512
        if name not in self.op.beparams:
7513
          self.op.beparams[name] = value
7514
    else:
7515
      # try to read the parameters old style, from the main section
7516
      for name in constants.BES_PARAMETERS:
7517
        if (name not in self.op.beparams and
7518
            einfo.has_option(constants.INISECT_INS, name)):
7519
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
7520

    
7521
    if einfo.has_section(constants.INISECT_OSP):
7522
      # use the parameters, without overriding
7523
      for name, value in einfo.items(constants.INISECT_OSP):
7524
        if name not in self.op.osparams:
7525
          self.op.osparams[name] = value
7526

    
7527
  def _RevertToDefaults(self, cluster):
7528
    """Revert the instance parameters to the default values.
7529

7530
    """
7531
    # hvparams
7532
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
7533
    for name in self.op.hvparams.keys():
7534
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
7535
        del self.op.hvparams[name]
7536
    # beparams
7537
    be_defs = cluster.SimpleFillBE({})
7538
    for name in self.op.beparams.keys():
7539
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
7540
        del self.op.beparams[name]
7541
    # nic params
7542
    nic_defs = cluster.SimpleFillNIC({})
7543
    for nic in self.op.nics:
7544
      for name in constants.NICS_PARAMETERS:
7545
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
7546
          del nic[name]
7547
    # osparams
7548
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
7549
    for name in self.op.osparams.keys():
7550
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
7551
        del self.op.osparams[name]
7552

    
7553
  def CheckPrereq(self):
7554
    """Check prerequisites.
7555

7556
    """
7557
    if self.op.mode == constants.INSTANCE_IMPORT:
7558
      export_info = self._ReadExportInfo()
7559
      self._ReadExportParams(export_info)
7560

    
7561
    if (not self.cfg.GetVGName() and
7562
        self.op.disk_template not in constants.DTS_NOT_LVM):
7563
      raise errors.OpPrereqError("Cluster does not support lvm-based"
7564
                                 " instances", errors.ECODE_STATE)
7565

    
7566
    if self.op.hypervisor is None:
7567
      self.op.hypervisor = self.cfg.GetHypervisorType()
7568

    
7569
    cluster = self.cfg.GetClusterInfo()
7570
    enabled_hvs = cluster.enabled_hypervisors
7571
    if self.op.hypervisor not in enabled_hvs:
7572
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
7573
                                 " cluster (%s)" % (self.op.hypervisor,
7574
                                  ",".join(enabled_hvs)),
7575
                                 errors.ECODE_STATE)
7576

    
7577
    # check hypervisor parameter syntax (locally)
7578
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
7579
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
7580
                                      self.op.hvparams)
7581
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
7582
    hv_type.CheckParameterSyntax(filled_hvp)
7583
    self.hv_full = filled_hvp
7584
    # check that we don't specify global parameters on an instance
7585
    _CheckGlobalHvParams(self.op.hvparams)
7586

    
7587
    # fill and remember the beparams dict
7588
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
7589
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
7590

    
7591
    # build os parameters
7592
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
7593

    
7594
    # now that hvp/bep are in final format, let's reset to defaults,
7595
    # if told to do so
7596
    if self.op.identify_defaults:
7597
      self._RevertToDefaults(cluster)
7598

    
7599
    # NIC buildup
7600
    self.nics = []
7601
    for idx, nic in enumerate(self.op.nics):
7602
      nic_mode_req = nic.get("mode", None)
7603
      nic_mode = nic_mode_req
7604
      if nic_mode is None:
7605
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
7606

    
7607
      # in routed mode, for the first nic, the default ip is 'auto'
7608
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
7609
        default_ip_mode = constants.VALUE_AUTO
7610
      else:
7611
        default_ip_mode = constants.VALUE_NONE
7612

    
7613
      # ip validity checks
7614
      ip = nic.get("ip", default_ip_mode)
7615
      if ip is None or ip.lower() == constants.VALUE_NONE:
7616
        nic_ip = None
7617
      elif ip.lower() == constants.VALUE_AUTO:
7618
        if not self.op.name_check:
7619
          raise errors.OpPrereqError("IP address set to auto but name checks"
7620
                                     " have been skipped",
7621
                                     errors.ECODE_INVAL)
7622
        nic_ip = self.hostname1.ip
7623
      else:
7624
        if not netutils.IPAddress.IsValid(ip):
7625
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
7626
                                     errors.ECODE_INVAL)
7627
        nic_ip = ip
7628

    
7629
      # TODO: check the ip address for uniqueness
7630
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
7631
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
7632
                                   errors.ECODE_INVAL)
7633

    
7634
      # MAC address verification
7635
      mac = nic.get("mac", constants.VALUE_AUTO)
7636
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7637
        mac = utils.NormalizeAndValidateMac(mac)
7638

    
7639
        try:
7640
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
7641
        except errors.ReservationError:
7642
          raise errors.OpPrereqError("MAC address %s already in use"
7643
                                     " in cluster" % mac,
7644
                                     errors.ECODE_NOTUNIQUE)
7645

    
7646
      # bridge verification
7647
      bridge = nic.get("bridge", None)
7648
      link = nic.get("link", None)
7649
      if bridge and link:
7650
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
7651
                                   " at the same time", errors.ECODE_INVAL)
7652
      elif bridge and nic_mode == constants.NIC_MODE_ROUTED:
7653
        raise errors.OpPrereqError("Cannot pass 'bridge' on a routed nic",
7654
                                   errors.ECODE_INVAL)
7655
      elif bridge:
7656
        link = bridge
7657

    
7658
      nicparams = {}
7659
      if nic_mode_req:
7660
        nicparams[constants.NIC_MODE] = nic_mode_req
7661
      if link:
7662
        nicparams[constants.NIC_LINK] = link
7663

    
7664
      check_params = cluster.SimpleFillNIC(nicparams)
7665
      objects.NIC.CheckParameterSyntax(check_params)
7666
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
7667

    
7668
    # disk checks/pre-build
7669
    self.disks = []
7670
    for disk in self.op.disks:
7671
      mode = disk.get("mode", constants.DISK_RDWR)
7672
      if mode not in constants.DISK_ACCESS_SET:
7673
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
7674
                                   mode, errors.ECODE_INVAL)
7675
      size = disk.get("size", None)
7676
      if size is None:
7677
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
7678
      try:
7679
        size = int(size)
7680
      except (TypeError, ValueError):
7681
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
7682
                                   errors.ECODE_INVAL)
7683
      vg = disk.get("vg", self.cfg.GetVGName())
7684
      new_disk = {"size": size, "mode": mode, "vg": vg}
7685
      if "adopt" in disk:
7686
        new_disk["adopt"] = disk["adopt"]
7687
      self.disks.append(new_disk)
7688

    
7689
    if self.op.mode == constants.INSTANCE_IMPORT:
7690

    
7691
      # Check that the new instance doesn't have less disks than the export
7692
      instance_disks = len(self.disks)
7693
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
7694
      if instance_disks < export_disks:
7695
        raise errors.OpPrereqError("Not enough disks to import."
7696
                                   " (instance: %d, export: %d)" %
7697
                                   (instance_disks, export_disks),
7698
                                   errors.ECODE_INVAL)
7699

    
7700
      disk_images = []
7701
      for idx in range(export_disks):
7702
        option = 'disk%d_dump' % idx
7703
        if export_info.has_option(constants.INISECT_INS, option):
7704
          # FIXME: are the old os-es, disk sizes, etc. useful?
7705
          export_name = export_info.get(constants.INISECT_INS, option)
7706
          image = utils.PathJoin(self.op.src_path, export_name)
7707
          disk_images.append(image)
7708
        else:
7709
          disk_images.append(False)
7710

    
7711
      self.src_images = disk_images
7712

    
7713
      old_name = export_info.get(constants.INISECT_INS, 'name')
7714
      try:
7715
        exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
7716
      except (TypeError, ValueError), err:
7717
        raise errors.OpPrereqError("Invalid export file, nic_count is not"
7718
                                   " an integer: %s" % str(err),
7719
                                   errors.ECODE_STATE)
7720
      if self.op.instance_name == old_name:
7721
        for idx, nic in enumerate(self.nics):
7722
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
7723
            nic_mac_ini = 'nic%d_mac' % idx
7724
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
7725

    
7726
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
7727

    
7728
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
7729
    if self.op.ip_check:
7730
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
7731
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
7732
                                   (self.check_ip, self.op.instance_name),
7733
                                   errors.ECODE_NOTUNIQUE)
7734

    
7735
    #### mac address generation
7736
    # By generating here the mac address both the allocator and the hooks get
7737
    # the real final mac address rather than the 'auto' or 'generate' value.
7738
    # There is a race condition between the generation and the instance object
7739
    # creation, which means that we know the mac is valid now, but we're not
7740
    # sure it will be when we actually add the instance. If things go bad
7741
    # adding the instance will abort because of a duplicate mac, and the
7742
    # creation job will fail.
7743
    for nic in self.nics:
7744
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7745
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
7746

    
7747
    #### allocator run
7748

    
7749
    if self.op.iallocator is not None:
7750
      self._RunAllocator()
7751

    
7752
    #### node related checks
7753

    
7754
    # check primary node
7755
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
7756
    assert self.pnode is not None, \
7757
      "Cannot retrieve locked node %s" % self.op.pnode
7758
    if pnode.offline:
7759
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
7760
                                 pnode.name, errors.ECODE_STATE)
7761
    if pnode.drained:
7762
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
7763
                                 pnode.name, errors.ECODE_STATE)
7764
    if not pnode.vm_capable:
7765
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
7766
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
7767

    
7768
    self.secondaries = []
7769

    
7770
    # mirror node verification
7771
    if self.op.disk_template in constants.DTS_INT_MIRROR:
7772
      if self.op.snode == pnode.name:
7773
        raise errors.OpPrereqError("The secondary node cannot be the"
7774
                                   " primary node.", errors.ECODE_INVAL)
7775
      _CheckNodeOnline(self, self.op.snode)
7776
      _CheckNodeNotDrained(self, self.op.snode)
7777
      _CheckNodeVmCapable(self, self.op.snode)
7778
      self.secondaries.append(self.op.snode)
7779

    
7780
    nodenames = [pnode.name] + self.secondaries
7781

    
7782
    if not self.adopt_disks:
7783
      # Check lv size requirements, if not adopting
7784
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
7785
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
7786

    
7787
    elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
7788
      all_lvs = set([i["vg"] + "/" + i["adopt"] for i in self.disks])
7789
      if len(all_lvs) != len(self.disks):
7790
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
7791
                                   errors.ECODE_INVAL)
7792
      for lv_name in all_lvs:
7793
        try:
7794
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
7795
          # to ReserveLV uses the same syntax
7796
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
7797
        except errors.ReservationError:
7798
          raise errors.OpPrereqError("LV named %s used by another instance" %
7799
                                     lv_name, errors.ECODE_NOTUNIQUE)
7800

    
7801
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
7802
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
7803

    
7804
      node_lvs = self.rpc.call_lv_list([pnode.name],
7805
                                       vg_names.payload.keys())[pnode.name]
7806
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
7807
      node_lvs = node_lvs.payload
7808

    
7809
      delta = all_lvs.difference(node_lvs.keys())
7810
      if delta:
7811
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
7812
                                   utils.CommaJoin(delta),
7813
                                   errors.ECODE_INVAL)
7814
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
7815
      if online_lvs:
7816
        raise errors.OpPrereqError("Online logical volumes found, cannot"
7817
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
7818
                                   errors.ECODE_STATE)
7819
      # update the size of disk based on what is found
7820
      for dsk in self.disks:
7821
        dsk["size"] = int(float(node_lvs[dsk["vg"] + "/" + dsk["adopt"]][0]))
7822

    
7823
    elif self.op.disk_template == constants.DT_BLOCK:
7824
      # Normalize and de-duplicate device paths
7825
      all_disks = set([os.path.abspath(i["adopt"]) for i in self.disks])
7826
      if len(all_disks) != len(self.disks):
7827
        raise errors.OpPrereqError("Duplicate disk names given for adoption",
7828
                                   errors.ECODE_INVAL)
7829
      baddisks = [d for d in all_disks
7830
                  if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
7831
      if baddisks:
7832
        raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
7833
                                   " cannot be adopted" %
7834
                                   (", ".join(baddisks),
7835
                                    constants.ADOPTABLE_BLOCKDEV_ROOT),
7836
                                   errors.ECODE_INVAL)
7837

    
7838
      node_disks = self.rpc.call_bdev_sizes([pnode.name],
7839
                                            list(all_disks))[pnode.name]
7840
      node_disks.Raise("Cannot get block device information from node %s" %
7841
                       pnode.name)
7842
      node_disks = node_disks.payload
7843
      delta = all_disks.difference(node_disks.keys())
7844
      if delta:
7845
        raise errors.OpPrereqError("Missing block device(s): %s" %
7846
                                   utils.CommaJoin(delta),
7847
                                   errors.ECODE_INVAL)
7848
      for dsk in self.disks:
7849
        dsk["size"] = int(float(node_disks[dsk["adopt"]]))
7850

    
7851
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
7852

    
7853
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
7854
    # check OS parameters (remotely)
7855
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
7856

    
7857
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
7858

    
7859
    # memory check on primary node
7860
    if self.op.start:
7861
      _CheckNodeFreeMemory(self, self.pnode.name,
7862
                           "creating instance %s" % self.op.instance_name,
7863
                           self.be_full[constants.BE_MEMORY],
7864
                           self.op.hypervisor)
7865

    
7866
    self.dry_run_result = list(nodenames)
7867

    
7868
  def Exec(self, feedback_fn):
7869
    """Create and add the instance to the cluster.
7870

7871
    """
7872
    instance = self.op.instance_name
7873
    pnode_name = self.pnode.name
7874

    
7875
    ht_kind = self.op.hypervisor
7876
    if ht_kind in constants.HTS_REQ_PORT:
7877
      network_port = self.cfg.AllocatePort()
7878
    else:
7879
      network_port = None
7880

    
7881
    if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE:
7882
      # this is needed because os.path.join does not accept None arguments
7883
      if self.op.file_storage_dir is None:
7884
        string_file_storage_dir = ""
7885
      else:
7886
        string_file_storage_dir = self.op.file_storage_dir
7887

    
7888
      # build the full file storage dir path
7889
      if self.op.disk_template == constants.DT_SHARED_FILE:
7890
        get_fsd_fn = self.cfg.GetSharedFileStorageDir
7891
      else:
7892
        get_fsd_fn = self.cfg.GetFileStorageDir
7893

    
7894
      file_storage_dir = utils.PathJoin(get_fsd_fn(),
7895
                                        string_file_storage_dir, instance)
7896
    else:
7897
      file_storage_dir = ""
7898

    
7899
    disks = _GenerateDiskTemplate(self,
7900
                                  self.op.disk_template,
7901
                                  instance, pnode_name,
7902
                                  self.secondaries,
7903
                                  self.disks,
7904
                                  file_storage_dir,
7905
                                  self.op.file_driver,
7906
                                  0,
7907
                                  feedback_fn)
7908

    
7909
    iobj = objects.Instance(name=instance, os=self.op.os_type,
7910
                            primary_node=pnode_name,
7911
                            nics=self.nics, disks=disks,
7912
                            disk_template=self.op.disk_template,
7913
                            admin_up=False,
7914
                            network_port=network_port,
7915
                            beparams=self.op.beparams,
7916
                            hvparams=self.op.hvparams,
7917
                            hypervisor=self.op.hypervisor,
7918
                            osparams=self.op.osparams,
7919
                            )
7920

    
7921
    if self.adopt_disks:
7922
      if self.op.disk_template == constants.DT_PLAIN:
7923
        # rename LVs to the newly-generated names; we need to construct
7924
        # 'fake' LV disks with the old data, plus the new unique_id
7925
        tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
7926
        rename_to = []
7927
        for t_dsk, a_dsk in zip (tmp_disks, self.disks):
7928
          rename_to.append(t_dsk.logical_id)
7929
          t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk["adopt"])
7930
          self.cfg.SetDiskID(t_dsk, pnode_name)
7931
        result = self.rpc.call_blockdev_rename(pnode_name,
7932
                                               zip(tmp_disks, rename_to))
7933
        result.Raise("Failed to rename adoped LVs")
7934
    else:
7935
      feedback_fn("* creating instance disks...")
7936
      try:
7937
        _CreateDisks(self, iobj)
7938
      except errors.OpExecError:
7939
        self.LogWarning("Device creation failed, reverting...")
7940
        try:
7941
          _RemoveDisks(self, iobj)
7942
        finally:
7943
          self.cfg.ReleaseDRBDMinors(instance)
7944
          raise
7945

    
7946
      if self.cfg.GetClusterInfo().prealloc_wipe_disks:
7947
        feedback_fn("* wiping instance disks...")
7948
        try:
7949
          _WipeDisks(self, iobj)
7950
        except errors.OpExecError:
7951
          self.LogWarning("Device wiping failed, reverting...")
7952
          try:
7953
            _RemoveDisks(self, iobj)
7954
          finally:
7955
            self.cfg.ReleaseDRBDMinors(instance)
7956
            raise
7957

    
7958
    feedback_fn("adding instance %s to cluster config" % instance)
7959

    
7960
    self.cfg.AddInstance(iobj, self.proc.GetECId())
7961

    
7962
    # Declare that we don't want to remove the instance lock anymore, as we've
7963
    # added the instance to the config
7964
    del self.remove_locks[locking.LEVEL_INSTANCE]
7965
    # Unlock all the nodes
7966
    if self.op.mode == constants.INSTANCE_IMPORT:
7967
      nodes_keep = [self.op.src_node]
7968
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
7969
                       if node != self.op.src_node]
7970
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
7971
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
7972
    else:
7973
      self.context.glm.release(locking.LEVEL_NODE)
7974
      del self.acquired_locks[locking.LEVEL_NODE]
7975

    
7976
    if self.op.wait_for_sync:
7977
      disk_abort = not _WaitForSync(self, iobj)
7978
    elif iobj.disk_template in constants.DTS_INT_MIRROR:
7979
      # make sure the disks are not degraded (still sync-ing is ok)
7980
      time.sleep(15)
7981
      feedback_fn("* checking mirrors status")
7982
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
7983
    else:
7984
      disk_abort = False
7985

    
7986
    if disk_abort:
7987
      _RemoveDisks(self, iobj)
7988
      self.cfg.RemoveInstance(iobj.name)
7989
      # Make sure the instance lock gets removed
7990
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
7991
      raise errors.OpExecError("There are some degraded disks for"
7992
                               " this instance")
7993

    
7994
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
7995
      if self.op.mode == constants.INSTANCE_CREATE:
7996
        if not self.op.no_install:
7997
          feedback_fn("* running the instance OS create scripts...")
7998
          # FIXME: pass debug option from opcode to backend
7999
          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
8000
                                                 self.op.debug_level)
8001
          result.Raise("Could not add os for instance %s"
8002
                       " on node %s" % (instance, pnode_name))
8003

    
8004
      elif self.op.mode == constants.INSTANCE_IMPORT:
8005
        feedback_fn("* running the instance OS import scripts...")
8006

    
8007
        transfers = []
8008

    
8009
        for idx, image in enumerate(self.src_images):
8010
          if not image:
8011
            continue
8012

    
8013
          # FIXME: pass debug option from opcode to backend
8014
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
8015
                                             constants.IEIO_FILE, (image, ),
8016
                                             constants.IEIO_SCRIPT,
8017
                                             (iobj.disks[idx], idx),
8018
                                             None)
8019
          transfers.append(dt)
8020

    
8021
        import_result = \
8022
          masterd.instance.TransferInstanceData(self, feedback_fn,
8023
                                                self.op.src_node, pnode_name,
8024
                                                self.pnode.secondary_ip,
8025
                                                iobj, transfers)
8026
        if not compat.all(import_result):
8027
          self.LogWarning("Some disks for instance %s on node %s were not"
8028
                          " imported successfully" % (instance, pnode_name))
8029

    
8030
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
8031
        feedback_fn("* preparing remote import...")
8032
        # The source cluster will stop the instance before attempting to make a
8033
        # connection. In some cases stopping an instance can take a long time,
8034
        # hence the shutdown timeout is added to the connection timeout.
8035
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
8036
                           self.op.source_shutdown_timeout)
8037
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
8038

    
8039
        assert iobj.primary_node == self.pnode.name
8040
        disk_results = \
8041
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
8042
                                        self.source_x509_ca,
8043
                                        self._cds, timeouts)
8044
        if not compat.all(disk_results):
8045
          # TODO: Should the instance still be started, even if some disks
8046
          # failed to import (valid for local imports, too)?
8047
          self.LogWarning("Some disks for instance %s on node %s were not"
8048
                          " imported successfully" % (instance, pnode_name))
8049

    
8050
        # Run rename script on newly imported instance
8051
        assert iobj.name == instance
8052
        feedback_fn("Running rename script for %s" % instance)
8053
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
8054
                                                   self.source_instance_name,
8055
                                                   self.op.debug_level)
8056
        if result.fail_msg:
8057
          self.LogWarning("Failed to run rename script for %s on node"
8058
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
8059

    
8060
      else:
8061
        # also checked in the prereq part
8062
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
8063
                                     % self.op.mode)
8064

    
8065
    if self.op.start:
8066
      iobj.admin_up = True
8067
      self.cfg.Update(iobj, feedback_fn)
8068
      logging.info("Starting instance %s on node %s", instance, pnode_name)
8069
      feedback_fn("* starting instance...")
8070
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
8071
      result.Raise("Could not start instance")
8072

    
8073
    return list(iobj.all_nodes)
8074

    
8075

    
8076
class LUInstanceConsole(NoHooksLU):
8077
  """Connect to an instance's console.
8078

8079
  This is somewhat special in that it returns the command line that
8080
  you need to run on the master node in order to connect to the
8081
  console.
8082

8083
  """
8084
  REQ_BGL = False
8085

    
8086
  def ExpandNames(self):
8087
    self._ExpandAndLockInstance()
8088

    
8089
  def CheckPrereq(self):
8090
    """Check prerequisites.
8091

8092
    This checks that the instance is in the cluster.
8093

8094
    """
8095
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8096
    assert self.instance is not None, \
8097
      "Cannot retrieve locked instance %s" % self.op.instance_name
8098
    _CheckNodeOnline(self, self.instance.primary_node)
8099

    
8100
  def Exec(self, feedback_fn):
8101
    """Connect to the console of an instance
8102

8103
    """
8104
    instance = self.instance
8105
    node = instance.primary_node
8106

    
8107
    node_insts = self.rpc.call_instance_list([node],
8108
                                             [instance.hypervisor])[node]
8109
    node_insts.Raise("Can't get node information from %s" % node)
8110

    
8111
    if instance.name not in node_insts.payload:
8112
      if instance.admin_up:
8113
        state = "ERROR_down"
8114
      else:
8115
        state = "ADMIN_down"
8116
      raise errors.OpExecError("Instance %s is not running (state %s)" %
8117
                               (instance.name, state))
8118

    
8119
    logging.debug("Connecting to console of %s on %s", instance.name, node)
8120

    
8121
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
8122

    
8123

    
8124
def _GetInstanceConsole(cluster, instance):
8125
  """Returns console information for an instance.
8126

8127
  @type cluster: L{objects.Cluster}
8128
  @type instance: L{objects.Instance}
8129
  @rtype: dict
8130

8131
  """
8132
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
8133
  # beparams and hvparams are passed separately, to avoid editing the
8134
  # instance and then saving the defaults in the instance itself.
8135
  hvparams = cluster.FillHV(instance)
8136
  beparams = cluster.FillBE(instance)
8137
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
8138

    
8139
  assert console.instance == instance.name
8140
  assert console.Validate()
8141

    
8142
  return console.ToDict()
8143

    
8144

    
8145
class LUInstanceReplaceDisks(LogicalUnit):
8146
  """Replace the disks of an instance.
8147

8148
  """
8149
  HPATH = "mirrors-replace"
8150
  HTYPE = constants.HTYPE_INSTANCE
8151
  REQ_BGL = False
8152

    
8153
  def CheckArguments(self):
8154
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
8155
                                  self.op.iallocator)
8156

    
8157
  def ExpandNames(self):
8158
    self._ExpandAndLockInstance()
8159

    
8160
    if self.op.iallocator is not None:
8161
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8162

    
8163
    elif self.op.remote_node is not None:
8164
      remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8165
      self.op.remote_node = remote_node
8166

    
8167
      # Warning: do not remove the locking of the new secondary here
8168
      # unless DRBD8.AddChildren is changed to work in parallel;
8169
      # currently it doesn't since parallel invocations of
8170
      # FindUnusedMinor will conflict
8171
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
8172
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
8173

    
8174
    else:
8175
      self.needed_locks[locking.LEVEL_NODE] = []
8176
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8177

    
8178
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
8179
                                   self.op.iallocator, self.op.remote_node,
8180
                                   self.op.disks, False, self.op.early_release)
8181

    
8182
    self.tasklets = [self.replacer]
8183

    
8184
  def DeclareLocks(self, level):
8185
    # If we're not already locking all nodes in the set we have to declare the
8186
    # instance's primary/secondary nodes.
8187
    if (level == locking.LEVEL_NODE and
8188
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
8189
      self._LockInstancesNodes()
8190

    
8191
  def BuildHooksEnv(self):
8192
    """Build hooks env.
8193

8194
    This runs on the master, the primary and all the secondaries.
8195

8196
    """
8197
    instance = self.replacer.instance
8198
    env = {
8199
      "MODE": self.op.mode,
8200
      "NEW_SECONDARY": self.op.remote_node,
8201
      "OLD_SECONDARY": instance.secondary_nodes[0],
8202
      }
8203
    env.update(_BuildInstanceHookEnvByObject(self, instance))
8204
    nl = [
8205
      self.cfg.GetMasterNode(),
8206
      instance.primary_node,
8207
      ]
8208
    if self.op.remote_node is not None:
8209
      nl.append(self.op.remote_node)
8210
    return env, nl, nl
8211

    
8212

    
8213
class TLReplaceDisks(Tasklet):
8214
  """Replaces disks for an instance.
8215

8216
  Note: Locking is not within the scope of this class.
8217

8218
  """
8219
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
8220
               disks, delay_iallocator, early_release):
8221
    """Initializes this class.
8222

8223
    """
8224
    Tasklet.__init__(self, lu)
8225

    
8226
    # Parameters
8227
    self.instance_name = instance_name
8228
    self.mode = mode
8229
    self.iallocator_name = iallocator_name
8230
    self.remote_node = remote_node
8231
    self.disks = disks
8232
    self.delay_iallocator = delay_iallocator
8233
    self.early_release = early_release
8234

    
8235
    # Runtime data
8236
    self.instance = None
8237
    self.new_node = None
8238
    self.target_node = None
8239
    self.other_node = None
8240
    self.remote_node_info = None
8241
    self.node_secondary_ip = None
8242

    
8243
  @staticmethod
8244
  def CheckArguments(mode, remote_node, iallocator):
8245
    """Helper function for users of this class.
8246

8247
    """
8248
    # check for valid parameter combination
8249
    if mode == constants.REPLACE_DISK_CHG:
8250
      if remote_node is None and iallocator is None:
8251
        raise errors.OpPrereqError("When changing the secondary either an"
8252
                                   " iallocator script must be used or the"
8253
                                   " new node given", errors.ECODE_INVAL)
8254

    
8255
      if remote_node is not None and iallocator is not None:
8256
        raise errors.OpPrereqError("Give either the iallocator or the new"
8257
                                   " secondary, not both", errors.ECODE_INVAL)
8258

    
8259
    elif remote_node is not None or iallocator is not None:
8260
      # Not replacing the secondary
8261
      raise errors.OpPrereqError("The iallocator and new node options can"
8262
                                 " only be used when changing the"
8263
                                 " secondary node", errors.ECODE_INVAL)
8264

    
8265
  @staticmethod
8266
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
8267
    """Compute a new secondary node using an IAllocator.
8268

8269
    """
8270
    ial = IAllocator(lu.cfg, lu.rpc,
8271
                     mode=constants.IALLOCATOR_MODE_RELOC,
8272
                     name=instance_name,
8273
                     relocate_from=relocate_from)
8274

    
8275
    ial.Run(iallocator_name)
8276

    
8277
    if not ial.success:
8278
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
8279
                                 " %s" % (iallocator_name, ial.info),
8280
                                 errors.ECODE_NORES)
8281

    
8282
    if len(ial.result) != ial.required_nodes:
8283
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8284
                                 " of nodes (%s), required %s" %
8285
                                 (iallocator_name,
8286
                                  len(ial.result), ial.required_nodes),
8287
                                 errors.ECODE_FAULT)
8288

    
8289
    remote_node_name = ial.result[0]
8290

    
8291
    lu.LogInfo("Selected new secondary for instance '%s': %s",
8292
               instance_name, remote_node_name)
8293

    
8294
    return remote_node_name
8295

    
8296
  def _FindFaultyDisks(self, node_name):
8297
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
8298
                                    node_name, True)
8299

    
8300
  def CheckPrereq(self):
8301
    """Check prerequisites.
8302

8303
    This checks that the instance is in the cluster.
8304

8305
    """
8306
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
8307
    assert instance is not None, \
8308
      "Cannot retrieve locked instance %s" % self.instance_name
8309

    
8310
    if instance.disk_template != constants.DT_DRBD8:
8311
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
8312
                                 " instances", errors.ECODE_INVAL)
8313

    
8314
    if len(instance.secondary_nodes) != 1:
8315
      raise errors.OpPrereqError("The instance has a strange layout,"
8316
                                 " expected one secondary but found %d" %
8317
                                 len(instance.secondary_nodes),
8318
                                 errors.ECODE_FAULT)
8319

    
8320
    if not self.delay_iallocator:
8321
      self._CheckPrereq2()
8322

    
8323
  def _CheckPrereq2(self):
8324
    """Check prerequisites, second part.
8325

8326
    This function should always be part of CheckPrereq. It was separated and is
8327
    now called from Exec because during node evacuation iallocator was only
8328
    called with an unmodified cluster model, not taking planned changes into
8329
    account.
8330

8331
    """
8332
    instance = self.instance
8333
    secondary_node = instance.secondary_nodes[0]
8334

    
8335
    if self.iallocator_name is None:
8336
      remote_node = self.remote_node
8337
    else:
8338
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
8339
                                       instance.name, instance.secondary_nodes)
8340

    
8341
    if remote_node is not None:
8342
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
8343
      assert self.remote_node_info is not None, \
8344
        "Cannot retrieve locked node %s" % remote_node
8345
    else:
8346
      self.remote_node_info = None
8347

    
8348
    if remote_node == self.instance.primary_node:
8349
      raise errors.OpPrereqError("The specified node is the primary node of"
8350
                                 " the instance.", errors.ECODE_INVAL)
8351

    
8352
    if remote_node == secondary_node:
8353
      raise errors.OpPrereqError("The specified node is already the"
8354
                                 " secondary node of the instance.",
8355
                                 errors.ECODE_INVAL)
8356

    
8357
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
8358
                                    constants.REPLACE_DISK_CHG):
8359
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
8360
                                 errors.ECODE_INVAL)
8361

    
8362
    if self.mode == constants.REPLACE_DISK_AUTO:
8363
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
8364
      faulty_secondary = self._FindFaultyDisks(secondary_node)
8365

    
8366
      if faulty_primary and faulty_secondary:
8367
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
8368
                                   " one node and can not be repaired"
8369
                                   " automatically" % self.instance_name,
8370
                                   errors.ECODE_STATE)
8371

    
8372
      if faulty_primary:
8373
        self.disks = faulty_primary
8374
        self.target_node = instance.primary_node
8375
        self.other_node = secondary_node
8376
        check_nodes = [self.target_node, self.other_node]
8377
      elif faulty_secondary:
8378
        self.disks = faulty_secondary
8379
        self.target_node = secondary_node
8380
        self.other_node = instance.primary_node
8381
        check_nodes = [self.target_node, self.other_node]
8382
      else:
8383
        self.disks = []
8384
        check_nodes = []
8385

    
8386
    else:
8387
      # Non-automatic modes
8388
      if self.mode == constants.REPLACE_DISK_PRI:
8389
        self.target_node = instance.primary_node
8390
        self.other_node = secondary_node
8391
        check_nodes = [self.target_node, self.other_node]
8392

    
8393
      elif self.mode == constants.REPLACE_DISK_SEC:
8394
        self.target_node = secondary_node
8395
        self.other_node = instance.primary_node
8396
        check_nodes = [self.target_node, self.other_node]
8397

    
8398
      elif self.mode == constants.REPLACE_DISK_CHG:
8399
        self.new_node = remote_node
8400
        self.other_node = instance.primary_node
8401
        self.target_node = secondary_node
8402
        check_nodes = [self.new_node, self.other_node]
8403

    
8404
        _CheckNodeNotDrained(self.lu, remote_node)
8405
        _CheckNodeVmCapable(self.lu, remote_node)
8406

    
8407
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
8408
        assert old_node_info is not None
8409
        if old_node_info.offline and not self.early_release:
8410
          # doesn't make sense to delay the release
8411
          self.early_release = True
8412
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
8413
                          " early-release mode", secondary_node)
8414

    
8415
      else:
8416
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
8417
                                     self.mode)
8418

    
8419
      # If not specified all disks should be replaced
8420
      if not self.disks:
8421
        self.disks = range(len(self.instance.disks))
8422

    
8423
    for node in check_nodes:
8424
      _CheckNodeOnline(self.lu, node)
8425

    
8426
    # Check whether disks are valid
8427
    for disk_idx in self.disks:
8428
      instance.FindDisk(disk_idx)
8429

    
8430
    # Get secondary node IP addresses
8431
    node_2nd_ip = {}
8432

    
8433
    for node_name in [self.target_node, self.other_node, self.new_node]:
8434
      if node_name is not None:
8435
        node_2nd_ip[node_name] = self.cfg.GetNodeInfo(node_name).secondary_ip
8436

    
8437
    self.node_secondary_ip = node_2nd_ip
8438

    
8439
  def Exec(self, feedback_fn):
8440
    """Execute disk replacement.
8441

8442
    This dispatches the disk replacement to the appropriate handler.
8443

8444
    """
8445
    if self.delay_iallocator:
8446
      self._CheckPrereq2()
8447

    
8448
    if not self.disks:
8449
      feedback_fn("No disks need replacement")
8450
      return
8451

    
8452
    feedback_fn("Replacing disk(s) %s for %s" %
8453
                (utils.CommaJoin(self.disks), self.instance.name))
8454

    
8455
    activate_disks = (not self.instance.admin_up)
8456

    
8457
    # Activate the instance disks if we're replacing them on a down instance
8458
    if activate_disks:
8459
      _StartInstanceDisks(self.lu, self.instance, True)
8460

    
8461
    try:
8462
      # Should we replace the secondary node?
8463
      if self.new_node is not None:
8464
        fn = self._ExecDrbd8Secondary
8465
      else:
8466
        fn = self._ExecDrbd8DiskOnly
8467

    
8468
      return fn(feedback_fn)
8469

    
8470
    finally:
8471
      # Deactivate the instance disks if we're replacing them on a
8472
      # down instance
8473
      if activate_disks:
8474
        _SafeShutdownInstanceDisks(self.lu, self.instance)
8475

    
8476
  def _CheckVolumeGroup(self, nodes):
8477
    self.lu.LogInfo("Checking volume groups")
8478

    
8479
    vgname = self.cfg.GetVGName()
8480

    
8481
    # Make sure volume group exists on all involved nodes
8482
    results = self.rpc.call_vg_list(nodes)
8483
    if not results:
8484
      raise errors.OpExecError("Can't list volume groups on the nodes")
8485

    
8486
    for node in nodes:
8487
      res = results[node]
8488
      res.Raise("Error checking node %s" % node)
8489
      if vgname not in res.payload:
8490
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
8491
                                 (vgname, node))
8492

    
8493
  def _CheckDisksExistence(self, nodes):
8494
    # Check disk existence
8495
    for idx, dev in enumerate(self.instance.disks):
8496
      if idx not in self.disks:
8497
        continue
8498

    
8499
      for node in nodes:
8500
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
8501
        self.cfg.SetDiskID(dev, node)
8502

    
8503
        result = self.rpc.call_blockdev_find(node, dev)
8504

    
8505
        msg = result.fail_msg
8506
        if msg or not result.payload:
8507
          if not msg:
8508
            msg = "disk not found"
8509
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
8510
                                   (idx, node, msg))
8511

    
8512
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
8513
    for idx, dev in enumerate(self.instance.disks):
8514
      if idx not in self.disks:
8515
        continue
8516

    
8517
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
8518
                      (idx, node_name))
8519

    
8520
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
8521
                                   ldisk=ldisk):
8522
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
8523
                                 " replace disks for instance %s" %
8524
                                 (node_name, self.instance.name))
8525

    
8526
  def _CreateNewStorage(self, node_name):
8527
    vgname = self.cfg.GetVGName()
8528
    iv_names = {}
8529

    
8530
    for idx, dev in enumerate(self.instance.disks):
8531
      if idx not in self.disks:
8532
        continue
8533

    
8534
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
8535

    
8536
      self.cfg.SetDiskID(dev, node_name)
8537

    
8538
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
8539
      names = _GenerateUniqueNames(self.lu, lv_names)
8540

    
8541
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
8542
                             logical_id=(vgname, names[0]))
8543
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
8544
                             logical_id=(vgname, names[1]))
8545

    
8546
      new_lvs = [lv_data, lv_meta]
8547
      old_lvs = dev.children
8548
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
8549

    
8550
      # we pass force_create=True to force the LVM creation
8551
      for new_lv in new_lvs:
8552
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
8553
                        _GetInstanceInfoText(self.instance), False)
8554

    
8555
    return iv_names
8556

    
8557
  def _CheckDevices(self, node_name, iv_names):
8558
    for name, (dev, _, _) in iv_names.iteritems():
8559
      self.cfg.SetDiskID(dev, node_name)
8560

    
8561
      result = self.rpc.call_blockdev_find(node_name, dev)
8562

    
8563
      msg = result.fail_msg
8564
      if msg or not result.payload:
8565
        if not msg:
8566
          msg = "disk not found"
8567
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
8568
                                 (name, msg))
8569

    
8570
      if result.payload.is_degraded:
8571
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
8572

    
8573
  def _RemoveOldStorage(self, node_name, iv_names):
8574
    for name, (_, old_lvs, _) in iv_names.iteritems():
8575
      self.lu.LogInfo("Remove logical volumes for %s" % name)
8576

    
8577
      for lv in old_lvs:
8578
        self.cfg.SetDiskID(lv, node_name)
8579

    
8580
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
8581
        if msg:
8582
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
8583
                             hint="remove unused LVs manually")
8584

    
8585
  def _ReleaseNodeLock(self, node_name):
8586
    """Releases the lock for a given node."""
8587
    self.lu.context.glm.release(locking.LEVEL_NODE, node_name)
8588

    
8589
  def _ExecDrbd8DiskOnly(self, feedback_fn):
8590
    """Replace a disk on the primary or secondary for DRBD 8.
8591

8592
    The algorithm for replace is quite complicated:
8593

8594
      1. for each disk to be replaced:
8595

8596
        1. create new LVs on the target node with unique names
8597
        1. detach old LVs from the drbd device
8598
        1. rename old LVs to name_replaced.<time_t>
8599
        1. rename new LVs to old LVs
8600
        1. attach the new LVs (with the old names now) to the drbd device
8601

8602
      1. wait for sync across all devices
8603

8604
      1. for each modified disk:
8605

8606
        1. remove old LVs (which have the name name_replaces.<time_t>)
8607

8608
    Failures are not very well handled.
8609

8610
    """
8611
    steps_total = 6
8612

    
8613
    # Step: check device activation
8614
    self.lu.LogStep(1, steps_total, "Check device existence")
8615
    self._CheckDisksExistence([self.other_node, self.target_node])
8616
    self._CheckVolumeGroup([self.target_node, self.other_node])
8617

    
8618
    # Step: check other node consistency
8619
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8620
    self._CheckDisksConsistency(self.other_node,
8621
                                self.other_node == self.instance.primary_node,
8622
                                False)
8623

    
8624
    # Step: create new storage
8625
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8626
    iv_names = self._CreateNewStorage(self.target_node)
8627

    
8628
    # Step: for each lv, detach+rename*2+attach
8629
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8630
    for dev, old_lvs, new_lvs in iv_names.itervalues():
8631
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
8632

    
8633
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
8634
                                                     old_lvs)
8635
      result.Raise("Can't detach drbd from local storage on node"
8636
                   " %s for device %s" % (self.target_node, dev.iv_name))
8637
      #dev.children = []
8638
      #cfg.Update(instance)
8639

    
8640
      # ok, we created the new LVs, so now we know we have the needed
8641
      # storage; as such, we proceed on the target node to rename
8642
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
8643
      # using the assumption that logical_id == physical_id (which in
8644
      # turn is the unique_id on that node)
8645

    
8646
      # FIXME(iustin): use a better name for the replaced LVs
8647
      temp_suffix = int(time.time())
8648
      ren_fn = lambda d, suff: (d.physical_id[0],
8649
                                d.physical_id[1] + "_replaced-%s" % suff)
8650

    
8651
      # Build the rename list based on what LVs exist on the node
8652
      rename_old_to_new = []
8653
      for to_ren in old_lvs:
8654
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
8655
        if not result.fail_msg and result.payload:
8656
          # device exists
8657
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
8658

    
8659
      self.lu.LogInfo("Renaming the old LVs on the target node")
8660
      result = self.rpc.call_blockdev_rename(self.target_node,
8661
                                             rename_old_to_new)
8662
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
8663

    
8664
      # Now we rename the new LVs to the old LVs
8665
      self.lu.LogInfo("Renaming the new LVs on the target node")
8666
      rename_new_to_old = [(new, old.physical_id)
8667
                           for old, new in zip(old_lvs, new_lvs)]
8668
      result = self.rpc.call_blockdev_rename(self.target_node,
8669
                                             rename_new_to_old)
8670
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
8671

    
8672
      for old, new in zip(old_lvs, new_lvs):
8673
        new.logical_id = old.logical_id
8674
        self.cfg.SetDiskID(new, self.target_node)
8675

    
8676
      for disk in old_lvs:
8677
        disk.logical_id = ren_fn(disk, temp_suffix)
8678
        self.cfg.SetDiskID(disk, self.target_node)
8679

    
8680
      # Now that the new lvs have the old name, we can add them to the device
8681
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
8682
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
8683
                                                  new_lvs)
8684
      msg = result.fail_msg
8685
      if msg:
8686
        for new_lv in new_lvs:
8687
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
8688
                                               new_lv).fail_msg
8689
          if msg2:
8690
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
8691
                               hint=("cleanup manually the unused logical"
8692
                                     "volumes"))
8693
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
8694

    
8695
      dev.children = new_lvs
8696

    
8697
      self.cfg.Update(self.instance, feedback_fn)
8698

    
8699
    cstep = 5
8700
    if self.early_release:
8701
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8702
      cstep += 1
8703
      self._RemoveOldStorage(self.target_node, iv_names)
8704
      # WARNING: we release both node locks here, do not do other RPCs
8705
      # than WaitForSync to the primary node
8706
      self._ReleaseNodeLock([self.target_node, self.other_node])
8707

    
8708
    # Wait for sync
8709
    # This can fail as the old devices are degraded and _WaitForSync
8710
    # does a combined result over all disks, so we don't check its return value
8711
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8712
    cstep += 1
8713
    _WaitForSync(self.lu, self.instance)
8714

    
8715
    # Check all devices manually
8716
    self._CheckDevices(self.instance.primary_node, iv_names)
8717

    
8718
    # Step: remove old storage
8719
    if not self.early_release:
8720
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8721
      cstep += 1
8722
      self._RemoveOldStorage(self.target_node, iv_names)
8723

    
8724
  def _ExecDrbd8Secondary(self, feedback_fn):
8725
    """Replace the secondary node for DRBD 8.
8726

8727
    The algorithm for replace is quite complicated:
8728
      - for all disks of the instance:
8729
        - create new LVs on the new node with same names
8730
        - shutdown the drbd device on the old secondary
8731
        - disconnect the drbd network on the primary
8732
        - create the drbd device on the new secondary
8733
        - network attach the drbd on the primary, using an artifice:
8734
          the drbd code for Attach() will connect to the network if it
8735
          finds a device which is connected to the good local disks but
8736
          not network enabled
8737
      - wait for sync across all devices
8738
      - remove all disks from the old secondary
8739

8740
    Failures are not very well handled.
8741

8742
    """
8743
    steps_total = 6
8744

    
8745
    # Step: check device activation
8746
    self.lu.LogStep(1, steps_total, "Check device existence")
8747
    self._CheckDisksExistence([self.instance.primary_node])
8748
    self._CheckVolumeGroup([self.instance.primary_node])
8749

    
8750
    # Step: check other node consistency
8751
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8752
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
8753

    
8754
    # Step: create new storage
8755
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8756
    for idx, dev in enumerate(self.instance.disks):
8757
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
8758
                      (self.new_node, idx))
8759
      # we pass force_create=True to force LVM creation
8760
      for new_lv in dev.children:
8761
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
8762
                        _GetInstanceInfoText(self.instance), False)
8763

    
8764
    # Step 4: dbrd minors and drbd setups changes
8765
    # after this, we must manually remove the drbd minors on both the
8766
    # error and the success paths
8767
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8768
    minors = self.cfg.AllocateDRBDMinor([self.new_node
8769
                                         for dev in self.instance.disks],
8770
                                        self.instance.name)
8771
    logging.debug("Allocated minors %r", minors)
8772

    
8773
    iv_names = {}
8774
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
8775
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
8776
                      (self.new_node, idx))
8777
      # create new devices on new_node; note that we create two IDs:
8778
      # one without port, so the drbd will be activated without
8779
      # networking information on the new node at this stage, and one
8780
      # with network, for the latter activation in step 4
8781
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
8782
      if self.instance.primary_node == o_node1:
8783
        p_minor = o_minor1
8784
      else:
8785
        assert self.instance.primary_node == o_node2, "Three-node instance?"
8786
        p_minor = o_minor2
8787

    
8788
      new_alone_id = (self.instance.primary_node, self.new_node, None,
8789
                      p_minor, new_minor, o_secret)
8790
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
8791
                    p_minor, new_minor, o_secret)
8792

    
8793
      iv_names[idx] = (dev, dev.children, new_net_id)
8794
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
8795
                    new_net_id)
8796
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
8797
                              logical_id=new_alone_id,
8798
                              children=dev.children,
8799
                              size=dev.size)
8800
      try:
8801
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
8802
                              _GetInstanceInfoText(self.instance), False)
8803
      except errors.GenericError:
8804
        self.cfg.ReleaseDRBDMinors(self.instance.name)
8805
        raise
8806

    
8807
    # We have new devices, shutdown the drbd on the old secondary
8808
    for idx, dev in enumerate(self.instance.disks):
8809
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
8810
      self.cfg.SetDiskID(dev, self.target_node)
8811
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
8812
      if msg:
8813
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
8814
                           "node: %s" % (idx, msg),
8815
                           hint=("Please cleanup this device manually as"
8816
                                 " soon as possible"))
8817

    
8818
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
8819
    result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
8820
                                               self.node_secondary_ip,
8821
                                               self.instance.disks)\
8822
                                              [self.instance.primary_node]
8823

    
8824
    msg = result.fail_msg
8825
    if msg:
8826
      # detaches didn't succeed (unlikely)
8827
      self.cfg.ReleaseDRBDMinors(self.instance.name)
8828
      raise errors.OpExecError("Can't detach the disks from the network on"
8829
                               " old node: %s" % (msg,))
8830

    
8831
    # if we managed to detach at least one, we update all the disks of
8832
    # the instance to point to the new secondary
8833
    self.lu.LogInfo("Updating instance configuration")
8834
    for dev, _, new_logical_id in iv_names.itervalues():
8835
      dev.logical_id = new_logical_id
8836
      self.cfg.SetDiskID(dev, self.instance.primary_node)
8837

    
8838
    self.cfg.Update(self.instance, feedback_fn)
8839

    
8840
    # and now perform the drbd attach
8841
    self.lu.LogInfo("Attaching primary drbds to new secondary"
8842
                    " (standalone => connected)")
8843
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
8844
                                            self.new_node],
8845
                                           self.node_secondary_ip,
8846
                                           self.instance.disks,
8847
                                           self.instance.name,
8848
                                           False)
8849
    for to_node, to_result in result.items():
8850
      msg = to_result.fail_msg
8851
      if msg:
8852
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
8853
                           to_node, msg,
8854
                           hint=("please do a gnt-instance info to see the"
8855
                                 " status of disks"))
8856
    cstep = 5
8857
    if self.early_release:
8858
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8859
      cstep += 1
8860
      self._RemoveOldStorage(self.target_node, iv_names)
8861
      # WARNING: we release all node locks here, do not do other RPCs
8862
      # than WaitForSync to the primary node
8863
      self._ReleaseNodeLock([self.instance.primary_node,
8864
                             self.target_node,
8865
                             self.new_node])
8866

    
8867
    # Wait for sync
8868
    # This can fail as the old devices are degraded and _WaitForSync
8869
    # does a combined result over all disks, so we don't check its return value
8870
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8871
    cstep += 1
8872
    _WaitForSync(self.lu, self.instance)
8873

    
8874
    # Check all devices manually
8875
    self._CheckDevices(self.instance.primary_node, iv_names)
8876

    
8877
    # Step: remove old storage
8878
    if not self.early_release:
8879
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8880
      self._RemoveOldStorage(self.target_node, iv_names)
8881

    
8882

    
8883
class LURepairNodeStorage(NoHooksLU):
8884
  """Repairs the volume group on a node.
8885

8886
  """
8887
  REQ_BGL = False
8888

    
8889
  def CheckArguments(self):
8890
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
8891

    
8892
    storage_type = self.op.storage_type
8893

    
8894
    if (constants.SO_FIX_CONSISTENCY not in
8895
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
8896
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
8897
                                 " repaired" % storage_type,
8898
                                 errors.ECODE_INVAL)
8899

    
8900
  def ExpandNames(self):
8901
    self.needed_locks = {
8902
      locking.LEVEL_NODE: [self.op.node_name],
8903
      }
8904

    
8905
  def _CheckFaultyDisks(self, instance, node_name):
8906
    """Ensure faulty disks abort the opcode or at least warn."""
8907
    try:
8908
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
8909
                                  node_name, True):
8910
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
8911
                                   " node '%s'" % (instance.name, node_name),
8912
                                   errors.ECODE_STATE)
8913
    except errors.OpPrereqError, err:
8914
      if self.op.ignore_consistency:
8915
        self.proc.LogWarning(str(err.args[0]))
8916
      else:
8917
        raise
8918

    
8919
  def CheckPrereq(self):
8920
    """Check prerequisites.
8921

8922
    """
8923
    # Check whether any instance on this node has faulty disks
8924
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
8925
      if not inst.admin_up:
8926
        continue
8927
      check_nodes = set(inst.all_nodes)
8928
      check_nodes.discard(self.op.node_name)
8929
      for inst_node_name in check_nodes:
8930
        self._CheckFaultyDisks(inst, inst_node_name)
8931

    
8932
  def Exec(self, feedback_fn):
8933
    feedback_fn("Repairing storage unit '%s' on %s ..." %
8934
                (self.op.name, self.op.node_name))
8935

    
8936
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
8937
    result = self.rpc.call_storage_execute(self.op.node_name,
8938
                                           self.op.storage_type, st_args,
8939
                                           self.op.name,
8940
                                           constants.SO_FIX_CONSISTENCY)
8941
    result.Raise("Failed to repair storage unit '%s' on %s" %
8942
                 (self.op.name, self.op.node_name))
8943

    
8944

    
8945
class LUNodeEvacStrategy(NoHooksLU):
8946
  """Computes the node evacuation strategy.
8947

8948
  """
8949
  REQ_BGL = False
8950

    
8951
  def CheckArguments(self):
8952
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
8953

    
8954
  def ExpandNames(self):
8955
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
8956
    self.needed_locks = locks = {}
8957
    if self.op.remote_node is None:
8958
      locks[locking.LEVEL_NODE] = locking.ALL_SET
8959
    else:
8960
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8961
      locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
8962

    
8963
  def Exec(self, feedback_fn):
8964
    if self.op.remote_node is not None:
8965
      instances = []
8966
      for node in self.op.nodes:
8967
        instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
8968
      result = []
8969
      for i in instances:
8970
        if i.primary_node == self.op.remote_node:
8971
          raise errors.OpPrereqError("Node %s is the primary node of"
8972
                                     " instance %s, cannot use it as"
8973
                                     " secondary" %
8974
                                     (self.op.remote_node, i.name),
8975
                                     errors.ECODE_INVAL)
8976
        result.append([i.name, self.op.remote_node])
8977
    else:
8978
      ial = IAllocator(self.cfg, self.rpc,
8979
                       mode=constants.IALLOCATOR_MODE_MEVAC,
8980
                       evac_nodes=self.op.nodes)
8981
      ial.Run(self.op.iallocator, validate=True)
8982
      if not ial.success:
8983
        raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
8984
                                 errors.ECODE_NORES)
8985
      result = ial.result
8986
    return result
8987

    
8988

    
8989
class LUInstanceGrowDisk(LogicalUnit):
8990
  """Grow a disk of an instance.
8991

8992
  """
8993
  HPATH = "disk-grow"
8994
  HTYPE = constants.HTYPE_INSTANCE
8995
  REQ_BGL = False
8996

    
8997
  def ExpandNames(self):
8998
    self._ExpandAndLockInstance()
8999
    self.needed_locks[locking.LEVEL_NODE] = []
9000
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9001

    
9002
  def DeclareLocks(self, level):
9003
    if level == locking.LEVEL_NODE:
9004
      self._LockInstancesNodes()
9005

    
9006
  def BuildHooksEnv(self):
9007
    """Build hooks env.
9008

9009
    This runs on the master, the primary and all the secondaries.
9010

9011
    """
9012
    env = {
9013
      "DISK": self.op.disk,
9014
      "AMOUNT": self.op.amount,
9015
      }
9016
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
9017
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
9018
    return env, nl, nl
9019

    
9020
  def CheckPrereq(self):
9021
    """Check prerequisites.
9022

9023
    This checks that the instance is in the cluster.
9024

9025
    """
9026
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9027
    assert instance is not None, \
9028
      "Cannot retrieve locked instance %s" % self.op.instance_name
9029
    nodenames = list(instance.all_nodes)
9030
    for node in nodenames:
9031
      _CheckNodeOnline(self, node)
9032

    
9033
    self.instance = instance
9034

    
9035
    if instance.disk_template not in constants.DTS_GROWABLE:
9036
      raise errors.OpPrereqError("Instance's disk layout does not support"
9037
                                 " growing.", errors.ECODE_INVAL)
9038

    
9039
    self.disk = instance.FindDisk(self.op.disk)
9040

    
9041
    if instance.disk_template not in (constants.DT_FILE,
9042
                                      constants.DT_SHARED_FILE):
9043
      # TODO: check the free disk space for file, when that feature will be
9044
      # supported
9045
      _CheckNodesFreeDiskPerVG(self, nodenames,
9046
                               self.disk.ComputeGrowth(self.op.amount))
9047

    
9048
  def Exec(self, feedback_fn):
9049
    """Execute disk grow.
9050

9051
    """
9052
    instance = self.instance
9053
    disk = self.disk
9054

    
9055
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
9056
    if not disks_ok:
9057
      raise errors.OpExecError("Cannot activate block device to grow")
9058

    
9059
    for node in instance.all_nodes:
9060
      self.cfg.SetDiskID(disk, node)
9061
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
9062
      result.Raise("Grow request failed to node %s" % node)
9063

    
9064
      # TODO: Rewrite code to work properly
9065
      # DRBD goes into sync mode for a short amount of time after executing the
9066
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
9067
      # calling "resize" in sync mode fails. Sleeping for a short amount of
9068
      # time is a work-around.
9069
      time.sleep(5)
9070

    
9071
    disk.RecordGrow(self.op.amount)
9072
    self.cfg.Update(instance, feedback_fn)
9073
    if self.op.wait_for_sync:
9074
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
9075
      if disk_abort:
9076
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
9077
                             " status.\nPlease check the instance.")
9078
      if not instance.admin_up:
9079
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
9080
    elif not instance.admin_up:
9081
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
9082
                           " not supposed to be running because no wait for"
9083
                           " sync mode was requested.")
9084

    
9085

    
9086
class LUInstanceQueryData(NoHooksLU):
9087
  """Query runtime instance data.
9088

9089
  """
9090
  REQ_BGL = False
9091

    
9092
  def ExpandNames(self):
9093
    self.needed_locks = {}
9094
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
9095

    
9096
    if self.op.instances:
9097
      self.wanted_names = []
9098
      for name in self.op.instances:
9099
        full_name = _ExpandInstanceName(self.cfg, name)
9100
        self.wanted_names.append(full_name)
9101
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
9102
    else:
9103
      self.wanted_names = None
9104
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
9105

    
9106
    self.needed_locks[locking.LEVEL_NODE] = []
9107
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9108

    
9109
  def DeclareLocks(self, level):
9110
    if level == locking.LEVEL_NODE:
9111
      self._LockInstancesNodes()
9112

    
9113
  def CheckPrereq(self):
9114
    """Check prerequisites.
9115

9116
    This only checks the optional instance list against the existing names.
9117

9118
    """
9119
    if self.wanted_names is None:
9120
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
9121

    
9122
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
9123
                             in self.wanted_names]
9124

    
9125
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
9126
    """Returns the status of a block device
9127

9128
    """
9129
    if self.op.static or not node:
9130
      return None
9131

    
9132
    self.cfg.SetDiskID(dev, node)
9133

    
9134
    result = self.rpc.call_blockdev_find(node, dev)
9135
    if result.offline:
9136
      return None
9137

    
9138
    result.Raise("Can't compute disk status for %s" % instance_name)
9139

    
9140
    status = result.payload
9141
    if status is None:
9142
      return None
9143

    
9144
    return (status.dev_path, status.major, status.minor,
9145
            status.sync_percent, status.estimated_time,
9146
            status.is_degraded, status.ldisk_status)
9147

    
9148
  def _ComputeDiskStatus(self, instance, snode, dev):
9149
    """Compute block device status.
9150

9151
    """
9152
    if dev.dev_type in constants.LDS_DRBD:
9153
      # we change the snode then (otherwise we use the one passed in)
9154
      if dev.logical_id[0] == instance.primary_node:
9155
        snode = dev.logical_id[1]
9156
      else:
9157
        snode = dev.logical_id[0]
9158

    
9159
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
9160
                                              instance.name, dev)
9161
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
9162

    
9163
    if dev.children:
9164
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
9165
                      for child in dev.children]
9166
    else:
9167
      dev_children = []
9168

    
9169
    data = {
9170
      "iv_name": dev.iv_name,
9171
      "dev_type": dev.dev_type,
9172
      "logical_id": dev.logical_id,
9173
      "physical_id": dev.physical_id,
9174
      "pstatus": dev_pstatus,
9175
      "sstatus": dev_sstatus,
9176
      "children": dev_children,
9177
      "mode": dev.mode,
9178
      "size": dev.size,
9179
      }
9180

    
9181
    return data
9182

    
9183
  def Exec(self, feedback_fn):
9184
    """Gather and return data"""
9185
    result = {}
9186

    
9187
    cluster = self.cfg.GetClusterInfo()
9188

    
9189
    for instance in self.wanted_instances:
9190
      if not self.op.static:
9191
        remote_info = self.rpc.call_instance_info(instance.primary_node,
9192
                                                  instance.name,
9193
                                                  instance.hypervisor)
9194
        remote_info.Raise("Error checking node %s" % instance.primary_node)
9195
        remote_info = remote_info.payload
9196
        if remote_info and "state" in remote_info:
9197
          remote_state = "up"
9198
        else:
9199
          remote_state = "down"
9200
      else:
9201
        remote_state = None
9202
      if instance.admin_up:
9203
        config_state = "up"
9204
      else:
9205
        config_state = "down"
9206

    
9207
      disks = [self._ComputeDiskStatus(instance, None, device)
9208
               for device in instance.disks]
9209

    
9210
      idict = {
9211
        "name": instance.name,
9212
        "config_state": config_state,
9213
        "run_state": remote_state,
9214
        "pnode": instance.primary_node,
9215
        "snodes": instance.secondary_nodes,
9216
        "os": instance.os,
9217
        # this happens to be the same format used for hooks
9218
        "nics": _NICListToTuple(self, instance.nics),
9219
        "disk_template": instance.disk_template,
9220
        "disks": disks,
9221
        "hypervisor": instance.hypervisor,
9222
        "network_port": instance.network_port,
9223
        "hv_instance": instance.hvparams,
9224
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
9225
        "be_instance": instance.beparams,
9226
        "be_actual": cluster.FillBE(instance),
9227
        "os_instance": instance.osparams,
9228
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
9229
        "serial_no": instance.serial_no,
9230
        "mtime": instance.mtime,
9231
        "ctime": instance.ctime,
9232
        "uuid": instance.uuid,
9233
        }
9234

    
9235
      result[instance.name] = idict
9236

    
9237
    return result
9238

    
9239

    
9240
class LUInstanceSetParams(LogicalUnit):
9241
  """Modifies an instances's parameters.
9242

9243
  """
9244
  HPATH = "instance-modify"
9245
  HTYPE = constants.HTYPE_INSTANCE
9246
  REQ_BGL = False
9247

    
9248
  def CheckArguments(self):
9249
    if not (self.op.nics or self.op.disks or self.op.disk_template or
9250
            self.op.hvparams or self.op.beparams or self.op.os_name):
9251
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
9252

    
9253
    if self.op.hvparams:
9254
      _CheckGlobalHvParams(self.op.hvparams)
9255

    
9256
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9257

    
9258
    # Disk validation
9259
    disk_addremove = 0
9260
    for disk_op, disk_dict in self.op.disks:
9261
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
9262
      if disk_op == constants.DDM_REMOVE:
9263
        disk_addremove += 1
9264
        continue
9265
      elif disk_op == constants.DDM_ADD:
9266
        disk_addremove += 1
9267
      else:
9268
        if not isinstance(disk_op, int):
9269
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
9270
        if not isinstance(disk_dict, dict):
9271
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
9272
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9273

    
9274
      if disk_op == constants.DDM_ADD:
9275
        if "adopt" in disk_dict:
9276
          has_adopt = True
9277
        else:
9278
          has_adopt = False
9279

    
9280
        if has_adopt:
9281
          if instance.disk_template not in constants.DTS_MAY_ADOPT:
9282
            raise errors.OpPrereqError("Disk adoption is not supported for the"
9283
                                       " '%s' disk template" %
9284
                                       instance.disk_template,
9285
                                       errors.ECODE_INVAL)
9286
        self.adopt_disks = has_adopt
9287

    
9288
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
9289
        if mode not in constants.DISK_ACCESS_SET:
9290
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
9291
                                     errors.ECODE_INVAL)
9292
        if not has_adopt:
9293
          size = disk_dict.get('size', None)
9294
          if size is None:
9295
            raise errors.OpPrereqError("Required disk parameter size missing",
9296
                                       errors.ECODE_INVAL)
9297
          try:
9298
            size = int(size)
9299
          except (TypeError, ValueError), err:
9300
            raise errors.OpPrereqError("Invalid disk size parameter: %s" %
9301
                                       str(err), errors.ECODE_INVAL)
9302
          disk_dict['size'] = size
9303
      else:
9304
        # modification of disk
9305
        if 'size' in disk_dict:
9306
          raise errors.OpPrereqError("Disk size change not possible, use"
9307
                                     " grow-disk", errors.ECODE_INVAL)
9308

    
9309
    if disk_addremove > 1:
9310
      raise errors.OpPrereqError("Only one disk add or remove operation"
9311
                                 " supported at a time", errors.ECODE_INVAL)
9312

    
9313
    if self.op.disks and self.op.disk_template is not None:
9314
      raise errors.OpPrereqError("Disk template conversion and other disk"
9315
                                 " changes not supported at the same time",
9316
                                 errors.ECODE_INVAL)
9317

    
9318
    if (self.op.disk_template and
9319
        self.op.disk_template in constants.DTS_INT_MIRROR and
9320
        self.op.remote_node is None):
9321
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
9322
                                 " one requires specifying a secondary node",
9323
                                 errors.ECODE_INVAL)
9324

    
9325
    # NIC validation
9326
    nic_addremove = 0
9327
    for nic_op, nic_dict in self.op.nics:
9328
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
9329
      if nic_op == constants.DDM_REMOVE:
9330
        nic_addremove += 1
9331
        continue
9332
      elif nic_op == constants.DDM_ADD:
9333
        nic_addremove += 1
9334
      else:
9335
        if not isinstance(nic_op, int):
9336
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
9337
        if not isinstance(nic_dict, dict):
9338
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
9339
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9340

    
9341
      # nic_dict should be a dict
9342
      nic_ip = nic_dict.get('ip', None)
9343
      if nic_ip is not None:
9344
        if nic_ip.lower() == constants.VALUE_NONE:
9345
          nic_dict['ip'] = None
9346
        else:
9347
          if not netutils.IPAddress.IsValid(nic_ip):
9348
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
9349
                                       errors.ECODE_INVAL)
9350

    
9351
      nic_bridge = nic_dict.get('bridge', None)
9352
      nic_link = nic_dict.get('link', None)
9353
      if nic_bridge and nic_link:
9354
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
9355
                                   " at the same time", errors.ECODE_INVAL)
9356
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
9357
        nic_dict['bridge'] = None
9358
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
9359
        nic_dict['link'] = None
9360

    
9361
      if nic_op == constants.DDM_ADD:
9362
        nic_mac = nic_dict.get('mac', None)
9363
        if nic_mac is None:
9364
          nic_dict['mac'] = constants.VALUE_AUTO
9365

    
9366
      if 'mac' in nic_dict:
9367
        nic_mac = nic_dict['mac']
9368
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9369
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
9370

    
9371
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
9372
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
9373
                                     " modifying an existing nic",
9374
                                     errors.ECODE_INVAL)
9375

    
9376
    if nic_addremove > 1:
9377
      raise errors.OpPrereqError("Only one NIC add or remove operation"
9378
                                 " supported at a time", errors.ECODE_INVAL)
9379

    
9380
  def ExpandNames(self):
9381
    self._ExpandAndLockInstance()
9382
    self.needed_locks[locking.LEVEL_NODE] = []
9383
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9384

    
9385
  def DeclareLocks(self, level):
9386
    if level == locking.LEVEL_NODE:
9387
      self._LockInstancesNodes()
9388
      if self.op.disk_template and self.op.remote_node:
9389
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9390
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
9391

    
9392
  def BuildHooksEnv(self):
9393
    """Build hooks env.
9394

9395
    This runs on the master, primary and secondaries.
9396

9397
    """
9398
    args = dict()
9399
    if constants.BE_MEMORY in self.be_new:
9400
      args['memory'] = self.be_new[constants.BE_MEMORY]
9401
    if constants.BE_VCPUS in self.be_new:
9402
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
9403
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
9404
    # information at all.
9405
    if self.op.nics:
9406
      args['nics'] = []
9407
      nic_override = dict(self.op.nics)
9408
      for idx, nic in enumerate(self.instance.nics):
9409
        if idx in nic_override:
9410
          this_nic_override = nic_override[idx]
9411
        else:
9412
          this_nic_override = {}
9413
        if 'ip' in this_nic_override:
9414
          ip = this_nic_override['ip']
9415
        else:
9416
          ip = nic.ip
9417
        if 'mac' in this_nic_override:
9418
          mac = this_nic_override['mac']
9419
        else:
9420
          mac = nic.mac
9421
        if idx in self.nic_pnew:
9422
          nicparams = self.nic_pnew[idx]
9423
        else:
9424
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
9425
        mode = nicparams[constants.NIC_MODE]
9426
        link = nicparams[constants.NIC_LINK]
9427
        args['nics'].append((ip, mac, mode, link))
9428
      if constants.DDM_ADD in nic_override:
9429
        ip = nic_override[constants.DDM_ADD].get('ip', None)
9430
        mac = nic_override[constants.DDM_ADD]['mac']
9431
        nicparams = self.nic_pnew[constants.DDM_ADD]
9432
        mode = nicparams[constants.NIC_MODE]
9433
        link = nicparams[constants.NIC_LINK]
9434
        args['nics'].append((ip, mac, mode, link))
9435
      elif constants.DDM_REMOVE in nic_override:
9436
        del args['nics'][-1]
9437

    
9438
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
9439
    if self.op.disk_template:
9440
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
9441
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
9442
    return env, nl, nl
9443

    
9444
  def CheckPrereq(self):
9445
    """Check prerequisites.
9446

9447
    This only checks the instance list against the existing names.
9448

9449
    """
9450
    # checking the new params on the primary/secondary nodes
9451

    
9452
    instance = self.instance
9453
    cluster = self.cluster = self.cfg.GetClusterInfo()
9454
    assert self.instance is not None, \
9455
      "Cannot retrieve locked instance %s" % self.op.instance_name
9456
    pnode = instance.primary_node
9457
    nodelist = list(instance.all_nodes)
9458

    
9459
    # OS change
9460
    if self.op.os_name and not self.op.force:
9461
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
9462
                      self.op.force_variant)
9463
      instance_os = self.op.os_name
9464
    else:
9465
      instance_os = instance.os
9466

    
9467
    if self.op.disk_template:
9468
      if instance.disk_template == self.op.disk_template:
9469
        raise errors.OpPrereqError("Instance already has disk template %s" %
9470
                                   instance.disk_template, errors.ECODE_INVAL)
9471

    
9472
      if (instance.disk_template,
9473
          self.op.disk_template) not in self._DISK_CONVERSIONS:
9474
        raise errors.OpPrereqError("Unsupported disk template conversion from"
9475
                                   " %s to %s" % (instance.disk_template,
9476
                                                  self.op.disk_template),
9477
                                   errors.ECODE_INVAL)
9478
      _CheckInstanceDown(self, instance, "cannot change disk template")
9479
      if self.op.disk_template in constants.DTS_INT_MIRROR:
9480
        if self.op.remote_node == pnode:
9481
          raise errors.OpPrereqError("Given new secondary node %s is the same"
9482
                                     " as the primary node of the instance" %
9483
                                     self.op.remote_node, errors.ECODE_STATE)
9484
        _CheckNodeOnline(self, self.op.remote_node)
9485
        _CheckNodeNotDrained(self, self.op.remote_node)
9486
        # FIXME: here we assume that the old instance type is DT_PLAIN
9487
        assert instance.disk_template == constants.DT_PLAIN
9488
        disks = [{"size": d.size, "vg": d.logical_id[0]}
9489
                 for d in instance.disks]
9490
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
9491
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
9492

    
9493
    # hvparams processing
9494
    if self.op.hvparams:
9495
      hv_type = instance.hypervisor
9496
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
9497
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
9498
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
9499

    
9500
      # local check
9501
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
9502
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
9503
      self.hv_new = hv_new # the new actual values
9504
      self.hv_inst = i_hvdict # the new dict (without defaults)
9505
    else:
9506
      self.hv_new = self.hv_inst = {}
9507

    
9508
    # beparams processing
9509
    if self.op.beparams:
9510
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
9511
                                   use_none=True)
9512
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
9513
      be_new = cluster.SimpleFillBE(i_bedict)
9514
      self.be_new = be_new # the new actual values
9515
      self.be_inst = i_bedict # the new dict (without defaults)
9516
    else:
9517
      self.be_new = self.be_inst = {}
9518

    
9519
    # osparams processing
9520
    if self.op.osparams:
9521
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
9522
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
9523
      self.os_inst = i_osdict # the new dict (without defaults)
9524
    else:
9525
      self.os_inst = {}
9526

    
9527
    self.warn = []
9528

    
9529
    if constants.BE_MEMORY in self.op.beparams and not self.op.force:
9530
      mem_check_list = [pnode]
9531
      if be_new[constants.BE_AUTO_BALANCE]:
9532
        # either we changed auto_balance to yes or it was from before
9533
        mem_check_list.extend(instance.secondary_nodes)
9534
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
9535
                                                  instance.hypervisor)
9536
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
9537
                                         instance.hypervisor)
9538
      pninfo = nodeinfo[pnode]
9539
      msg = pninfo.fail_msg
9540
      if msg:
9541
        # Assume the primary node is unreachable and go ahead
9542
        self.warn.append("Can't get info from primary node %s: %s" %
9543
                         (pnode,  msg))
9544
      elif not isinstance(pninfo.payload.get('memory_free', None), int):
9545
        self.warn.append("Node data from primary node %s doesn't contain"
9546
                         " free memory information" % pnode)
9547
      elif instance_info.fail_msg:
9548
        self.warn.append("Can't get instance runtime information: %s" %
9549
                        instance_info.fail_msg)
9550
      else:
9551
        if instance_info.payload:
9552
          current_mem = int(instance_info.payload['memory'])
9553
        else:
9554
          # Assume instance not running
9555
          # (there is a slight race condition here, but it's not very probable,
9556
          # and we have no other way to check)
9557
          current_mem = 0
9558
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
9559
                    pninfo.payload['memory_free'])
9560
        if miss_mem > 0:
9561
          raise errors.OpPrereqError("This change will prevent the instance"
9562
                                     " from starting, due to %d MB of memory"
9563
                                     " missing on its primary node" % miss_mem,
9564
                                     errors.ECODE_NORES)
9565

    
9566
      if be_new[constants.BE_AUTO_BALANCE]:
9567
        for node, nres in nodeinfo.items():
9568
          if node not in instance.secondary_nodes:
9569
            continue
9570
          msg = nres.fail_msg
9571
          if msg:
9572
            self.warn.append("Can't get info from secondary node %s: %s" %
9573
                             (node, msg))
9574
          elif not isinstance(nres.payload.get('memory_free', None), int):
9575
            self.warn.append("Secondary node %s didn't return free"
9576
                             " memory information" % node)
9577
          elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
9578
            self.warn.append("Not enough memory to failover instance to"
9579
                             " secondary node %s" % node)
9580

    
9581
    # NIC processing
9582
    self.nic_pnew = {}
9583
    self.nic_pinst = {}
9584
    for nic_op, nic_dict in self.op.nics:
9585
      if nic_op == constants.DDM_REMOVE:
9586
        if not instance.nics:
9587
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
9588
                                     errors.ECODE_INVAL)
9589
        continue
9590
      if nic_op != constants.DDM_ADD:
9591
        # an existing nic
9592
        if not instance.nics:
9593
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
9594
                                     " no NICs" % nic_op,
9595
                                     errors.ECODE_INVAL)
9596
        if nic_op < 0 or nic_op >= len(instance.nics):
9597
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
9598
                                     " are 0 to %d" %
9599
                                     (nic_op, len(instance.nics) - 1),
9600
                                     errors.ECODE_INVAL)
9601
        old_nic_params = instance.nics[nic_op].nicparams
9602
        old_nic_ip = instance.nics[nic_op].ip
9603
      else:
9604
        old_nic_params = {}
9605
        old_nic_ip = None
9606

    
9607
      update_params_dict = dict([(key, nic_dict[key])
9608
                                 for key in constants.NICS_PARAMETERS
9609
                                 if key in nic_dict])
9610

    
9611
      if 'bridge' in nic_dict:
9612
        update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
9613

    
9614
      new_nic_params = _GetUpdatedParams(old_nic_params,
9615
                                         update_params_dict)
9616
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
9617
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
9618
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
9619
      self.nic_pinst[nic_op] = new_nic_params
9620
      self.nic_pnew[nic_op] = new_filled_nic_params
9621
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
9622

    
9623
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
9624
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
9625
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
9626
        if msg:
9627
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
9628
          if self.op.force:
9629
            self.warn.append(msg)
9630
          else:
9631
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
9632
      if new_nic_mode == constants.NIC_MODE_ROUTED:
9633
        if 'ip' in nic_dict:
9634
          nic_ip = nic_dict['ip']
9635
        else:
9636
          nic_ip = old_nic_ip
9637
        if nic_ip is None:
9638
          raise errors.OpPrereqError('Cannot set the nic ip to None'
9639
                                     ' on a routed nic', errors.ECODE_INVAL)
9640
      if 'mac' in nic_dict:
9641
        nic_mac = nic_dict['mac']
9642
        if nic_mac is None:
9643
          raise errors.OpPrereqError('Cannot set the nic mac to None',
9644
                                     errors.ECODE_INVAL)
9645
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9646
          # otherwise generate the mac
9647
          nic_dict['mac'] = self.cfg.GenerateMAC(self.proc.GetECId())
9648
        else:
9649
          # or validate/reserve the current one
9650
          try:
9651
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
9652
          except errors.ReservationError:
9653
            raise errors.OpPrereqError("MAC address %s already in use"
9654
                                       " in cluster" % nic_mac,
9655
                                       errors.ECODE_NOTUNIQUE)
9656

    
9657
    # DISK processing
9658
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
9659
      raise errors.OpPrereqError("Disk operations not supported for"
9660
                                 " diskless instances",
9661
                                 errors.ECODE_INVAL)
9662
    for disk_op, disk_dict in self.op.disks:
9663
      if disk_op == constants.DDM_REMOVE:
9664
        if len(instance.disks) == 1:
9665
          raise errors.OpPrereqError("Cannot remove the last disk of"
9666
                                     " an instance", errors.ECODE_INVAL)
9667
        _CheckInstanceDown(self, instance, "cannot remove disks")
9668

    
9669
      if (disk_op == constants.DDM_ADD and
9670
          len(instance.disks) >= constants.MAX_DISKS):
9671
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
9672
                                   " add more" % constants.MAX_DISKS,
9673
                                   errors.ECODE_STATE)
9674
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
9675
        # an existing disk
9676
        if disk_op < 0 or disk_op >= len(instance.disks):
9677
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
9678
                                     " are 0 to %d" %
9679
                                     (disk_op, len(instance.disks)),
9680
                                     errors.ECODE_INVAL)
9681

    
9682
      if disk_op == constants.DDM_ADD and self.adopt_disks:
9683
        if instance.disk_template == constants.DT_PLAIN:
9684
          # Check the adoption data
9685
          vg = disk_dict.get("vg", self.cfg.GetVGName())
9686
          lv_name = vg + "/" + disk_dict["adopt"]
9687
          try:
9688
            # FIXME: lv_name here is "vg/lv" need to ensure that other calls
9689
            # to ReserveLV uses the same syntax
9690
            self.cfg.ReserveLV(lv_name, self.proc.GetECId())
9691
          except errors.ReservationError:
9692
            raise errors.OpPrereqError("LV named %s used by another instance" %
9693
                                       lv_name, errors.ECODE_NOTUNIQUE)
9694

    
9695
          vg_names = self.rpc.call_vg_list([instance.primary_node])
9696
          vg_names = vg_names[instance.primary_node]
9697
          vg_names.Raise("Cannot get VG information from node %s" %
9698
                         instance.primary_node)
9699

    
9700
          node_lvs = self.rpc.call_lv_list([instance.primary_node],
9701
                                           vg_names.payload.keys()
9702
                                          )[instance.primary_node]
9703
          node_lvs.Raise("Cannot get LV information from node %s" %
9704
                         instance.primary_node)
9705
          node_lvs = node_lvs.payload
9706

    
9707
          if lv_name not in node_lvs:
9708
            raise errors.OpPrereqError("Logical volume: %s not present "
9709
                                       " on node %s" %
9710
                                       (lv_name, instance.primary_node),
9711
                                       errors.ECODE_INVAL)
9712
          if node_lvs[lv_name][2]:
9713
            raise errors.OpPrereqError("Logical volume %s is online, cannot"
9714
                                       " adopt." % lv_name,
9715
                                       errors.ECODE_STATE)
9716
          # update the size of disk based on what is found
9717
          disk_dict["size"] = int(float(node_lvs[lv_name][0]))
9718

    
9719
        elif instance.disk_template == constants.DT_BLOCK:
9720
          disk = disk_dict["adopt"]
9721
          node_disks = self.rpc.call_bdev_sizes([instance.primary_node],
9722
                                                [disk])[instance.primary_node]
9723
          node_disks = node_disks.payload
9724
          if disk not in node_disks:
9725
            raise errors.OpPrereqError("Missing block device %s" % disk,
9726
                                       errors.ECODE_INVAL)
9727
          disk_dict["size"] = int(float(node_disks[disk]))
9728

    
9729
    return
9730

    
9731
  def _ConvertPlainToDrbd(self, feedback_fn):
9732
    """Converts an instance from plain to drbd.
9733

9734
    """
9735
    feedback_fn("Converting template to drbd")
9736
    instance = self.instance
9737
    pnode = instance.primary_node
9738
    snode = self.op.remote_node
9739

    
9740
    # create a fake disk info for _GenerateDiskTemplate
9741
    disk_info = [{"size": d.size, "mode": d.mode} for d in instance.disks]
9742
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
9743
                                      instance.name, pnode, [snode],
9744
                                      disk_info, None, None, 0, feedback_fn)
9745
    info = _GetInstanceInfoText(instance)
9746
    feedback_fn("Creating aditional volumes...")
9747
    # first, create the missing data and meta devices
9748
    for disk in new_disks:
9749
      # unfortunately this is... not too nice
9750
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
9751
                            info, True)
9752
      for child in disk.children:
9753
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
9754
    # at this stage, all new LVs have been created, we can rename the
9755
    # old ones
9756
    feedback_fn("Renaming original volumes...")
9757
    rename_list = [(o, n.children[0].logical_id)
9758
                   for (o, n) in zip(instance.disks, new_disks)]
9759
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
9760
    result.Raise("Failed to rename original LVs")
9761

    
9762
    feedback_fn("Initializing DRBD devices...")
9763
    # all child devices are in place, we can now create the DRBD devices
9764
    for disk in new_disks:
9765
      for node in [pnode, snode]:
9766
        f_create = node == pnode
9767
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
9768

    
9769
    # at this point, the instance has been modified
9770
    instance.disk_template = constants.DT_DRBD8
9771
    instance.disks = new_disks
9772
    self.cfg.Update(instance, feedback_fn)
9773

    
9774
    # disks are created, waiting for sync
9775
    disk_abort = not _WaitForSync(self, instance)
9776
    if disk_abort:
9777
      raise errors.OpExecError("There are some degraded disks for"
9778
                               " this instance, please cleanup manually")
9779

    
9780
  def _ConvertDrbdToPlain(self, feedback_fn):
9781
    """Converts an instance from drbd to plain.
9782

9783
    """
9784
    instance = self.instance
9785
    assert len(instance.secondary_nodes) == 1
9786
    pnode = instance.primary_node
9787
    snode = instance.secondary_nodes[0]
9788
    feedback_fn("Converting template to plain")
9789

    
9790
    old_disks = instance.disks
9791
    new_disks = [d.children[0] for d in old_disks]
9792

    
9793
    # copy over size and mode
9794
    for parent, child in zip(old_disks, new_disks):
9795
      child.size = parent.size
9796
      child.mode = parent.mode
9797

    
9798
    # update instance structure
9799
    instance.disks = new_disks
9800
    instance.disk_template = constants.DT_PLAIN
9801
    self.cfg.Update(instance, feedback_fn)
9802

    
9803
    feedback_fn("Removing volumes on the secondary node...")
9804
    for disk in old_disks:
9805
      self.cfg.SetDiskID(disk, snode)
9806
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
9807
      if msg:
9808
        self.LogWarning("Could not remove block device %s on node %s,"
9809
                        " continuing anyway: %s", disk.iv_name, snode, msg)
9810

    
9811
    feedback_fn("Removing unneeded volumes on the primary node...")
9812
    for idx, disk in enumerate(old_disks):
9813
      meta = disk.children[1]
9814
      self.cfg.SetDiskID(meta, pnode)
9815
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
9816
      if msg:
9817
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
9818
                        " continuing anyway: %s", idx, pnode, msg)
9819

    
9820
  def Exec(self, feedback_fn):
9821
    """Modifies an instance.
9822

9823
    All parameters take effect only at the next restart of the instance.
9824

9825
    """
9826
    # Process here the warnings from CheckPrereq, as we don't have a
9827
    # feedback_fn there.
9828
    for warn in self.warn:
9829
      feedback_fn("WARNING: %s" % warn)
9830

    
9831
    result = []
9832
    instance = self.instance
9833
    # disk changes
9834
    for disk_op, disk_dict in self.op.disks:
9835
      if disk_op == constants.DDM_REMOVE:
9836
        # remove the last disk
9837
        device = instance.disks.pop()
9838
        device_idx = len(instance.disks)
9839
        for node, disk in device.ComputeNodeTree(instance.primary_node):
9840
          self.cfg.SetDiskID(disk, node)
9841
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
9842
          if msg:
9843
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
9844
                            " continuing anyway", device_idx, node, msg)
9845
        result.append(("disk/%d" % device_idx, "remove"))
9846
      elif disk_op == constants.DDM_ADD:
9847
        # add a new disk
9848
        if instance.disk_template in (constants.DT_FILE,
9849
                                        constants.DT_SHARED_FILE):
9850
          file_driver, file_path = instance.disks[0].logical_id
9851
          file_path = os.path.dirname(file_path)
9852
        else:
9853
          file_driver = file_path = None
9854
        disk_idx_base = len(instance.disks)
9855
        new_disk = _GenerateDiskTemplate(self,
9856
                                         instance.disk_template,
9857
                                         instance.name, instance.primary_node,
9858
                                         instance.secondary_nodes,
9859
                                         [disk_dict],
9860
                                         file_path,
9861
                                         file_driver,
9862
                                         disk_idx_base, feedback_fn)[0]
9863
        instance.disks.append(new_disk)
9864
        info = _GetInstanceInfoText(instance)
9865

    
9866
        if self.adopt_disks:
9867
          if instance.disk_template == constants.DT_PLAIN:
9868
            # rename LV to the a newly-generated name; we need to construct
9869
            # 'fake' LV disk with the old data, plus the new unique_id
9870
            tmp_disk = objects.Disk.FromDict(disk_dict)
9871
            rename_to = tmp_disk.logical_id
9872
            tmp_disk.logical_id = (tmp_disk.logical_id[0], disk_dict["adopt"])
9873
            self.cfg.SetDiskID(tmp_disk, instance.primary_node)
9874
            result = self.rpc.call_blockdev_rename(instance.primary_node,
9875
                                                   [(tmp_disk, rename_to)])
9876
            result.Raise("Failed to rename adopted LV")
9877
          result.append(("disk/%d" % disk_idx_base, "add:adopt=%s,mode=%s" %
9878
                         (disk_dict["adopt"], new_disk.mode)))
9879
        else:
9880
          logging.info("Creating volume %s for instance %s",
9881
                       new_disk.iv_name, instance.name)
9882
          # Note: this needs to be kept in sync with _CreateDisks
9883
          #HARDCODE
9884
          for node in instance.all_nodes:
9885
            f_create = node == instance.primary_node
9886
            try:
9887
              _CreateBlockDev(self, node, instance, new_disk,
9888
                              f_create, info, f_create)
9889
            except errors.OpExecError, err:
9890
              self.LogWarning("Failed to create volume %s (%s) on"
9891
                              " node %s: %s",
9892
                              new_disk.iv_name, new_disk, node, err)
9893
          result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
9894
                         (new_disk.size, new_disk.mode)))
9895
      else:
9896
        # change a given disk
9897
        instance.disks[disk_op].mode = disk_dict['mode']
9898
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
9899

    
9900
    if self.op.disk_template:
9901
      r_shut = _ShutdownInstanceDisks(self, instance)
9902
      if not r_shut:
9903
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
9904
                                 " proceed with disk template conversion")
9905
      mode = (instance.disk_template, self.op.disk_template)
9906
      try:
9907
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
9908
      except:
9909
        self.cfg.ReleaseDRBDMinors(instance.name)
9910
        raise
9911
      result.append(("disk_template", self.op.disk_template))
9912

    
9913
    # NIC changes
9914
    for nic_op, nic_dict in self.op.nics:
9915
      if nic_op == constants.DDM_REMOVE:
9916
        # remove the last nic
9917
        del instance.nics[-1]
9918
        result.append(("nic.%d" % len(instance.nics), "remove"))
9919
      elif nic_op == constants.DDM_ADD:
9920
        # mac and bridge should be set, by now
9921
        mac = nic_dict['mac']
9922
        ip = nic_dict.get('ip', None)
9923
        nicparams = self.nic_pinst[constants.DDM_ADD]
9924
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
9925
        instance.nics.append(new_nic)
9926
        result.append(("nic.%d" % (len(instance.nics) - 1),
9927
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
9928
                       (new_nic.mac, new_nic.ip,
9929
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
9930
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
9931
                       )))
9932
      else:
9933
        for key in 'mac', 'ip':
9934
          if key in nic_dict:
9935
            setattr(instance.nics[nic_op], key, nic_dict[key])
9936
        if nic_op in self.nic_pinst:
9937
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
9938
        for key, val in nic_dict.iteritems():
9939
          result.append(("nic.%s/%d" % (key, nic_op), val))
9940

    
9941
    # hvparams changes
9942
    if self.op.hvparams:
9943
      instance.hvparams = self.hv_inst
9944
      for key, val in self.op.hvparams.iteritems():
9945
        result.append(("hv/%s" % key, val))
9946

    
9947
    # beparams changes
9948
    if self.op.beparams:
9949
      instance.beparams = self.be_inst
9950
      for key, val in self.op.beparams.iteritems():
9951
        result.append(("be/%s" % key, val))
9952

    
9953
    # OS change
9954
    if self.op.os_name:
9955
      instance.os = self.op.os_name
9956

    
9957
    # osparams changes
9958
    if self.op.osparams:
9959
      instance.osparams = self.os_inst
9960
      for key, val in self.op.osparams.iteritems():
9961
        result.append(("os/%s" % key, val))
9962

    
9963
    self.cfg.Update(instance, feedback_fn)
9964

    
9965
    return result
9966

    
9967
  _DISK_CONVERSIONS = {
9968
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
9969
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
9970
    }
9971

    
9972

    
9973
class LUBackupQuery(NoHooksLU):
9974
  """Query the exports list
9975

9976
  """
9977
  REQ_BGL = False
9978

    
9979
  def ExpandNames(self):
9980
    self.needed_locks = {}
9981
    self.share_locks[locking.LEVEL_NODE] = 1
9982
    if not self.op.nodes:
9983
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9984
    else:
9985
      self.needed_locks[locking.LEVEL_NODE] = \
9986
        _GetWantedNodes(self, self.op.nodes)
9987

    
9988
  def Exec(self, feedback_fn):
9989
    """Compute the list of all the exported system images.
9990

9991
    @rtype: dict
9992
    @return: a dictionary with the structure node->(export-list)
9993
        where export-list is a list of the instances exported on
9994
        that node.
9995

9996
    """
9997
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
9998
    rpcresult = self.rpc.call_export_list(self.nodes)
9999
    result = {}
10000
    for node in rpcresult:
10001
      if rpcresult[node].fail_msg:
10002
        result[node] = False
10003
      else:
10004
        result[node] = rpcresult[node].payload
10005

    
10006
    return result
10007

    
10008

    
10009
class LUBackupPrepare(NoHooksLU):
10010
  """Prepares an instance for an export and returns useful information.
10011

10012
  """
10013
  REQ_BGL = False
10014

    
10015
  def ExpandNames(self):
10016
    self._ExpandAndLockInstance()
10017

    
10018
  def CheckPrereq(self):
10019
    """Check prerequisites.
10020

10021
    """
10022
    instance_name = self.op.instance_name
10023

    
10024
    self.instance = self.cfg.GetInstanceInfo(instance_name)
10025
    assert self.instance is not None, \
10026
          "Cannot retrieve locked instance %s" % self.op.instance_name
10027
    _CheckNodeOnline(self, self.instance.primary_node)
10028

    
10029
    self._cds = _GetClusterDomainSecret()
10030

    
10031
  def Exec(self, feedback_fn):
10032
    """Prepares an instance for an export.
10033

10034
    """
10035
    instance = self.instance
10036

    
10037
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
10038
      salt = utils.GenerateSecret(8)
10039

    
10040
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
10041
      result = self.rpc.call_x509_cert_create(instance.primary_node,
10042
                                              constants.RIE_CERT_VALIDITY)
10043
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
10044

    
10045
      (name, cert_pem) = result.payload
10046

    
10047
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
10048
                                             cert_pem)
10049

    
10050
      return {
10051
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
10052
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
10053
                          salt),
10054
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
10055
        }
10056

    
10057
    return None
10058

    
10059

    
10060
class LUBackupExport(LogicalUnit):
10061
  """Export an instance to an image in the cluster.
10062

10063
  """
10064
  HPATH = "instance-export"
10065
  HTYPE = constants.HTYPE_INSTANCE
10066
  REQ_BGL = False
10067

    
10068
  def CheckArguments(self):
10069
    """Check the arguments.
10070

10071
    """
10072
    self.x509_key_name = self.op.x509_key_name
10073
    self.dest_x509_ca_pem = self.op.destination_x509_ca
10074

    
10075
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
10076
      if not self.x509_key_name:
10077
        raise errors.OpPrereqError("Missing X509 key name for encryption",
10078
                                   errors.ECODE_INVAL)
10079

    
10080
      if not self.dest_x509_ca_pem:
10081
        raise errors.OpPrereqError("Missing destination X509 CA",
10082
                                   errors.ECODE_INVAL)
10083

    
10084
  def ExpandNames(self):
10085
    self._ExpandAndLockInstance()
10086

    
10087
    # Lock all nodes for local exports
10088
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10089
      # FIXME: lock only instance primary and destination node
10090
      #
10091
      # Sad but true, for now we have do lock all nodes, as we don't know where
10092
      # the previous export might be, and in this LU we search for it and
10093
      # remove it from its current node. In the future we could fix this by:
10094
      #  - making a tasklet to search (share-lock all), then create the
10095
      #    new one, then one to remove, after
10096
      #  - removing the removal operation altogether
10097
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10098

    
10099
  def DeclareLocks(self, level):
10100
    """Last minute lock declaration."""
10101
    # All nodes are locked anyway, so nothing to do here.
10102

    
10103
  def BuildHooksEnv(self):
10104
    """Build hooks env.
10105

10106
    This will run on the master, primary node and target node.
10107

10108
    """
10109
    env = {
10110
      "EXPORT_MODE": self.op.mode,
10111
      "EXPORT_NODE": self.op.target_node,
10112
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
10113
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
10114
      # TODO: Generic function for boolean env variables
10115
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
10116
      }
10117

    
10118
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
10119

    
10120
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
10121

    
10122
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10123
      nl.append(self.op.target_node)
10124

    
10125
    return env, nl, nl
10126

    
10127
  def CheckPrereq(self):
10128
    """Check prerequisites.
10129

10130
    This checks that the instance and node names are valid.
10131

10132
    """
10133
    instance_name = self.op.instance_name
10134

    
10135
    self.instance = self.cfg.GetInstanceInfo(instance_name)
10136
    assert self.instance is not None, \
10137
          "Cannot retrieve locked instance %s" % self.op.instance_name
10138
    _CheckNodeOnline(self, self.instance.primary_node)
10139

    
10140
    if (self.op.remove_instance and self.instance.admin_up and
10141
        not self.op.shutdown):
10142
      raise errors.OpPrereqError("Can not remove instance without shutting it"
10143
                                 " down before")
10144

    
10145
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10146
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
10147
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
10148
      assert self.dst_node is not None
10149

    
10150
      _CheckNodeOnline(self, self.dst_node.name)
10151
      _CheckNodeNotDrained(self, self.dst_node.name)
10152

    
10153
      self._cds = None
10154
      self.dest_disk_info = None
10155
      self.dest_x509_ca = None
10156

    
10157
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10158
      self.dst_node = None
10159

    
10160
      if len(self.op.target_node) != len(self.instance.disks):
10161
        raise errors.OpPrereqError(("Received destination information for %s"
10162
                                    " disks, but instance %s has %s disks") %
10163
                                   (len(self.op.target_node), instance_name,
10164
                                    len(self.instance.disks)),
10165
                                   errors.ECODE_INVAL)
10166

    
10167
      cds = _GetClusterDomainSecret()
10168

    
10169
      # Check X509 key name
10170
      try:
10171
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
10172
      except (TypeError, ValueError), err:
10173
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
10174

    
10175
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
10176
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
10177
                                   errors.ECODE_INVAL)
10178

    
10179
      # Load and verify CA
10180
      try:
10181
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
10182
      except OpenSSL.crypto.Error, err:
10183
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
10184
                                   (err, ), errors.ECODE_INVAL)
10185

    
10186
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
10187
      if errcode is not None:
10188
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
10189
                                   (msg, ), errors.ECODE_INVAL)
10190

    
10191
      self.dest_x509_ca = cert
10192

    
10193
      # Verify target information
10194
      disk_info = []
10195
      for idx, disk_data in enumerate(self.op.target_node):
10196
        try:
10197
          (host, port, magic) = \
10198
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
10199
        except errors.GenericError, err:
10200
          raise errors.OpPrereqError("Target info for disk %s: %s" %
10201
                                     (idx, err), errors.ECODE_INVAL)
10202

    
10203
        disk_info.append((host, port, magic))
10204

    
10205
      assert len(disk_info) == len(self.op.target_node)
10206
      self.dest_disk_info = disk_info
10207

    
10208
    else:
10209
      raise errors.ProgrammerError("Unhandled export mode %r" %
10210
                                   self.op.mode)
10211

    
10212
    # instance disk type verification
10213
    # TODO: Implement export support for file-based disks
10214
    for disk in self.instance.disks:
10215
      if disk.dev_type == constants.LD_FILE:
10216
        raise errors.OpPrereqError("Export not supported for instances with"
10217
                                   " file-based disks", errors.ECODE_INVAL)
10218

    
10219
  def _CleanupExports(self, feedback_fn):
10220
    """Removes exports of current instance from all other nodes.
10221

10222
    If an instance in a cluster with nodes A..D was exported to node C, its
10223
    exports will be removed from the nodes A, B and D.
10224

10225
    """
10226
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
10227

    
10228
    nodelist = self.cfg.GetNodeList()
10229
    nodelist.remove(self.dst_node.name)
10230

    
10231
    # on one-node clusters nodelist will be empty after the removal
10232
    # if we proceed the backup would be removed because OpBackupQuery
10233
    # substitutes an empty list with the full cluster node list.
10234
    iname = self.instance.name
10235
    if nodelist:
10236
      feedback_fn("Removing old exports for instance %s" % iname)
10237
      exportlist = self.rpc.call_export_list(nodelist)
10238
      for node in exportlist:
10239
        if exportlist[node].fail_msg:
10240
          continue
10241
        if iname in exportlist[node].payload:
10242
          msg = self.rpc.call_export_remove(node, iname).fail_msg
10243
          if msg:
10244
            self.LogWarning("Could not remove older export for instance %s"
10245
                            " on node %s: %s", iname, node, msg)
10246

    
10247
  def Exec(self, feedback_fn):
10248
    """Export an instance to an image in the cluster.
10249

10250
    """
10251
    assert self.op.mode in constants.EXPORT_MODES
10252

    
10253
    instance = self.instance
10254
    src_node = instance.primary_node
10255

    
10256
    if self.op.shutdown:
10257
      # shutdown the instance, but not the disks
10258
      feedback_fn("Shutting down instance %s" % instance.name)
10259
      result = self.rpc.call_instance_shutdown(src_node, instance,
10260
                                               self.op.shutdown_timeout)
10261
      # TODO: Maybe ignore failures if ignore_remove_failures is set
10262
      result.Raise("Could not shutdown instance %s on"
10263
                   " node %s" % (instance.name, src_node))
10264

    
10265
    # set the disks ID correctly since call_instance_start needs the
10266
    # correct drbd minor to create the symlinks
10267
    for disk in instance.disks:
10268
      self.cfg.SetDiskID(disk, src_node)
10269

    
10270
    activate_disks = (not instance.admin_up)
10271

    
10272
    if activate_disks:
10273
      # Activate the instance disks if we'exporting a stopped instance
10274
      feedback_fn("Activating disks for %s" % instance.name)
10275
      _StartInstanceDisks(self, instance, None)
10276

    
10277
    try:
10278
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
10279
                                                     instance)
10280

    
10281
      helper.CreateSnapshots()
10282
      try:
10283
        if (self.op.shutdown and instance.admin_up and
10284
            not self.op.remove_instance):
10285
          assert not activate_disks
10286
          feedback_fn("Starting instance %s" % instance.name)
10287
          result = self.rpc.call_instance_start(src_node, instance, None, None)
10288
          msg = result.fail_msg
10289
          if msg:
10290
            feedback_fn("Failed to start instance: %s" % msg)
10291
            _ShutdownInstanceDisks(self, instance)
10292
            raise errors.OpExecError("Could not start instance: %s" % msg)
10293

    
10294
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
10295
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
10296
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10297
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
10298
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
10299

    
10300
          (key_name, _, _) = self.x509_key_name
10301

    
10302
          dest_ca_pem = \
10303
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
10304
                                            self.dest_x509_ca)
10305

    
10306
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
10307
                                                     key_name, dest_ca_pem,
10308
                                                     timeouts)
10309
      finally:
10310
        helper.Cleanup()
10311

    
10312
      # Check for backwards compatibility
10313
      assert len(dresults) == len(instance.disks)
10314
      assert compat.all(isinstance(i, bool) for i in dresults), \
10315
             "Not all results are boolean: %r" % dresults
10316

    
10317
    finally:
10318
      if activate_disks:
10319
        feedback_fn("Deactivating disks for %s" % instance.name)
10320
        _ShutdownInstanceDisks(self, instance)
10321

    
10322
    if not (compat.all(dresults) and fin_resu):
10323
      failures = []
10324
      if not fin_resu:
10325
        failures.append("export finalization")
10326
      if not compat.all(dresults):
10327
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
10328
                               if not dsk)
10329
        failures.append("disk export: disk(s) %s" % fdsk)
10330

    
10331
      raise errors.OpExecError("Export failed, errors in %s" %
10332
                               utils.CommaJoin(failures))
10333

    
10334
    # At this point, the export was successful, we can cleanup/finish
10335

    
10336
    # Remove instance if requested
10337
    if self.op.remove_instance:
10338
      feedback_fn("Removing instance %s" % instance.name)
10339
      _RemoveInstance(self, feedback_fn, instance,
10340
                      self.op.ignore_remove_failures)
10341

    
10342
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10343
      self._CleanupExports(feedback_fn)
10344

    
10345
    return fin_resu, dresults
10346

    
10347

    
10348
class LUBackupRemove(NoHooksLU):
10349
  """Remove exports related to the named instance.
10350

10351
  """
10352
  REQ_BGL = False
10353

    
10354
  def ExpandNames(self):
10355
    self.needed_locks = {}
10356
    # We need all nodes to be locked in order for RemoveExport to work, but we
10357
    # don't need to lock the instance itself, as nothing will happen to it (and
10358
    # we can remove exports also for a removed instance)
10359
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10360

    
10361
  def Exec(self, feedback_fn):
10362
    """Remove any export.
10363

10364
    """
10365
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
10366
    # If the instance was not found we'll try with the name that was passed in.
10367
    # This will only work if it was an FQDN, though.
10368
    fqdn_warn = False
10369
    if not instance_name:
10370
      fqdn_warn = True
10371
      instance_name = self.op.instance_name
10372

    
10373
    locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
10374
    exportlist = self.rpc.call_export_list(locked_nodes)
10375
    found = False
10376
    for node in exportlist:
10377
      msg = exportlist[node].fail_msg
10378
      if msg:
10379
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
10380
        continue
10381
      if instance_name in exportlist[node].payload:
10382
        found = True
10383
        result = self.rpc.call_export_remove(node, instance_name)
10384
        msg = result.fail_msg
10385
        if msg:
10386
          logging.error("Could not remove export for instance %s"
10387
                        " on node %s: %s", instance_name, node, msg)
10388

    
10389
    if fqdn_warn and not found:
10390
      feedback_fn("Export not found. If trying to remove an export belonging"
10391
                  " to a deleted instance please use its Fully Qualified"
10392
                  " Domain Name.")
10393

    
10394

    
10395
class LUGroupAdd(LogicalUnit):
10396
  """Logical unit for creating node groups.
10397

10398
  """
10399
  HPATH = "group-add"
10400
  HTYPE = constants.HTYPE_GROUP
10401
  REQ_BGL = False
10402

    
10403
  def ExpandNames(self):
10404
    # We need the new group's UUID here so that we can create and acquire the
10405
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
10406
    # that it should not check whether the UUID exists in the configuration.
10407
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
10408
    self.needed_locks = {}
10409
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10410

    
10411
  def CheckPrereq(self):
10412
    """Check prerequisites.
10413

10414
    This checks that the given group name is not an existing node group
10415
    already.
10416

10417
    """
10418
    try:
10419
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10420
    except errors.OpPrereqError:
10421
      pass
10422
    else:
10423
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
10424
                                 " node group (UUID: %s)" %
10425
                                 (self.op.group_name, existing_uuid),
10426
                                 errors.ECODE_EXISTS)
10427

    
10428
    if self.op.ndparams:
10429
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10430

    
10431
  def BuildHooksEnv(self):
10432
    """Build hooks env.
10433

10434
    """
10435
    env = {
10436
      "GROUP_NAME": self.op.group_name,
10437
      }
10438
    mn = self.cfg.GetMasterNode()
10439
    return env, [mn], [mn]
10440

    
10441
  def Exec(self, feedback_fn):
10442
    """Add the node group to the cluster.
10443

10444
    """
10445
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
10446
                                  uuid=self.group_uuid,
10447
                                  alloc_policy=self.op.alloc_policy,
10448
                                  ndparams=self.op.ndparams)
10449

    
10450
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
10451
    del self.remove_locks[locking.LEVEL_NODEGROUP]
10452

    
10453

    
10454
class LUGroupAssignNodes(NoHooksLU):
10455
  """Logical unit for assigning nodes to groups.
10456

10457
  """
10458
  REQ_BGL = False
10459

    
10460
  def ExpandNames(self):
10461
    # These raise errors.OpPrereqError on their own:
10462
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10463
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
10464

    
10465
    # We want to lock all the affected nodes and groups. We have readily
10466
    # available the list of nodes, and the *destination* group. To gather the
10467
    # list of "source" groups, we need to fetch node information.
10468
    self.node_data = self.cfg.GetAllNodesInfo()
10469
    affected_groups = set(self.node_data[node].group for node in self.op.nodes)
10470
    affected_groups.add(self.group_uuid)
10471

    
10472
    self.needed_locks = {
10473
      locking.LEVEL_NODEGROUP: list(affected_groups),
10474
      locking.LEVEL_NODE: self.op.nodes,
10475
      }
10476

    
10477
  def CheckPrereq(self):
10478
    """Check prerequisites.
10479

10480
    """
10481
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10482
    instance_data = self.cfg.GetAllInstancesInfo()
10483

    
10484
    if self.group is None:
10485
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10486
                               (self.op.group_name, self.group_uuid))
10487

    
10488
    (new_splits, previous_splits) = \
10489
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
10490
                                             for node in self.op.nodes],
10491
                                            self.node_data, instance_data)
10492

    
10493
    if new_splits:
10494
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
10495

    
10496
      if not self.op.force:
10497
        raise errors.OpExecError("The following instances get split by this"
10498
                                 " change and --force was not given: %s" %
10499
                                 fmt_new_splits)
10500
      else:
10501
        self.LogWarning("This operation will split the following instances: %s",
10502
                        fmt_new_splits)
10503

    
10504
        if previous_splits:
10505
          self.LogWarning("In addition, these already-split instances continue"
10506
                          " to be spit across groups: %s",
10507
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
10508

    
10509
  def Exec(self, feedback_fn):
10510
    """Assign nodes to a new group.
10511

10512
    """
10513
    for node in self.op.nodes:
10514
      self.node_data[node].group = self.group_uuid
10515

    
10516
    self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
10517

    
10518
  @staticmethod
10519
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
10520
    """Check for split instances after a node assignment.
10521

10522
    This method considers a series of node assignments as an atomic operation,
10523
    and returns information about split instances after applying the set of
10524
    changes.
10525

10526
    In particular, it returns information about newly split instances, and
10527
    instances that were already split, and remain so after the change.
10528

10529
    Only instances whose disk template is listed in constants.DTS_INT_MIRROR are
10530
    considered.
10531

10532
    @type changes: list of (node_name, new_group_uuid) pairs.
10533
    @param changes: list of node assignments to consider.
10534
    @param node_data: a dict with data for all nodes
10535
    @param instance_data: a dict with all instances to consider
10536
    @rtype: a two-tuple
10537
    @return: a list of instances that were previously okay and result split as a
10538
      consequence of this change, and a list of instances that were previously
10539
      split and this change does not fix.
10540

10541
    """
10542
    changed_nodes = dict((node, group) for node, group in changes
10543
                         if node_data[node].group != group)
10544

    
10545
    all_split_instances = set()
10546
    previously_split_instances = set()
10547

    
10548
    def InstanceNodes(instance):
10549
      return [instance.primary_node] + list(instance.secondary_nodes)
10550

    
10551
    for inst in instance_data.values():
10552
      if inst.disk_template not in constants.DTS_INT_MIRROR:
10553
        continue
10554

    
10555
      instance_nodes = InstanceNodes(inst)
10556

    
10557
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
10558
        previously_split_instances.add(inst.name)
10559

    
10560
      if len(set(changed_nodes.get(node, node_data[node].group)
10561
                 for node in instance_nodes)) > 1:
10562
        all_split_instances.add(inst.name)
10563

    
10564
    return (list(all_split_instances - previously_split_instances),
10565
            list(previously_split_instances & all_split_instances))
10566

    
10567

    
10568
class _GroupQuery(_QueryBase):
10569

    
10570
  FIELDS = query.GROUP_FIELDS
10571

    
10572
  def ExpandNames(self, lu):
10573
    lu.needed_locks = {}
10574

    
10575
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
10576
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
10577

    
10578
    if not self.names:
10579
      self.wanted = [name_to_uuid[name]
10580
                     for name in utils.NiceSort(name_to_uuid.keys())]
10581
    else:
10582
      # Accept names to be either names or UUIDs.
10583
      missing = []
10584
      self.wanted = []
10585
      all_uuid = frozenset(self._all_groups.keys())
10586

    
10587
      for name in self.names:
10588
        if name in all_uuid:
10589
          self.wanted.append(name)
10590
        elif name in name_to_uuid:
10591
          self.wanted.append(name_to_uuid[name])
10592
        else:
10593
          missing.append(name)
10594

    
10595
      if missing:
10596
        raise errors.OpPrereqError("Some groups do not exist: %s" % missing,
10597
                                   errors.ECODE_NOENT)
10598

    
10599
  def DeclareLocks(self, lu, level):
10600
    pass
10601

    
10602
  def _GetQueryData(self, lu):
10603
    """Computes the list of node groups and their attributes.
10604

10605
    """
10606
    do_nodes = query.GQ_NODE in self.requested_data
10607
    do_instances = query.GQ_INST in self.requested_data
10608

    
10609
    group_to_nodes = None
10610
    group_to_instances = None
10611

    
10612
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
10613
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
10614
    # latter GetAllInstancesInfo() is not enough, for we have to go through
10615
    # instance->node. Hence, we will need to process nodes even if we only need
10616
    # instance information.
10617
    if do_nodes or do_instances:
10618
      all_nodes = lu.cfg.GetAllNodesInfo()
10619
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
10620
      node_to_group = {}
10621

    
10622
      for node in all_nodes.values():
10623
        if node.group in group_to_nodes:
10624
          group_to_nodes[node.group].append(node.name)
10625
          node_to_group[node.name] = node.group
10626

    
10627
      if do_instances:
10628
        all_instances = lu.cfg.GetAllInstancesInfo()
10629
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
10630

    
10631
        for instance in all_instances.values():
10632
          node = instance.primary_node
10633
          if node in node_to_group:
10634
            group_to_instances[node_to_group[node]].append(instance.name)
10635

    
10636
        if not do_nodes:
10637
          # Do not pass on node information if it was not requested.
10638
          group_to_nodes = None
10639

    
10640
    return query.GroupQueryData([self._all_groups[uuid]
10641
                                 for uuid in self.wanted],
10642
                                group_to_nodes, group_to_instances)
10643

    
10644

    
10645
class LUGroupQuery(NoHooksLU):
10646
  """Logical unit for querying node groups.
10647

10648
  """
10649
  REQ_BGL = False
10650

    
10651
  def CheckArguments(self):
10652
    self.gq = _GroupQuery(self.op.names, self.op.output_fields, False)
10653

    
10654
  def ExpandNames(self):
10655
    self.gq.ExpandNames(self)
10656

    
10657
  def Exec(self, feedback_fn):
10658
    return self.gq.OldStyleQuery(self)
10659

    
10660

    
10661
class LUGroupSetParams(LogicalUnit):
10662
  """Modifies the parameters of a node group.
10663

10664
  """
10665
  HPATH = "group-modify"
10666
  HTYPE = constants.HTYPE_GROUP
10667
  REQ_BGL = False
10668

    
10669
  def CheckArguments(self):
10670
    all_changes = [
10671
      self.op.ndparams,
10672
      self.op.alloc_policy,
10673
      ]
10674

    
10675
    if all_changes.count(None) == len(all_changes):
10676
      raise errors.OpPrereqError("Please pass at least one modification",
10677
                                 errors.ECODE_INVAL)
10678

    
10679
  def ExpandNames(self):
10680
    # This raises errors.OpPrereqError on its own:
10681
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10682

    
10683
    self.needed_locks = {
10684
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10685
      }
10686

    
10687
  def CheckPrereq(self):
10688
    """Check prerequisites.
10689

10690
    """
10691
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10692

    
10693
    if self.group is None:
10694
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10695
                               (self.op.group_name, self.group_uuid))
10696

    
10697
    if self.op.ndparams:
10698
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
10699
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10700
      self.new_ndparams = new_ndparams
10701

    
10702
  def BuildHooksEnv(self):
10703
    """Build hooks env.
10704

10705
    """
10706
    env = {
10707
      "GROUP_NAME": self.op.group_name,
10708
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
10709
      }
10710
    mn = self.cfg.GetMasterNode()
10711
    return env, [mn], [mn]
10712

    
10713
  def Exec(self, feedback_fn):
10714
    """Modifies the node group.
10715

10716
    """
10717
    result = []
10718

    
10719
    if self.op.ndparams:
10720
      self.group.ndparams = self.new_ndparams
10721
      result.append(("ndparams", str(self.group.ndparams)))
10722

    
10723
    if self.op.alloc_policy:
10724
      self.group.alloc_policy = self.op.alloc_policy
10725

    
10726
    self.cfg.Update(self.group, feedback_fn)
10727
    return result
10728

    
10729

    
10730

    
10731
class LUGroupRemove(LogicalUnit):
10732
  HPATH = "group-remove"
10733
  HTYPE = constants.HTYPE_GROUP
10734
  REQ_BGL = False
10735

    
10736
  def ExpandNames(self):
10737
    # This will raises errors.OpPrereqError on its own:
10738
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10739
    self.needed_locks = {
10740
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10741
      }
10742

    
10743
  def CheckPrereq(self):
10744
    """Check prerequisites.
10745

10746
    This checks that the given group name exists as a node group, that is
10747
    empty (i.e., contains no nodes), and that is not the last group of the
10748
    cluster.
10749

10750
    """
10751
    # Verify that the group is empty.
10752
    group_nodes = [node.name
10753
                   for node in self.cfg.GetAllNodesInfo().values()
10754
                   if node.group == self.group_uuid]
10755

    
10756
    if group_nodes:
10757
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
10758
                                 " nodes: %s" %
10759
                                 (self.op.group_name,
10760
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
10761
                                 errors.ECODE_STATE)
10762

    
10763
    # Verify the cluster would not be left group-less.
10764
    if len(self.cfg.GetNodeGroupList()) == 1:
10765
      raise errors.OpPrereqError("Group '%s' is the only group,"
10766
                                 " cannot be removed" %
10767
                                 self.op.group_name,
10768
                                 errors.ECODE_STATE)
10769

    
10770
  def BuildHooksEnv(self):
10771
    """Build hooks env.
10772

10773
    """
10774
    env = {
10775
      "GROUP_NAME": self.op.group_name,
10776
      }
10777
    mn = self.cfg.GetMasterNode()
10778
    return env, [mn], [mn]
10779

    
10780
  def Exec(self, feedback_fn):
10781
    """Remove the node group.
10782

10783
    """
10784
    try:
10785
      self.cfg.RemoveNodeGroup(self.group_uuid)
10786
    except errors.ConfigurationError:
10787
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
10788
                               (self.op.group_name, self.group_uuid))
10789

    
10790
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10791

    
10792

    
10793
class LUGroupRename(LogicalUnit):
10794
  HPATH = "group-rename"
10795
  HTYPE = constants.HTYPE_GROUP
10796
  REQ_BGL = False
10797

    
10798
  def ExpandNames(self):
10799
    # This raises errors.OpPrereqError on its own:
10800
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.old_name)
10801

    
10802
    self.needed_locks = {
10803
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10804
      }
10805

    
10806
  def CheckPrereq(self):
10807
    """Check prerequisites.
10808

10809
    This checks that the given old_name exists as a node group, and that
10810
    new_name doesn't.
10811

10812
    """
10813
    try:
10814
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
10815
    except errors.OpPrereqError:
10816
      pass
10817
    else:
10818
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
10819
                                 " node group (UUID: %s)" %
10820
                                 (self.op.new_name, new_name_uuid),
10821
                                 errors.ECODE_EXISTS)
10822

    
10823
  def BuildHooksEnv(self):
10824
    """Build hooks env.
10825

10826
    """
10827
    env = {
10828
      "OLD_NAME": self.op.old_name,
10829
      "NEW_NAME": self.op.new_name,
10830
      }
10831

    
10832
    mn = self.cfg.GetMasterNode()
10833
    all_nodes = self.cfg.GetAllNodesInfo()
10834
    run_nodes = [mn]
10835
    all_nodes.pop(mn, None)
10836

    
10837
    for node in all_nodes.values():
10838
      if node.group == self.group_uuid:
10839
        run_nodes.append(node.name)
10840

    
10841
    return env, run_nodes, run_nodes
10842

    
10843
  def Exec(self, feedback_fn):
10844
    """Rename the node group.
10845

10846
    """
10847
    group = self.cfg.GetNodeGroup(self.group_uuid)
10848

    
10849
    if group is None:
10850
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10851
                               (self.op.old_name, self.group_uuid))
10852

    
10853
    group.name = self.op.new_name
10854
    self.cfg.Update(group, feedback_fn)
10855

    
10856
    return self.op.new_name
10857

    
10858

    
10859
class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
10860
  """Generic tags LU.
10861

10862
  This is an abstract class which is the parent of all the other tags LUs.
10863

10864
  """
10865

    
10866
  def ExpandNames(self):
10867
    self.needed_locks = {}
10868
    if self.op.kind == constants.TAG_NODE:
10869
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
10870
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
10871
    elif self.op.kind == constants.TAG_INSTANCE:
10872
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
10873
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
10874

    
10875
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
10876
    # not possible to acquire the BGL based on opcode parameters)
10877

    
10878
  def CheckPrereq(self):
10879
    """Check prerequisites.
10880

10881
    """
10882
    if self.op.kind == constants.TAG_CLUSTER:
10883
      self.target = self.cfg.GetClusterInfo()
10884
    elif self.op.kind == constants.TAG_NODE:
10885
      self.target = self.cfg.GetNodeInfo(self.op.name)
10886
    elif self.op.kind == constants.TAG_INSTANCE:
10887
      self.target = self.cfg.GetInstanceInfo(self.op.name)
10888
    else:
10889
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
10890
                                 str(self.op.kind), errors.ECODE_INVAL)
10891

    
10892

    
10893
class LUTagsGet(TagsLU):
10894
  """Returns the tags of a given object.
10895

10896
  """
10897
  REQ_BGL = False
10898

    
10899
  def ExpandNames(self):
10900
    TagsLU.ExpandNames(self)
10901

    
10902
    # Share locks as this is only a read operation
10903
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
10904

    
10905
  def Exec(self, feedback_fn):
10906
    """Returns the tag list.
10907

10908
    """
10909
    return list(self.target.GetTags())
10910

    
10911

    
10912
class LUTagsSearch(NoHooksLU):
10913
  """Searches the tags for a given pattern.
10914

10915
  """
10916
  REQ_BGL = False
10917

    
10918
  def ExpandNames(self):
10919
    self.needed_locks = {}
10920

    
10921
  def CheckPrereq(self):
10922
    """Check prerequisites.
10923

10924
    This checks the pattern passed for validity by compiling it.
10925

10926
    """
10927
    try:
10928
      self.re = re.compile(self.op.pattern)
10929
    except re.error, err:
10930
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
10931
                                 (self.op.pattern, err), errors.ECODE_INVAL)
10932

    
10933
  def Exec(self, feedback_fn):
10934
    """Returns the tag list.
10935

10936
    """
10937
    cfg = self.cfg
10938
    tgts = [("/cluster", cfg.GetClusterInfo())]
10939
    ilist = cfg.GetAllInstancesInfo().values()
10940
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
10941
    nlist = cfg.GetAllNodesInfo().values()
10942
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
10943
    results = []
10944
    for path, target in tgts:
10945
      for tag in target.GetTags():
10946
        if self.re.search(tag):
10947
          results.append((path, tag))
10948
    return results
10949

    
10950

    
10951
class LUTagsSet(TagsLU):
10952
  """Sets a tag on a given object.
10953

10954
  """
10955
  REQ_BGL = False
10956

    
10957
  def CheckPrereq(self):
10958
    """Check prerequisites.
10959

10960
    This checks the type and length of the tag name and value.
10961

10962
    """
10963
    TagsLU.CheckPrereq(self)
10964
    for tag in self.op.tags:
10965
      objects.TaggableObject.ValidateTag(tag)
10966

    
10967
  def Exec(self, feedback_fn):
10968
    """Sets the tag.
10969

10970
    """
10971
    try:
10972
      for tag in self.op.tags:
10973
        self.target.AddTag(tag)
10974
    except errors.TagError, err:
10975
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
10976
    self.cfg.Update(self.target, feedback_fn)
10977

    
10978

    
10979
class LUTagsDel(TagsLU):
10980
  """Delete a list of tags from a given object.
10981

10982
  """
10983
  REQ_BGL = False
10984

    
10985
  def CheckPrereq(self):
10986
    """Check prerequisites.
10987

10988
    This checks that we have the given tag.
10989

10990
    """
10991
    TagsLU.CheckPrereq(self)
10992
    for tag in self.op.tags:
10993
      objects.TaggableObject.ValidateTag(tag)
10994
    del_tags = frozenset(self.op.tags)
10995
    cur_tags = self.target.GetTags()
10996

    
10997
    diff_tags = del_tags - cur_tags
10998
    if diff_tags:
10999
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
11000
      raise errors.OpPrereqError("Tag(s) %s not found" %
11001
                                 (utils.CommaJoin(diff_names), ),
11002
                                 errors.ECODE_NOENT)
11003

    
11004
  def Exec(self, feedback_fn):
11005
    """Remove the tag from the object.
11006

11007
    """
11008
    for tag in self.op.tags:
11009
      self.target.RemoveTag(tag)
11010
    self.cfg.Update(self.target, feedback_fn)
11011

    
11012

    
11013
class LUTestDelay(NoHooksLU):
11014
  """Sleep for a specified amount of time.
11015

11016
  This LU sleeps on the master and/or nodes for a specified amount of
11017
  time.
11018

11019
  """
11020
  REQ_BGL = False
11021

    
11022
  def ExpandNames(self):
11023
    """Expand names and set required locks.
11024

11025
    This expands the node list, if any.
11026

11027
    """
11028
    self.needed_locks = {}
11029
    if self.op.on_nodes:
11030
      # _GetWantedNodes can be used here, but is not always appropriate to use
11031
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
11032
      # more information.
11033
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
11034
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
11035

    
11036
  def _TestDelay(self):
11037
    """Do the actual sleep.
11038

11039
    """
11040
    if self.op.on_master:
11041
      if not utils.TestDelay(self.op.duration):
11042
        raise errors.OpExecError("Error during master delay test")
11043
    if self.op.on_nodes:
11044
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
11045
      for node, node_result in result.items():
11046
        node_result.Raise("Failure during rpc call to node %s" % node)
11047

    
11048
  def Exec(self, feedback_fn):
11049
    """Execute the test delay opcode, with the wanted repetitions.
11050

11051
    """
11052
    if self.op.repeat == 0:
11053
      self._TestDelay()
11054
    else:
11055
      top_value = self.op.repeat - 1
11056
      for i in range(self.op.repeat):
11057
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
11058
        self._TestDelay()
11059

    
11060

    
11061
class LUTestJqueue(NoHooksLU):
11062
  """Utility LU to test some aspects of the job queue.
11063

11064
  """
11065
  REQ_BGL = False
11066

    
11067
  # Must be lower than default timeout for WaitForJobChange to see whether it
11068
  # notices changed jobs
11069
  _CLIENT_CONNECT_TIMEOUT = 20.0
11070
  _CLIENT_CONFIRM_TIMEOUT = 60.0
11071

    
11072
  @classmethod
11073
  def _NotifyUsingSocket(cls, cb, errcls):
11074
    """Opens a Unix socket and waits for another program to connect.
11075

11076
    @type cb: callable
11077
    @param cb: Callback to send socket name to client
11078
    @type errcls: class
11079
    @param errcls: Exception class to use for errors
11080

11081
    """
11082
    # Using a temporary directory as there's no easy way to create temporary
11083
    # sockets without writing a custom loop around tempfile.mktemp and
11084
    # socket.bind
11085
    tmpdir = tempfile.mkdtemp()
11086
    try:
11087
      tmpsock = utils.PathJoin(tmpdir, "sock")
11088

    
11089
      logging.debug("Creating temporary socket at %s", tmpsock)
11090
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
11091
      try:
11092
        sock.bind(tmpsock)
11093
        sock.listen(1)
11094

    
11095
        # Send details to client
11096
        cb(tmpsock)
11097

    
11098
        # Wait for client to connect before continuing
11099
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
11100
        try:
11101
          (conn, _) = sock.accept()
11102
        except socket.error, err:
11103
          raise errcls("Client didn't connect in time (%s)" % err)
11104
      finally:
11105
        sock.close()
11106
    finally:
11107
      # Remove as soon as client is connected
11108
      shutil.rmtree(tmpdir)
11109

    
11110
    # Wait for client to close
11111
    try:
11112
      try:
11113
        # pylint: disable-msg=E1101
11114
        # Instance of '_socketobject' has no ... member
11115
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
11116
        conn.recv(1)
11117
      except socket.error, err:
11118
        raise errcls("Client failed to confirm notification (%s)" % err)
11119
    finally:
11120
      conn.close()
11121

    
11122
  def _SendNotification(self, test, arg, sockname):
11123
    """Sends a notification to the client.
11124

11125
    @type test: string
11126
    @param test: Test name
11127
    @param arg: Test argument (depends on test)
11128
    @type sockname: string
11129
    @param sockname: Socket path
11130

11131
    """
11132
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
11133

    
11134
  def _Notify(self, prereq, test, arg):
11135
    """Notifies the client of a test.
11136

11137
    @type prereq: bool
11138
    @param prereq: Whether this is a prereq-phase test
11139
    @type test: string
11140
    @param test: Test name
11141
    @param arg: Test argument (depends on test)
11142

11143
    """
11144
    if prereq:
11145
      errcls = errors.OpPrereqError
11146
    else:
11147
      errcls = errors.OpExecError
11148

    
11149
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
11150
                                                  test, arg),
11151
                                   errcls)
11152

    
11153
  def CheckArguments(self):
11154
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
11155
    self.expandnames_calls = 0
11156

    
11157
  def ExpandNames(self):
11158
    checkargs_calls = getattr(self, "checkargs_calls", 0)
11159
    if checkargs_calls < 1:
11160
      raise errors.ProgrammerError("CheckArguments was not called")
11161

    
11162
    self.expandnames_calls += 1
11163

    
11164
    if self.op.notify_waitlock:
11165
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
11166

    
11167
    self.LogInfo("Expanding names")
11168

    
11169
    # Get lock on master node (just to get a lock, not for a particular reason)
11170
    self.needed_locks = {
11171
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
11172
      }
11173

    
11174
  def Exec(self, feedback_fn):
11175
    if self.expandnames_calls < 1:
11176
      raise errors.ProgrammerError("ExpandNames was not called")
11177

    
11178
    if self.op.notify_exec:
11179
      self._Notify(False, constants.JQT_EXEC, None)
11180

    
11181
    self.LogInfo("Executing")
11182

    
11183
    if self.op.log_messages:
11184
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
11185
      for idx, msg in enumerate(self.op.log_messages):
11186
        self.LogInfo("Sending log message %s", idx + 1)
11187
        feedback_fn(constants.JQT_MSGPREFIX + msg)
11188
        # Report how many test messages have been sent
11189
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
11190

    
11191
    if self.op.fail:
11192
      raise errors.OpExecError("Opcode failure was requested")
11193

    
11194
    return True
11195

    
11196

    
11197
class IAllocator(object):
11198
  """IAllocator framework.
11199

11200
  An IAllocator instance has three sets of attributes:
11201
    - cfg that is needed to query the cluster
11202
    - input data (all members of the _KEYS class attribute are required)
11203
    - four buffer attributes (in|out_data|text), that represent the
11204
      input (to the external script) in text and data structure format,
11205
      and the output from it, again in two formats
11206
    - the result variables from the script (success, info, nodes) for
11207
      easy usage
11208

11209
  """
11210
  # pylint: disable-msg=R0902
11211
  # lots of instance attributes
11212
  _ALLO_KEYS = [
11213
    "name", "mem_size", "disks", "disk_template",
11214
    "os", "tags", "nics", "vcpus", "hypervisor",
11215
    ]
11216
  _RELO_KEYS = [
11217
    "name", "relocate_from",
11218
    ]
11219
  _EVAC_KEYS = [
11220
    "evac_nodes",
11221
    ]
11222

    
11223
  def __init__(self, cfg, rpc, mode, **kwargs):
11224
    self.cfg = cfg
11225
    self.rpc = rpc
11226
    # init buffer variables
11227
    self.in_text = self.out_text = self.in_data = self.out_data = None
11228
    # init all input fields so that pylint is happy
11229
    self.mode = mode
11230
    self.mem_size = self.disks = self.disk_template = None
11231
    self.os = self.tags = self.nics = self.vcpus = None
11232
    self.hypervisor = None
11233
    self.relocate_from = None
11234
    self.name = None
11235
    self.evac_nodes = None
11236
    # computed fields
11237
    self.required_nodes = None
11238
    # init result fields
11239
    self.success = self.info = self.result = None
11240
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11241
      keyset = self._ALLO_KEYS
11242
      fn = self._AddNewInstance
11243
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11244
      keyset = self._RELO_KEYS
11245
      fn = self._AddRelocateInstance
11246
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11247
      keyset = self._EVAC_KEYS
11248
      fn = self._AddEvacuateNodes
11249
    else:
11250
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
11251
                                   " IAllocator" % self.mode)
11252
    for key in kwargs:
11253
      if key not in keyset:
11254
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
11255
                                     " IAllocator" % key)
11256
      setattr(self, key, kwargs[key])
11257

    
11258
    for key in keyset:
11259
      if key not in kwargs:
11260
        raise errors.ProgrammerError("Missing input parameter '%s' to"
11261
                                     " IAllocator" % key)
11262
    self._BuildInputData(fn)
11263

    
11264
  def _ComputeClusterData(self):
11265
    """Compute the generic allocator input data.
11266

11267
    This is the data that is independent of the actual operation.
11268

11269
    """
11270
    cfg = self.cfg
11271
    cluster_info = cfg.GetClusterInfo()
11272
    # cluster data
11273
    data = {
11274
      "version": constants.IALLOCATOR_VERSION,
11275
      "cluster_name": cfg.GetClusterName(),
11276
      "cluster_tags": list(cluster_info.GetTags()),
11277
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
11278
      # we don't have job IDs
11279
      }
11280
    ninfo = cfg.GetAllNodesInfo()
11281
    iinfo = cfg.GetAllInstancesInfo().values()
11282
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
11283

    
11284
    # node data
11285
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
11286

    
11287
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11288
      hypervisor_name = self.hypervisor
11289
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11290
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
11291
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11292
      hypervisor_name = cluster_info.enabled_hypervisors[0]
11293

    
11294
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
11295
                                        hypervisor_name)
11296
    node_iinfo = \
11297
      self.rpc.call_all_instances_info(node_list,
11298
                                       cluster_info.enabled_hypervisors)
11299

    
11300
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
11301

    
11302
    config_ndata = self._ComputeBasicNodeData(ninfo)
11303
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
11304
                                                 i_list, config_ndata)
11305
    assert len(data["nodes"]) == len(ninfo), \
11306
        "Incomplete node data computed"
11307

    
11308
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
11309

    
11310
    self.in_data = data
11311

    
11312
  @staticmethod
11313
  def _ComputeNodeGroupData(cfg):
11314
    """Compute node groups data.
11315

11316
    """
11317
    ng = {}
11318
    for guuid, gdata in cfg.GetAllNodeGroupsInfo().items():
11319
      ng[guuid] = {
11320
        "name": gdata.name,
11321
        "alloc_policy": gdata.alloc_policy,
11322
        }
11323
    return ng
11324

    
11325
  @staticmethod
11326
  def _ComputeBasicNodeData(node_cfg):
11327
    """Compute global node data.
11328

11329
    @rtype: dict
11330
    @returns: a dict of name: (node dict, node config)
11331

11332
    """
11333
    node_results = {}
11334
    for ninfo in node_cfg.values():
11335
      # fill in static (config-based) values
11336
      pnr = {
11337
        "tags": list(ninfo.GetTags()),
11338
        "primary_ip": ninfo.primary_ip,
11339
        "secondary_ip": ninfo.secondary_ip,
11340
        "offline": ninfo.offline,
11341
        "drained": ninfo.drained,
11342
        "master_candidate": ninfo.master_candidate,
11343
        "group": ninfo.group,
11344
        "master_capable": ninfo.master_capable,
11345
        "vm_capable": ninfo.vm_capable,
11346
        }
11347

    
11348
      node_results[ninfo.name] = pnr
11349

    
11350
    return node_results
11351

    
11352
  @staticmethod
11353
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
11354
                              node_results):
11355
    """Compute global node data.
11356

11357
    @param node_results: the basic node structures as filled from the config
11358

11359
    """
11360
    # make a copy of the current dict
11361
    node_results = dict(node_results)
11362
    for nname, nresult in node_data.items():
11363
      assert nname in node_results, "Missing basic data for node %s" % nname
11364
      ninfo = node_cfg[nname]
11365

    
11366
      if not (ninfo.offline or ninfo.drained):
11367
        nresult.Raise("Can't get data for node %s" % nname)
11368
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
11369
                                nname)
11370
        remote_info = nresult.payload
11371

    
11372
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
11373
                     'vg_size', 'vg_free', 'cpu_total']:
11374
          if attr not in remote_info:
11375
            raise errors.OpExecError("Node '%s' didn't return attribute"
11376
                                     " '%s'" % (nname, attr))
11377
          if not isinstance(remote_info[attr], int):
11378
            raise errors.OpExecError("Node '%s' returned invalid value"
11379
                                     " for '%s': %s" %
11380
                                     (nname, attr, remote_info[attr]))
11381
        # compute memory used by primary instances
11382
        i_p_mem = i_p_up_mem = 0
11383
        for iinfo, beinfo in i_list:
11384
          if iinfo.primary_node == nname:
11385
            i_p_mem += beinfo[constants.BE_MEMORY]
11386
            if iinfo.name not in node_iinfo[nname].payload:
11387
              i_used_mem = 0
11388
            else:
11389
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
11390
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
11391
            remote_info['memory_free'] -= max(0, i_mem_diff)
11392

    
11393
            if iinfo.admin_up:
11394
              i_p_up_mem += beinfo[constants.BE_MEMORY]
11395

    
11396
        # compute memory used by instances
11397
        pnr_dyn = {
11398
          "total_memory": remote_info['memory_total'],
11399
          "reserved_memory": remote_info['memory_dom0'],
11400
          "free_memory": remote_info['memory_free'],
11401
          "total_disk": remote_info['vg_size'],
11402
          "free_disk": remote_info['vg_free'],
11403
          "total_cpus": remote_info['cpu_total'],
11404
          "i_pri_memory": i_p_mem,
11405
          "i_pri_up_memory": i_p_up_mem,
11406
          }
11407
        pnr_dyn.update(node_results[nname])
11408
        node_results[nname] = pnr_dyn
11409

    
11410
    return node_results
11411

    
11412
  @staticmethod
11413
  def _ComputeInstanceData(cluster_info, i_list):
11414
    """Compute global instance data.
11415

11416
    """
11417
    instance_data = {}
11418
    for iinfo, beinfo in i_list:
11419
      nic_data = []
11420
      for nic in iinfo.nics:
11421
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
11422
        nic_dict = {"mac": nic.mac,
11423
                    "ip": nic.ip,
11424
                    "mode": filled_params[constants.NIC_MODE],
11425
                    "link": filled_params[constants.NIC_LINK],
11426
                   }
11427
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
11428
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
11429
        nic_data.append(nic_dict)
11430
      pir = {
11431
        "tags": list(iinfo.GetTags()),
11432
        "admin_up": iinfo.admin_up,
11433
        "vcpus": beinfo[constants.BE_VCPUS],
11434
        "memory": beinfo[constants.BE_MEMORY],
11435
        "os": iinfo.os,
11436
        "nics": nic_data,
11437
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
11438
        "disk_template": iinfo.disk_template,
11439
        "hypervisor": iinfo.hypervisor,
11440
        }
11441
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
11442
                                                 pir["disks"])
11443
      # hail's relocation mode does not work without secondaries,
11444
      # as it exclusively tries replace-secondary moves. So, let's trick hail
11445
      # by specifying our primary and secondary node to be the same.
11446
      if iinfo.disk_template in constants.DTS_EXT_MIRROR:
11447
        pir["nodes"] = [iinfo.primary_node, iinfo.primary_node]
11448
      else:
11449
        pir["nodes"] = [iinfo.primary_node] + list(iinfo.secondary_nodes)
11450
      instance_data[iinfo.name] = pir
11451

    
11452
    return instance_data
11453

    
11454
  def _AddNewInstance(self):
11455
    """Add new instance data to allocator structure.
11456

11457
    This in combination with _AllocatorGetClusterData will create the
11458
    correct structure needed as input for the allocator.
11459

11460
    The checks for the completeness of the opcode must have already been
11461
    done.
11462

11463
    """
11464
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
11465

    
11466
    if self.disk_template in constants.DTS_INT_MIRROR:
11467
      self.required_nodes = 2
11468
    else:
11469
      self.required_nodes = 1
11470
    request = {
11471
      "name": self.name,
11472
      "disk_template": self.disk_template,
11473
      "tags": self.tags,
11474
      "os": self.os,
11475
      "vcpus": self.vcpus,
11476
      "memory": self.mem_size,
11477
      "disks": self.disks,
11478
      "disk_space_total": disk_space,
11479
      "nics": self.nics,
11480
      "required_nodes": self.required_nodes,
11481
      }
11482
    return request
11483

    
11484
  def _AddRelocateInstance(self):
11485
    """Add relocate instance data to allocator structure.
11486

11487
    This in combination with _IAllocatorGetClusterData will create the
11488
    correct structure needed as input for the allocator.
11489

11490
    The checks for the completeness of the opcode must have already been
11491
    done.
11492

11493
    """
11494
    instance = self.cfg.GetInstanceInfo(self.name)
11495
    if instance is None:
11496
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
11497
                                   " IAllocator" % self.name)
11498

    
11499
    if instance.disk_template not in constants.DTS_MIRRORED:
11500
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
11501
                                 errors.ECODE_INVAL)
11502

    
11503
    if instance.disk_template in constants.DTS_INT_MIRROR and \
11504
        len(instance.secondary_nodes) != 1:
11505
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
11506
                                 errors.ECODE_STATE)
11507

    
11508
    self.required_nodes = 1
11509
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
11510
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
11511

    
11512
    request = {
11513
      "name": self.name,
11514
      "disk_space_total": disk_space,
11515
      "required_nodes": self.required_nodes,
11516
      "relocate_from": self.relocate_from,
11517
      }
11518
    return request
11519

    
11520
  def _AddEvacuateNodes(self):
11521
    """Add evacuate nodes data to allocator structure.
11522

11523
    """
11524
    request = {
11525
      "evac_nodes": self.evac_nodes
11526
      }
11527
    return request
11528

    
11529
  def _BuildInputData(self, fn):
11530
    """Build input data structures.
11531

11532
    """
11533
    self._ComputeClusterData()
11534

    
11535
    request = fn()
11536
    request["type"] = self.mode
11537
    self.in_data["request"] = request
11538

    
11539
    self.in_text = serializer.Dump(self.in_data)
11540

    
11541
  def Run(self, name, validate=True, call_fn=None):
11542
    """Run an instance allocator and return the results.
11543

11544
    """
11545
    if call_fn is None:
11546
      call_fn = self.rpc.call_iallocator_runner
11547

    
11548
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
11549
    result.Raise("Failure while running the iallocator script")
11550

    
11551
    self.out_text = result.payload
11552
    if validate:
11553
      self._ValidateResult()
11554

    
11555
  def _ValidateResult(self):
11556
    """Process the allocator results.
11557

11558
    This will process and if successful save the result in
11559
    self.out_data and the other parameters.
11560

11561
    """
11562
    try:
11563
      rdict = serializer.Load(self.out_text)
11564
    except Exception, err:
11565
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
11566

    
11567
    if not isinstance(rdict, dict):
11568
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
11569

    
11570
    # TODO: remove backwards compatiblity in later versions
11571
    if "nodes" in rdict and "result" not in rdict:
11572
      rdict["result"] = rdict["nodes"]
11573
      del rdict["nodes"]
11574

    
11575
    for key in "success", "info", "result":
11576
      if key not in rdict:
11577
        raise errors.OpExecError("Can't parse iallocator results:"
11578
                                 " missing key '%s'" % key)
11579
      setattr(self, key, rdict[key])
11580

    
11581
    if not isinstance(rdict["result"], list):
11582
      raise errors.OpExecError("Can't parse iallocator results: 'result' key"
11583
                               " is not a list")
11584
    self.out_data = rdict
11585

    
11586

    
11587
class LUTestAllocator(NoHooksLU):
11588
  """Run allocator tests.
11589

11590
  This LU runs the allocator tests
11591

11592
  """
11593
  def CheckPrereq(self):
11594
    """Check prerequisites.
11595

11596
    This checks the opcode parameters depending on the director and mode test.
11597

11598
    """
11599
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11600
      for attr in ["mem_size", "disks", "disk_template",
11601
                   "os", "tags", "nics", "vcpus"]:
11602
        if not hasattr(self.op, attr):
11603
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
11604
                                     attr, errors.ECODE_INVAL)
11605
      iname = self.cfg.ExpandInstanceName(self.op.name)
11606
      if iname is not None:
11607
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
11608
                                   iname, errors.ECODE_EXISTS)
11609
      if not isinstance(self.op.nics, list):
11610
        raise errors.OpPrereqError("Invalid parameter 'nics'",
11611
                                   errors.ECODE_INVAL)
11612
      if not isinstance(self.op.disks, list):
11613
        raise errors.OpPrereqError("Invalid parameter 'disks'",
11614
                                   errors.ECODE_INVAL)
11615
      for row in self.op.disks:
11616
        if (not isinstance(row, dict) or
11617
            "size" not in row or
11618
            not isinstance(row["size"], int) or
11619
            "mode" not in row or
11620
            row["mode"] not in ['r', 'w']):
11621
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
11622
                                     " parameter", errors.ECODE_INVAL)
11623
      if self.op.hypervisor is None:
11624
        self.op.hypervisor = self.cfg.GetHypervisorType()
11625
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11626
      fname = _ExpandInstanceName(self.cfg, self.op.name)
11627
      self.op.name = fname
11628
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
11629
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11630
      if not hasattr(self.op, "evac_nodes"):
11631
        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
11632
                                   " opcode input", errors.ECODE_INVAL)
11633
    else:
11634
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
11635
                                 self.op.mode, errors.ECODE_INVAL)
11636

    
11637
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
11638
      if self.op.allocator is None:
11639
        raise errors.OpPrereqError("Missing allocator name",
11640
                                   errors.ECODE_INVAL)
11641
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
11642
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
11643
                                 self.op.direction, errors.ECODE_INVAL)
11644

    
11645
  def Exec(self, feedback_fn):
11646
    """Run the allocator test.
11647

11648
    """
11649
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11650
      ial = IAllocator(self.cfg, self.rpc,
11651
                       mode=self.op.mode,
11652
                       name=self.op.name,
11653
                       mem_size=self.op.mem_size,
11654
                       disks=self.op.disks,
11655
                       disk_template=self.op.disk_template,
11656
                       os=self.op.os,
11657
                       tags=self.op.tags,
11658
                       nics=self.op.nics,
11659
                       vcpus=self.op.vcpus,
11660
                       hypervisor=self.op.hypervisor,
11661
                       )
11662
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11663
      ial = IAllocator(self.cfg, self.rpc,
11664
                       mode=self.op.mode,
11665
                       name=self.op.name,
11666
                       relocate_from=list(self.relocate_from),
11667
                       )
11668
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11669
      ial = IAllocator(self.cfg, self.rpc,
11670
                       mode=self.op.mode,
11671
                       evac_nodes=self.op.evac_nodes)
11672
    else:
11673
      raise errors.ProgrammerError("Uncatched mode %s in"
11674
                                   " LUTestAllocator.Exec", self.op.mode)
11675

    
11676
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
11677
      result = ial.in_text
11678
    else:
11679
      ial.Run(self.op.allocator, validate=False)
11680
      result = ial.out_text
11681
    return result
11682

    
11683

    
11684
#: Query type implementations
11685
_QUERY_IMPL = {
11686
  constants.QR_INSTANCE: _InstanceQuery,
11687
  constants.QR_NODE: _NodeQuery,
11688
  constants.QR_GROUP: _GroupQuery,
11689
  }
11690

    
11691

    
11692
def _GetQueryImplementation(name):
11693
  """Returns the implemtnation for a query type.
11694

11695
  @param name: Query type, must be one of L{constants.QR_OP_QUERY}
11696

11697
  """
11698
  try:
11699
    return _QUERY_IMPL[name]
11700
  except KeyError:
11701
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
11702
                               errors.ECODE_INVAL)