Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 27485a97

History | View | Annotate | Download (405.2 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_NET_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_NET_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 ExpandNames(self):
5668
    self._ExpandAndLockInstance()
5669
    self.needed_locks[locking.LEVEL_NODE] = []
5670
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5671

    
5672
  def DeclareLocks(self, level):
5673
    if level == locking.LEVEL_NODE:
5674
      self._LockInstancesNodes()
5675

    
5676
  def BuildHooksEnv(self):
5677
    """Build hooks env.
5678

5679
    This runs on master, primary and secondary nodes of the instance.
5680

5681
    """
5682
    instance = self.instance
5683
    source_node = instance.primary_node
5684
    target_node = instance.secondary_nodes[0]
5685
    env = {
5686
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
5687
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5688
      "OLD_PRIMARY": source_node,
5689
      "OLD_SECONDARY": target_node,
5690
      "NEW_PRIMARY": target_node,
5691
      "NEW_SECONDARY": source_node,
5692
      }
5693
    env.update(_BuildInstanceHookEnvByObject(self, instance))
5694
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
5695
    nl_post = list(nl)
5696
    nl_post.append(source_node)
5697
    return env, nl, nl_post
5698

    
5699
  def CheckPrereq(self):
5700
    """Check prerequisites.
5701

5702
    This checks that the instance is in the cluster.
5703

5704
    """
5705
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5706
    assert self.instance is not None, \
5707
      "Cannot retrieve locked instance %s" % self.op.instance_name
5708

    
5709
    bep = self.cfg.GetClusterInfo().FillBE(instance)
5710
    if instance.disk_template not in constants.DTS_NET_MIRROR:
5711
      raise errors.OpPrereqError("Instance's disk layout is not"
5712
                                 " network mirrored, cannot failover.",
5713
                                 errors.ECODE_STATE)
5714

    
5715
    secondary_nodes = instance.secondary_nodes
5716
    if not secondary_nodes:
5717
      raise errors.ProgrammerError("no secondary node but using "
5718
                                   "a mirrored disk template")
5719

    
5720
    target_node = secondary_nodes[0]
5721
    _CheckNodeOnline(self, target_node)
5722
    _CheckNodeNotDrained(self, target_node)
5723
    if instance.admin_up:
5724
      # check memory requirements on the secondary node
5725
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
5726
                           instance.name, bep[constants.BE_MEMORY],
5727
                           instance.hypervisor)
5728
    else:
5729
      self.LogInfo("Not checking memory on the secondary node as"
5730
                   " instance will not be started")
5731

    
5732
    # check bridge existance
5733
    _CheckInstanceBridgesExist(self, instance, node=target_node)
5734

    
5735
  def Exec(self, feedback_fn):
5736
    """Failover an instance.
5737

5738
    The failover is done by shutting it down on its present node and
5739
    starting it on the secondary.
5740

5741
    """
5742
    instance = self.instance
5743
    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
5744

    
5745
    source_node = instance.primary_node
5746
    target_node = instance.secondary_nodes[0]
5747

    
5748
    if instance.admin_up:
5749
      feedback_fn("* checking disk consistency between source and target")
5750
      for dev in instance.disks:
5751
        # for drbd, these are drbd over lvm
5752
        if not _CheckDiskConsistency(self, dev, target_node, False):
5753
          if not self.op.ignore_consistency:
5754
            raise errors.OpExecError("Disk %s is degraded on target node,"
5755
                                     " aborting failover." % dev.iv_name)
5756
    else:
5757
      feedback_fn("* not checking disk consistency as instance is not running")
5758

    
5759
    feedback_fn("* shutting down instance on source node")
5760
    logging.info("Shutting down instance %s on node %s",
5761
                 instance.name, source_node)
5762

    
5763
    result = self.rpc.call_instance_shutdown(source_node, instance,
5764
                                             self.op.shutdown_timeout)
5765
    msg = result.fail_msg
5766
    if msg:
5767
      if self.op.ignore_consistency or primary_node.offline:
5768
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
5769
                             " Proceeding anyway. Please make sure node"
5770
                             " %s is down. Error details: %s",
5771
                             instance.name, source_node, source_node, msg)
5772
      else:
5773
        raise errors.OpExecError("Could not shutdown instance %s on"
5774
                                 " node %s: %s" %
5775
                                 (instance.name, source_node, msg))
5776

    
5777
    feedback_fn("* deactivating the instance's disks on source node")
5778
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
5779
      raise errors.OpExecError("Can't shut down the instance's disks.")
5780

    
5781
    instance.primary_node = target_node
5782
    # distribute new instance config to the other nodes
5783
    self.cfg.Update(instance, feedback_fn)
5784

    
5785
    # Only start the instance if it's marked as up
5786
    if instance.admin_up:
5787
      feedback_fn("* activating the instance's disks on target node")
5788
      logging.info("Starting instance %s on node %s",
5789
                   instance.name, target_node)
5790

    
5791
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
5792
                                           ignore_secondaries=True)
5793
      if not disks_ok:
5794
        _ShutdownInstanceDisks(self, instance)
5795
        raise errors.OpExecError("Can't activate the instance's disks")
5796

    
5797
      feedback_fn("* starting the instance on the target node")
5798
      result = self.rpc.call_instance_start(target_node, instance, None, None)
5799
      msg = result.fail_msg
5800
      if msg:
5801
        _ShutdownInstanceDisks(self, instance)
5802
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
5803
                                 (instance.name, target_node, msg))
5804

    
5805

    
5806
class LUInstanceMigrate(LogicalUnit):
5807
  """Migrate an instance.
5808

5809
  This is migration without shutting down, compared to the failover,
5810
  which is done with shutdown.
5811

5812
  """
5813
  HPATH = "instance-migrate"
5814
  HTYPE = constants.HTYPE_INSTANCE
5815
  REQ_BGL = False
5816

    
5817
  def CheckArguments(self):
5818
    _CheckIAllocatorOrNode(self, "iallocator", "target_node")
5819

    
5820
  def ExpandNames(self):
5821
    self._ExpandAndLockInstance()
5822

    
5823
    if self.op.target_node is not None:
5824
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
5825

    
5826
    self.needed_locks[locking.LEVEL_NODE] = []
5827
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5828

    
5829
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
5830
                                       self.op.cleanup, self.op.iallocator,
5831
                                       self.op.target_node)
5832
    self.tasklets = [self._migrater]
5833

    
5834
  def DeclareLocks(self, level):
5835
    if level == locking.LEVEL_NODE:
5836
      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
5837
      if instance.disk_template in constants.DTS_EXT_MIRROR:
5838
        if self.op.target_node is None:
5839
          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5840
        else:
5841
          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
5842
                                                   self.op.target_node]
5843
        del self.recalculate_locks[locking.LEVEL_NODE]
5844
      else:
5845
        self._LockInstancesNodes()
5846

    
5847
  def BuildHooksEnv(self):
5848
    """Build hooks env.
5849

5850
    This runs on master, primary and secondary nodes of the instance.
5851

5852
    """
5853
    instance = self._migrater.instance
5854
    source_node = instance.primary_node
5855
    target_node = self._migrater.target_node
5856
    env = _BuildInstanceHookEnvByObject(self, instance)
5857
    env["MIGRATE_LIVE"] = self._migrater.live
5858
    env["MIGRATE_CLEANUP"] = self.op.cleanup
5859
    env.update({
5860
        "OLD_PRIMARY": source_node,
5861
        "NEW_PRIMARY": target_node,
5862
        })
5863

    
5864
    if instance.disk_template in constants.DTS_NET_MIRROR:
5865
      env["OLD_SECONDARY"] = target_node
5866
      env["NEW_SECONDARY"] = source_node
5867
    else:
5868
      env["OLD_SECONDARY"] = env["NEW_SECONDARY"] = None
5869

    
5870
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
5871
    nl_post = list(nl)
5872
    nl_post.append(source_node)
5873
    return env, nl, nl_post
5874

    
5875

    
5876
class LUInstanceMove(LogicalUnit):
5877
  """Move an instance by data-copying.
5878

5879
  """
5880
  HPATH = "instance-move"
5881
  HTYPE = constants.HTYPE_INSTANCE
5882
  REQ_BGL = False
5883

    
5884
  def ExpandNames(self):
5885
    self._ExpandAndLockInstance()
5886
    target_node = _ExpandNodeName(self.cfg, self.op.target_node)
5887
    self.op.target_node = target_node
5888
    self.needed_locks[locking.LEVEL_NODE] = [target_node]
5889
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5890

    
5891
  def DeclareLocks(self, level):
5892
    if level == locking.LEVEL_NODE:
5893
      self._LockInstancesNodes(primary_only=True)
5894

    
5895
  def BuildHooksEnv(self):
5896
    """Build hooks env.
5897

5898
    This runs on master, primary and secondary nodes of the instance.
5899

5900
    """
5901
    env = {
5902
      "TARGET_NODE": self.op.target_node,
5903
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5904
      }
5905
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5906
    nl = [self.cfg.GetMasterNode()] + [self.instance.primary_node,
5907
                                       self.op.target_node]
5908
    return env, nl, nl
5909

    
5910
  def CheckPrereq(self):
5911
    """Check prerequisites.
5912

5913
    This checks that the instance is in the cluster.
5914

5915
    """
5916
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5917
    assert self.instance is not None, \
5918
      "Cannot retrieve locked instance %s" % self.op.instance_name
5919

    
5920
    node = self.cfg.GetNodeInfo(self.op.target_node)
5921
    assert node is not None, \
5922
      "Cannot retrieve locked node %s" % self.op.target_node
5923

    
5924
    self.target_node = target_node = node.name
5925

    
5926
    if target_node == instance.primary_node:
5927
      raise errors.OpPrereqError("Instance %s is already on the node %s" %
5928
                                 (instance.name, target_node),
5929
                                 errors.ECODE_STATE)
5930

    
5931
    bep = self.cfg.GetClusterInfo().FillBE(instance)
5932

    
5933
    for idx, dsk in enumerate(instance.disks):
5934
      if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
5935
        raise errors.OpPrereqError("Instance disk %d has a complex layout,"
5936
                                   " cannot copy" % idx, errors.ECODE_STATE)
5937

    
5938
    _CheckNodeOnline(self, target_node)
5939
    _CheckNodeNotDrained(self, target_node)
5940
    _CheckNodeVmCapable(self, target_node)
5941

    
5942
    if instance.admin_up:
5943
      # check memory requirements on the secondary node
5944
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
5945
                           instance.name, bep[constants.BE_MEMORY],
5946
                           instance.hypervisor)
5947
    else:
5948
      self.LogInfo("Not checking memory on the secondary node as"
5949
                   " instance will not be started")
5950

    
5951
    # check bridge existance
5952
    _CheckInstanceBridgesExist(self, instance, node=target_node)
5953

    
5954
  def Exec(self, feedback_fn):
5955
    """Move an instance.
5956

5957
    The move is done by shutting it down on its present node, copying
5958
    the data over (slow) and starting it on the new node.
5959

5960
    """
5961
    instance = self.instance
5962

    
5963
    source_node = instance.primary_node
5964
    target_node = self.target_node
5965

    
5966
    self.LogInfo("Shutting down instance %s on source node %s",
5967
                 instance.name, source_node)
5968

    
5969
    result = self.rpc.call_instance_shutdown(source_node, instance,
5970
                                             self.op.shutdown_timeout)
5971
    msg = result.fail_msg
5972
    if msg:
5973
      if self.op.ignore_consistency:
5974
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
5975
                             " Proceeding anyway. Please make sure node"
5976
                             " %s is down. Error details: %s",
5977
                             instance.name, source_node, source_node, msg)
5978
      else:
5979
        raise errors.OpExecError("Could not shutdown instance %s on"
5980
                                 " node %s: %s" %
5981
                                 (instance.name, source_node, msg))
5982

    
5983
    # create the target disks
5984
    try:
5985
      _CreateDisks(self, instance, target_node=target_node)
5986
    except errors.OpExecError:
5987
      self.LogWarning("Device creation failed, reverting...")
5988
      try:
5989
        _RemoveDisks(self, instance, target_node=target_node)
5990
      finally:
5991
        self.cfg.ReleaseDRBDMinors(instance.name)
5992
        raise
5993

    
5994
    cluster_name = self.cfg.GetClusterInfo().cluster_name
5995

    
5996
    errs = []
5997
    # activate, get path, copy the data over
5998
    for idx, disk in enumerate(instance.disks):
5999
      self.LogInfo("Copying data for disk %d", idx)
6000
      result = self.rpc.call_blockdev_assemble(target_node, disk,
6001
                                               instance.name, True, idx)
6002
      if result.fail_msg:
6003
        self.LogWarning("Can't assemble newly created disk %d: %s",
6004
                        idx, result.fail_msg)
6005
        errs.append(result.fail_msg)
6006
        break
6007
      dev_path = result.payload
6008
      result = self.rpc.call_blockdev_export(source_node, disk,
6009
                                             target_node, dev_path,
6010
                                             cluster_name)
6011
      if result.fail_msg:
6012
        self.LogWarning("Can't copy data over for disk %d: %s",
6013
                        idx, result.fail_msg)
6014
        errs.append(result.fail_msg)
6015
        break
6016

    
6017
    if errs:
6018
      self.LogWarning("Some disks failed to copy, aborting")
6019
      try:
6020
        _RemoveDisks(self, instance, target_node=target_node)
6021
      finally:
6022
        self.cfg.ReleaseDRBDMinors(instance.name)
6023
        raise errors.OpExecError("Errors during disk copy: %s" %
6024
                                 (",".join(errs),))
6025

    
6026
    instance.primary_node = target_node
6027
    self.cfg.Update(instance, feedback_fn)
6028

    
6029
    self.LogInfo("Removing the disks on the original node")
6030
    _RemoveDisks(self, instance, target_node=source_node)
6031

    
6032
    # Only start the instance if it's marked as up
6033
    if instance.admin_up:
6034
      self.LogInfo("Starting instance %s on node %s",
6035
                   instance.name, target_node)
6036

    
6037
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
6038
                                           ignore_secondaries=True)
6039
      if not disks_ok:
6040
        _ShutdownInstanceDisks(self, instance)
6041
        raise errors.OpExecError("Can't activate the instance's disks")
6042

    
6043
      result = self.rpc.call_instance_start(target_node, instance, None, None)
6044
      msg = result.fail_msg
6045
      if msg:
6046
        _ShutdownInstanceDisks(self, instance)
6047
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
6048
                                 (instance.name, target_node, msg))
6049

    
6050

    
6051
class LUNodeMigrate(LogicalUnit):
6052
  """Migrate all instances from a node.
6053

6054
  """
6055
  HPATH = "node-migrate"
6056
  HTYPE = constants.HTYPE_NODE
6057
  REQ_BGL = False
6058

    
6059
  def ExpandNames(self):
6060
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6061

    
6062
    self.needed_locks = {
6063
      locking.LEVEL_NODE: [self.op.node_name],
6064
      }
6065

    
6066
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6067

    
6068
    # Create tasklets for migrating instances for all instances on this node
6069
    names = []
6070
    tasklets = []
6071

    
6072
    for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name):
6073
      logging.debug("Migrating instance %s", inst.name)
6074
      names.append(inst.name)
6075

    
6076
      tasklets.append(TLMigrateInstance(self, inst.name, False))
6077

    
6078
    self.tasklets = tasklets
6079

    
6080
    # Declare instance locks
6081
    self.needed_locks[locking.LEVEL_INSTANCE] = names
6082

    
6083
  def DeclareLocks(self, level):
6084
    if level == locking.LEVEL_NODE:
6085
      self._LockInstancesNodes()
6086

    
6087
  def BuildHooksEnv(self):
6088
    """Build hooks env.
6089

6090
    This runs on the master, the primary and all the secondaries.
6091

6092
    """
6093
    env = {
6094
      "NODE_NAME": self.op.node_name,
6095
      }
6096

    
6097
    nl = [self.cfg.GetMasterNode()]
6098

    
6099
    return (env, nl, nl)
6100

    
6101

    
6102
class TLMigrateInstance(Tasklet):
6103
  """Tasklet class for instance migration.
6104

6105
  @type live: boolean
6106
  @ivar live: whether the migration will be done live or non-live;
6107
      this variable is initalized only after CheckPrereq has run
6108

6109
  """
6110
  def __init__(self, lu, instance_name, cleanup,
6111
               iallocator=None, target_node=None):
6112
    """Initializes this class.
6113

6114
    """
6115
    Tasklet.__init__(self, lu)
6116

    
6117
    # Parameters
6118
    self.instance_name = instance_name
6119
    self.cleanup = cleanup
6120
    self.live = False # will be overridden later
6121
    self.iallocator = iallocator
6122
    self.target_node = target_node
6123

    
6124
  def CheckPrereq(self):
6125
    """Check prerequisites.
6126

6127
    This checks that the instance is in the cluster.
6128

6129
    """
6130
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
6131
    instance = self.cfg.GetInstanceInfo(instance_name)
6132
    assert instance is not None
6133
    self.instance = instance
6134

    
6135
    if instance.disk_template not in constants.DTS_MIRRORED:
6136
      raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
6137
                                 " migrations" % instance.disk_template,
6138
                                 errors.ECODE_STATE)
6139

    
6140
    if instance.disk_template in constants.DTS_EXT_MIRROR:
6141
      if [self.iallocator, self.target_node].count(None) != 1:
6142
        raise errors.OpPrereqError("Do not specify both, iallocator and"
6143
                                   " target node", errors.ECODE_INVAL)
6144

    
6145
      if self.iallocator:
6146
        self._RunAllocator()
6147

    
6148
      # self.target_node is already populated, either directly or by the
6149
      # iallocator run
6150
      target_node = self.target_node
6151

    
6152
      if len(self.lu.tasklets) == 1:
6153
        # It is safe to remove locks only when we're the only tasklet in the LU
6154
        nodes_keep = [instance.primary_node, self.target_node]
6155
        nodes_rel = [node for node in self.lu.acquired_locks[locking.LEVEL_NODE]
6156
                     if node not in nodes_keep]
6157
        self.lu.context.glm.release(locking.LEVEL_NODE, nodes_rel)
6158
        self.lu.acquired_locks[locking.LEVEL_NODE] = nodes_keep
6159

    
6160
    else:
6161
      secondary_nodes = instance.secondary_nodes
6162
      if not secondary_nodes:
6163
        raise errors.ConfigurationError("No secondary node but using"
6164
                                        " %s disk template" %
6165
                                        instance.disk_template)
6166
      target_node = secondary_nodes[0]
6167

    
6168
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
6169

    
6170
    # check memory requirements on the secondary node
6171
    _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
6172
                         instance.name, i_be[constants.BE_MEMORY],
6173
                         instance.hypervisor)
6174

    
6175
    # check bridge existance
6176
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
6177

    
6178
    if not self.cleanup:
6179
      _CheckNodeNotDrained(self.lu, target_node)
6180
      result = self.rpc.call_instance_migratable(instance.primary_node,
6181
                                                 instance)
6182
      result.Raise("Can't migrate, please use failover",
6183
                   prereq=True, ecode=errors.ECODE_STATE)
6184

    
6185

    
6186
  def _RunAllocator(self):
6187
    """Run the allocator based on input opcode.
6188

6189
    """
6190
    ial = IAllocator(self.cfg, self.rpc,
6191
                     mode=constants.IALLOCATOR_MODE_RELOC,
6192
                     name=self.instance_name,
6193
                     # TODO See why hail breaks with a single node below
6194
                     relocate_from=[self.instance.primary_node,
6195
                                    self.instance.primary_node],
6196
                     )
6197

    
6198
    ial.Run(self.iallocator)
6199

    
6200
    if not ial.success:
6201
      raise errors.OpPrereqError("Can't compute nodes using"
6202
                                 " iallocator '%s': %s" %
6203
                                 (self.iallocator, ial.info),
6204
                                 errors.ECODE_NORES)
6205
    if len(ial.result) != ial.required_nodes:
6206
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
6207
                                 " of nodes (%s), required %s" %
6208
                                 (self.iallocator, len(ial.result),
6209
                                  ial.required_nodes), errors.ECODE_FAULT)
6210
    self.target_node = ial.result[0]
6211
    self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
6212
                 self.instance_name, self.iallocator,
6213
                 utils.CommaJoin(ial.result))
6214

    
6215
    if self.lu.op.live is not None and self.lu.op.mode is not None:
6216
      raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
6217
                                 " parameters are accepted",
6218
                                 errors.ECODE_INVAL)
6219
    if self.lu.op.live is not None:
6220
      if self.lu.op.live:
6221
        self.lu.op.mode = constants.HT_MIGRATION_LIVE
6222
      else:
6223
        self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
6224
      # reset the 'live' parameter to None so that repeated
6225
      # invocations of CheckPrereq do not raise an exception
6226
      self.lu.op.live = None
6227
    elif self.lu.op.mode is None:
6228
      # read the default value from the hypervisor
6229
      i_hv = self.cfg.GetClusterInfo().FillHV(self.instance, skip_globals=False)
6230
      self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
6231

    
6232
    self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
6233

    
6234
  def _WaitUntilSync(self):
6235
    """Poll with custom rpc for disk sync.
6236

6237
    This uses our own step-based rpc call.
6238

6239
    """
6240
    self.feedback_fn("* wait until resync is done")
6241
    all_done = False
6242
    while not all_done:
6243
      all_done = True
6244
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
6245
                                            self.nodes_ip,
6246
                                            self.instance.disks)
6247
      min_percent = 100
6248
      for node, nres in result.items():
6249
        nres.Raise("Cannot resync disks on node %s" % node)
6250
        node_done, node_percent = nres.payload
6251
        all_done = all_done and node_done
6252
        if node_percent is not None:
6253
          min_percent = min(min_percent, node_percent)
6254
      if not all_done:
6255
        if min_percent < 100:
6256
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
6257
        time.sleep(2)
6258

    
6259
  def _EnsureSecondary(self, node):
6260
    """Demote a node to secondary.
6261

6262
    """
6263
    self.feedback_fn("* switching node %s to secondary mode" % node)
6264

    
6265
    for dev in self.instance.disks:
6266
      self.cfg.SetDiskID(dev, node)
6267

    
6268
    result = self.rpc.call_blockdev_close(node, self.instance.name,
6269
                                          self.instance.disks)
6270
    result.Raise("Cannot change disk to secondary on node %s" % node)
6271

    
6272
  def _GoStandalone(self):
6273
    """Disconnect from the network.
6274

6275
    """
6276
    self.feedback_fn("* changing into standalone mode")
6277
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
6278
                                               self.instance.disks)
6279
    for node, nres in result.items():
6280
      nres.Raise("Cannot disconnect disks node %s" % node)
6281

    
6282
  def _GoReconnect(self, multimaster):
6283
    """Reconnect to the network.
6284

6285
    """
6286
    if multimaster:
6287
      msg = "dual-master"
6288
    else:
6289
      msg = "single-master"
6290
    self.feedback_fn("* changing disks into %s mode" % msg)
6291
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
6292
                                           self.instance.disks,
6293
                                           self.instance.name, multimaster)
6294
    for node, nres in result.items():
6295
      nres.Raise("Cannot change disks config on node %s" % node)
6296

    
6297
  def _ExecCleanup(self):
6298
    """Try to cleanup after a failed migration.
6299

6300
    The cleanup is done by:
6301
      - check that the instance is running only on one node
6302
        (and update the config if needed)
6303
      - change disks on its secondary node to secondary
6304
      - wait until disks are fully synchronized
6305
      - disconnect from the network
6306
      - change disks into single-master mode
6307
      - wait again until disks are fully synchronized
6308

6309
    """
6310
    instance = self.instance
6311
    target_node = self.target_node
6312
    source_node = self.source_node
6313

    
6314
    # check running on only one node
6315
    self.feedback_fn("* checking where the instance actually runs"
6316
                     " (if this hangs, the hypervisor might be in"
6317
                     " a bad state)")
6318
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
6319
    for node, result in ins_l.items():
6320
      result.Raise("Can't contact node %s" % node)
6321

    
6322
    runningon_source = instance.name in ins_l[source_node].payload
6323
    runningon_target = instance.name in ins_l[target_node].payload
6324

    
6325
    if runningon_source and runningon_target:
6326
      raise errors.OpExecError("Instance seems to be running on two nodes,"
6327
                               " or the hypervisor is confused. You will have"
6328
                               " to ensure manually that it runs only on one"
6329
                               " and restart this operation.")
6330

    
6331
    if not (runningon_source or runningon_target):
6332
      raise errors.OpExecError("Instance does not seem to be running at all."
6333
                               " In this case, it's safer to repair by"
6334
                               " running 'gnt-instance stop' to ensure disk"
6335
                               " shutdown, and then restarting it.")
6336

    
6337
    if runningon_target:
6338
      # the migration has actually succeeded, we need to update the config
6339
      self.feedback_fn("* instance running on secondary node (%s),"
6340
                       " updating config" % target_node)
6341
      instance.primary_node = target_node
6342
      self.cfg.Update(instance, self.feedback_fn)
6343
      demoted_node = source_node
6344
    else:
6345
      self.feedback_fn("* instance confirmed to be running on its"
6346
                       " primary node (%s)" % source_node)
6347
      demoted_node = target_node
6348

    
6349
    if instance.disk_template in constants.DTS_NET_MIRROR:
6350
      self._EnsureSecondary(demoted_node)
6351
      try:
6352
        self._WaitUntilSync()
6353
      except errors.OpExecError:
6354
        # we ignore here errors, since if the device is standalone, it
6355
        # won't be able to sync
6356
        pass
6357
      self._GoStandalone()
6358
      self._GoReconnect(False)
6359
      self._WaitUntilSync()
6360

    
6361
    self.feedback_fn("* done")
6362

    
6363
  def _RevertDiskStatus(self):
6364
    """Try to revert the disk status after a failed migration.
6365

6366
    """
6367
    target_node = self.target_node
6368
    if self.instance.disk_template in constants.DTS_EXT_MIRROR:
6369
      return
6370

    
6371
    try:
6372
      self._EnsureSecondary(target_node)
6373
      self._GoStandalone()
6374
      self._GoReconnect(False)
6375
      self._WaitUntilSync()
6376
    except errors.OpExecError, err:
6377
      self.lu.LogWarning("Migration failed and I can't reconnect the"
6378
                         " drives: error '%s'\n"
6379
                         "Please look and recover the instance status" %
6380
                         str(err))
6381

    
6382
  def _AbortMigration(self):
6383
    """Call the hypervisor code to abort a started migration.
6384

6385
    """
6386
    instance = self.instance
6387
    target_node = self.target_node
6388
    migration_info = self.migration_info
6389

    
6390
    abort_result = self.rpc.call_finalize_migration(target_node,
6391
                                                    instance,
6392
                                                    migration_info,
6393
                                                    False)
6394
    abort_msg = abort_result.fail_msg
6395
    if abort_msg:
6396
      logging.error("Aborting migration failed on target node %s: %s",
6397
                    target_node, abort_msg)
6398
      # Don't raise an exception here, as we stil have to try to revert the
6399
      # disk status, even if this step failed.
6400

    
6401
  def _ExecMigration(self):
6402
    """Migrate an instance.
6403

6404
    The migrate is done by:
6405
      - change the disks into dual-master mode
6406
      - wait until disks are fully synchronized again
6407
      - migrate the instance
6408
      - change disks on the new secondary node (the old primary) to secondary
6409
      - wait until disks are fully synchronized
6410
      - change disks into single-master mode
6411

6412
    """
6413
    instance = self.instance
6414
    target_node = self.target_node
6415
    source_node = self.source_node
6416

    
6417
    self.feedback_fn("* checking disk consistency between source and target")
6418
    for dev in instance.disks:
6419
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
6420
        raise errors.OpExecError("Disk %s is degraded or not fully"
6421
                                 " synchronized on target node,"
6422
                                 " aborting migrate." % dev.iv_name)
6423

    
6424
    # First get the migration information from the remote node
6425
    result = self.rpc.call_migration_info(source_node, instance)
6426
    msg = result.fail_msg
6427
    if msg:
6428
      log_err = ("Failed fetching source migration information from %s: %s" %
6429
                 (source_node, msg))
6430
      logging.error(log_err)
6431
      raise errors.OpExecError(log_err)
6432

    
6433
    self.migration_info = migration_info = result.payload
6434

    
6435
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
6436
      # Then switch the disks to master/master mode
6437
      self._EnsureSecondary(target_node)
6438
      self._GoStandalone()
6439
      self._GoReconnect(True)
6440
      self._WaitUntilSync()
6441

    
6442
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
6443
    result = self.rpc.call_accept_instance(target_node,
6444
                                           instance,
6445
                                           migration_info,
6446
                                           self.nodes_ip[target_node])
6447

    
6448
    msg = result.fail_msg
6449
    if msg:
6450
      logging.error("Instance pre-migration failed, trying to revert"
6451
                    " disk status: %s", msg)
6452
      self.feedback_fn("Pre-migration failed, aborting")
6453
      self._AbortMigration()
6454
      self._RevertDiskStatus()
6455
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
6456
                               (instance.name, msg))
6457

    
6458
    self.feedback_fn("* migrating instance to %s" % target_node)
6459
    time.sleep(10)
6460
    result = self.rpc.call_instance_migrate(source_node, instance,
6461
                                            self.nodes_ip[target_node],
6462
                                            self.live)
6463
    msg = result.fail_msg
6464
    if msg:
6465
      logging.error("Instance migration failed, trying to revert"
6466
                    " disk status: %s", msg)
6467
      self.feedback_fn("Migration failed, aborting")
6468
      self._AbortMigration()
6469
      self._RevertDiskStatus()
6470
      raise errors.OpExecError("Could not migrate instance %s: %s" %
6471
                               (instance.name, msg))
6472
    time.sleep(10)
6473

    
6474
    instance.primary_node = target_node
6475
    # distribute new instance config to the other nodes
6476
    self.cfg.Update(instance, self.feedback_fn)
6477

    
6478
    result = self.rpc.call_finalize_migration(target_node,
6479
                                              instance,
6480
                                              migration_info,
6481
                                              True)
6482
    msg = result.fail_msg
6483
    if msg:
6484
      logging.error("Instance migration succeeded, but finalization failed:"
6485
                    " %s", msg)
6486
      raise errors.OpExecError("Could not finalize instance migration: %s" %
6487
                               msg)
6488

    
6489
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
6490
      self._EnsureSecondary(source_node)
6491
      self._WaitUntilSync()
6492
      self._GoStandalone()
6493
      self._GoReconnect(False)
6494
      self._WaitUntilSync()
6495

    
6496
    self.feedback_fn("* done")
6497

    
6498
  def Exec(self, feedback_fn):
6499
    """Perform the migration.
6500

6501
    """
6502
    feedback_fn("Migrating instance %s" % self.instance.name)
6503

    
6504
    self.feedback_fn = feedback_fn
6505

    
6506
    self.source_node = self.instance.primary_node
6507

    
6508
    # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
6509
    if self.instance.disk_template in constants.DTS_NET_MIRROR:
6510
      self.target_node = self.instance.secondary_nodes[0]
6511
      # Otherwise self.target_node has been populated either
6512
      # directly, or through an iallocator.
6513

    
6514
    self.all_nodes = [self.source_node, self.target_node]
6515
    self.nodes_ip = {
6516
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
6517
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
6518
      }
6519

    
6520
    if self.cleanup:
6521
      return self._ExecCleanup()
6522
    else:
6523
      return self._ExecMigration()
6524

    
6525

    
6526
def _CreateBlockDev(lu, node, instance, device, force_create,
6527
                    info, force_open):
6528
  """Create a tree of block devices on a given node.
6529

6530
  If this device type has to be created on secondaries, create it and
6531
  all its children.
6532

6533
  If not, just recurse to children keeping the same 'force' value.
6534

6535
  @param lu: the lu on whose behalf we execute
6536
  @param node: the node on which to create the device
6537
  @type instance: L{objects.Instance}
6538
  @param instance: the instance which owns the device
6539
  @type device: L{objects.Disk}
6540
  @param device: the device to create
6541
  @type force_create: boolean
6542
  @param force_create: whether to force creation of this device; this
6543
      will be change to True whenever we find a device which has
6544
      CreateOnSecondary() attribute
6545
  @param info: the extra 'metadata' we should attach to the device
6546
      (this will be represented as a LVM tag)
6547
  @type force_open: boolean
6548
  @param force_open: this parameter will be passes to the
6549
      L{backend.BlockdevCreate} function where it specifies
6550
      whether we run on primary or not, and it affects both
6551
      the child assembly and the device own Open() execution
6552

6553
  """
6554
  if device.CreateOnSecondary():
6555
    force_create = True
6556

    
6557
  if device.children:
6558
    for child in device.children:
6559
      _CreateBlockDev(lu, node, instance, child, force_create,
6560
                      info, force_open)
6561

    
6562
  if not force_create:
6563
    return
6564

    
6565
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
6566

    
6567

    
6568
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
6569
  """Create a single block device on a given node.
6570

6571
  This will not recurse over children of the device, so they must be
6572
  created in advance.
6573

6574
  @param lu: the lu on whose behalf we execute
6575
  @param node: the node on which to create the device
6576
  @type instance: L{objects.Instance}
6577
  @param instance: the instance which owns the device
6578
  @type device: L{objects.Disk}
6579
  @param device: the device to create
6580
  @param info: the extra 'metadata' we should attach to the device
6581
      (this will be represented as a LVM tag)
6582
  @type force_open: boolean
6583
  @param force_open: this parameter will be passes to the
6584
      L{backend.BlockdevCreate} function where it specifies
6585
      whether we run on primary or not, and it affects both
6586
      the child assembly and the device own Open() execution
6587

6588
  """
6589
  lu.cfg.SetDiskID(device, node)
6590
  result = lu.rpc.call_blockdev_create(node, device, device.size,
6591
                                       instance.name, force_open, info)
6592
  result.Raise("Can't create block device %s on"
6593
               " node %s for instance %s" % (device, node, instance.name))
6594
  if device.physical_id is None:
6595
    device.physical_id = result.payload
6596

    
6597

    
6598
def _GenerateUniqueNames(lu, exts):
6599
  """Generate a suitable LV name.
6600

6601
  This will generate a logical volume name for the given instance.
6602

6603
  """
6604
  results = []
6605
  for val in exts:
6606
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
6607
    results.append("%s%s" % (new_id, val))
6608
  return results
6609

    
6610

    
6611
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgname, names, iv_name,
6612
                         p_minor, s_minor):
6613
  """Generate a drbd8 device complete with its children.
6614

6615
  """
6616
  port = lu.cfg.AllocatePort()
6617
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
6618
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
6619
                          logical_id=(vgname, names[0]))
6620
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
6621
                          logical_id=(vgname, names[1]))
6622
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
6623
                          logical_id=(primary, secondary, port,
6624
                                      p_minor, s_minor,
6625
                                      shared_secret),
6626
                          children=[dev_data, dev_meta],
6627
                          iv_name=iv_name)
6628
  return drbd_dev
6629

    
6630

    
6631
def _GenerateDiskTemplate(lu, template_name,
6632
                          instance_name, primary_node,
6633
                          secondary_nodes, disk_info,
6634
                          file_storage_dir, file_driver,
6635
                          base_index, feedback_fn):
6636
  """Generate the entire disk layout for a given template type.
6637

6638
  """
6639
  #TODO: compute space requirements
6640

    
6641
  vgname = lu.cfg.GetVGName()
6642
  disk_count = len(disk_info)
6643
  disks = []
6644
  if template_name == constants.DT_DISKLESS:
6645
    pass
6646
  elif template_name == constants.DT_PLAIN:
6647
    if len(secondary_nodes) != 0:
6648
      raise errors.ProgrammerError("Wrong template configuration")
6649

    
6650
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6651
                                      for i in range(disk_count)])
6652
    for idx, disk in enumerate(disk_info):
6653
      disk_index = idx + base_index
6654
      vg = disk.get("vg", vgname)
6655
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
6656
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
6657
                              logical_id=(vg, names[idx]),
6658
                              iv_name="disk/%d" % disk_index,
6659
                              mode=disk["mode"])
6660
      disks.append(disk_dev)
6661
  elif template_name == constants.DT_DRBD8:
6662
    if len(secondary_nodes) != 1:
6663
      raise errors.ProgrammerError("Wrong template configuration")
6664
    remote_node = secondary_nodes[0]
6665
    minors = lu.cfg.AllocateDRBDMinor(
6666
      [primary_node, remote_node] * len(disk_info), instance_name)
6667

    
6668
    names = []
6669
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6670
                                               for i in range(disk_count)]):
6671
      names.append(lv_prefix + "_data")
6672
      names.append(lv_prefix + "_meta")
6673
    for idx, disk in enumerate(disk_info):
6674
      disk_index = idx + base_index
6675
      vg = disk.get("vg", vgname)
6676
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
6677
                                      disk["size"], vg, names[idx*2:idx*2+2],
6678
                                      "disk/%d" % disk_index,
6679
                                      minors[idx*2], minors[idx*2+1])
6680
      disk_dev.mode = disk["mode"]
6681
      disks.append(disk_dev)
6682
  elif template_name == constants.DT_FILE:
6683
    if len(secondary_nodes) != 0:
6684
      raise errors.ProgrammerError("Wrong template configuration")
6685

    
6686
    opcodes.RequireFileStorage()
6687

    
6688
    for idx, disk in enumerate(disk_info):
6689
      disk_index = idx + base_index
6690
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6691
                              iv_name="disk/%d" % disk_index,
6692
                              logical_id=(file_driver,
6693
                                          "%s/disk%d" % (file_storage_dir,
6694
                                                         disk_index)),
6695
                              mode=disk["mode"])
6696
      disks.append(disk_dev)
6697
  elif template_name == constants.DT_SHARED_FILE:
6698
    if len(secondary_nodes) != 0:
6699
      raise errors.ProgrammerError("Wrong template configuration")
6700

    
6701
    opcodes.RequireSharedFileStorage()
6702

    
6703
    for idx, disk in enumerate(disk_info):
6704
      disk_index = idx + base_index
6705
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6706
                              iv_name="disk/%d" % disk_index,
6707
                              logical_id=(file_driver,
6708
                                          "%s/disk%d" % (file_storage_dir,
6709
                                                         disk_index)),
6710
                              mode=disk["mode"])
6711
      disks.append(disk_dev)
6712
  elif template_name == constants.DT_BLOCK:
6713
    if len(secondary_nodes) != 0:
6714
      raise errors.ProgrammerError("Wrong template configuration")
6715

    
6716
    for idx, disk in enumerate(disk_info):
6717
      disk_index = idx + base_index
6718
      disk_dev = objects.Disk(dev_type=constants.LD_BLOCKDEV, size=disk["size"],
6719
                              logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
6720
                                          disk["adopt"]),
6721
                              iv_name="disk/%d" % disk_index,
6722
                              mode=disk["mode"])
6723
      disks.append(disk_dev)
6724

    
6725
  else:
6726
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
6727
  return disks
6728

    
6729

    
6730
def _GetInstanceInfoText(instance):
6731
  """Compute that text that should be added to the disk's metadata.
6732

6733
  """
6734
  return "originstname+%s" % instance.name
6735

    
6736

    
6737
def _CalcEta(time_taken, written, total_size):
6738
  """Calculates the ETA based on size written and total size.
6739

6740
  @param time_taken: The time taken so far
6741
  @param written: amount written so far
6742
  @param total_size: The total size of data to be written
6743
  @return: The remaining time in seconds
6744

6745
  """
6746
  avg_time = time_taken / float(written)
6747
  return (total_size - written) * avg_time
6748

    
6749

    
6750
def _WipeDisks(lu, instance):
6751
  """Wipes instance disks.
6752

6753
  @type lu: L{LogicalUnit}
6754
  @param lu: the logical unit on whose behalf we execute
6755
  @type instance: L{objects.Instance}
6756
  @param instance: the instance whose disks we should create
6757
  @return: the success of the wipe
6758

6759
  """
6760
  node = instance.primary_node
6761

    
6762
  for device in instance.disks:
6763
    lu.cfg.SetDiskID(device, node)
6764

    
6765
  logging.info("Pause sync of instance %s disks", instance.name)
6766
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
6767

    
6768
  for idx, success in enumerate(result.payload):
6769
    if not success:
6770
      logging.warn("pause-sync of instance %s for disks %d failed",
6771
                   instance.name, idx)
6772

    
6773
  try:
6774
    for idx, device in enumerate(instance.disks):
6775
      lu.LogInfo("* Wiping disk %d", idx)
6776
      logging.info("Wiping disk %d for instance %s, node %s",
6777
                   idx, instance.name, node)
6778

    
6779
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
6780
      # MAX_WIPE_CHUNK at max
6781
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
6782
                            constants.MIN_WIPE_CHUNK_PERCENT)
6783

    
6784
      offset = 0
6785
      size = device.size
6786
      last_output = 0
6787
      start_time = time.time()
6788

    
6789
      while offset < size:
6790
        wipe_size = min(wipe_chunk_size, size - offset)
6791
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
6792
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
6793
                     (idx, offset, wipe_size))
6794
        now = time.time()
6795
        offset += wipe_size
6796
        if now - last_output >= 60:
6797
          eta = _CalcEta(now - start_time, offset, size)
6798
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
6799
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
6800
          last_output = now
6801
  finally:
6802
    logging.info("Resume sync of instance %s disks", instance.name)
6803

    
6804
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
6805

    
6806
    for idx, success in enumerate(result.payload):
6807
      if not success:
6808
        lu.LogWarning("Warning: Resume sync of disk %d failed. Please have a"
6809
                      " look at the status and troubleshoot the issue.", idx)
6810
        logging.warn("resume-sync of instance %s for disks %d failed",
6811
                     instance.name, idx)
6812

    
6813

    
6814
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
6815
  """Create all disks for an instance.
6816

6817
  This abstracts away some work from AddInstance.
6818

6819
  @type lu: L{LogicalUnit}
6820
  @param lu: the logical unit on whose behalf we execute
6821
  @type instance: L{objects.Instance}
6822
  @param instance: the instance whose disks we should create
6823
  @type to_skip: list
6824
  @param to_skip: list of indices to skip
6825
  @type target_node: string
6826
  @param target_node: if passed, overrides the target node for creation
6827
  @rtype: boolean
6828
  @return: the success of the creation
6829

6830
  """
6831
  info = _GetInstanceInfoText(instance)
6832
  if target_node is None:
6833
    pnode = instance.primary_node
6834
    all_nodes = instance.all_nodes
6835
  else:
6836
    pnode = target_node
6837
    all_nodes = [pnode]
6838

    
6839
  if instance.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE):
6840
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6841
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
6842

    
6843
    result.Raise("Failed to create directory '%s' on"
6844
                 " node %s" % (file_storage_dir, pnode))
6845

    
6846
  # Note: this needs to be kept in sync with adding of disks in
6847
  # LUInstanceSetParams
6848
  for idx, device in enumerate(instance.disks):
6849
    if to_skip and idx in to_skip:
6850
      continue
6851
    logging.info("Creating volume %s for instance %s",
6852
                 device.iv_name, instance.name)
6853
    #HARDCODE
6854
    for node in all_nodes:
6855
      f_create = node == pnode
6856
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
6857

    
6858

    
6859
def _RemoveDisks(lu, instance, target_node=None):
6860
  """Remove all disks for an instance.
6861

6862
  This abstracts away some work from `AddInstance()` and
6863
  `RemoveInstance()`. Note that in case some of the devices couldn't
6864
  be removed, the removal will continue with the other ones (compare
6865
  with `_CreateDisks()`).
6866

6867
  @type lu: L{LogicalUnit}
6868
  @param lu: the logical unit on whose behalf we execute
6869
  @type instance: L{objects.Instance}
6870
  @param instance: the instance whose disks we should remove
6871
  @type target_node: string
6872
  @param target_node: used to override the node on which to remove the disks
6873
  @rtype: boolean
6874
  @return: the success of the removal
6875

6876
  """
6877
  logging.info("Removing block devices for instance %s", instance.name)
6878

    
6879
  all_result = True
6880
  for device in instance.disks:
6881
    if target_node:
6882
      edata = [(target_node, device)]
6883
    else:
6884
      edata = device.ComputeNodeTree(instance.primary_node)
6885
    for node, disk in edata:
6886
      lu.cfg.SetDiskID(disk, node)
6887
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
6888
      if msg:
6889
        lu.LogWarning("Could not remove block device %s on node %s,"
6890
                      " continuing anyway: %s", device.iv_name, node, msg)
6891
        all_result = False
6892

    
6893
  if instance.disk_template == constants.DT_FILE:
6894
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6895
    if target_node:
6896
      tgt = target_node
6897
    else:
6898
      tgt = instance.primary_node
6899
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
6900
    if result.fail_msg:
6901
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
6902
                    file_storage_dir, instance.primary_node, result.fail_msg)
6903
      all_result = False
6904

    
6905
  return all_result
6906

    
6907

    
6908
def _ComputeDiskSizePerVG(disk_template, disks):
6909
  """Compute disk size requirements in the volume group
6910

6911
  """
6912
  def _compute(disks, payload):
6913
    """Universal algorithm
6914

6915
    """
6916
    vgs = {}
6917
    for disk in disks:
6918
      vgs[disk["vg"]] = vgs.get("vg", 0) + disk["size"] + payload
6919

    
6920
    return vgs
6921

    
6922
  # Required free disk space as a function of disk and swap space
6923
  req_size_dict = {
6924
    constants.DT_DISKLESS: {},
6925
    constants.DT_PLAIN: _compute(disks, 0),
6926
    # 128 MB are added for drbd metadata for each disk
6927
    constants.DT_DRBD8: _compute(disks, 128),
6928
    constants.DT_FILE: {},
6929
    constants.DT_SHARED_FILE: {},
6930
  }
6931

    
6932
  if disk_template not in req_size_dict:
6933
    raise errors.ProgrammerError("Disk template '%s' size requirement"
6934
                                 " is unknown" %  disk_template)
6935

    
6936
  return req_size_dict[disk_template]
6937

    
6938

    
6939
def _ComputeDiskSize(disk_template, disks):
6940
  """Compute disk size requirements in the volume group
6941

6942
  """
6943
  # Required free disk space as a function of disk and swap space
6944
  req_size_dict = {
6945
    constants.DT_DISKLESS: None,
6946
    constants.DT_PLAIN: sum(d["size"] for d in disks),
6947
    # 128 MB are added for drbd metadata for each disk
6948
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
6949
    constants.DT_FILE: None,
6950
    constants.DT_SHARED_FILE: 0,
6951
    constants.DT_BLOCK: 0,
6952
  }
6953

    
6954
  if disk_template not in req_size_dict:
6955
    raise errors.ProgrammerError("Disk template '%s' size requirement"
6956
                                 " is unknown" %  disk_template)
6957

    
6958
  return req_size_dict[disk_template]
6959

    
6960

    
6961
def _FilterVmNodes(lu, nodenames):
6962
  """Filters out non-vm_capable nodes from a list.
6963

6964
  @type lu: L{LogicalUnit}
6965
  @param lu: the logical unit for which we check
6966
  @type nodenames: list
6967
  @param nodenames: the list of nodes on which we should check
6968
  @rtype: list
6969
  @return: the list of vm-capable nodes
6970

6971
  """
6972
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
6973
  return [name for name in nodenames if name not in vm_nodes]
6974

    
6975

    
6976
def _CheckHVParams(lu, nodenames, hvname, hvparams):
6977
  """Hypervisor parameter validation.
6978

6979
  This function abstract the hypervisor parameter validation to be
6980
  used in both instance create and instance modify.
6981

6982
  @type lu: L{LogicalUnit}
6983
  @param lu: the logical unit for which we check
6984
  @type nodenames: list
6985
  @param nodenames: the list of nodes on which we should check
6986
  @type hvname: string
6987
  @param hvname: the name of the hypervisor we should use
6988
  @type hvparams: dict
6989
  @param hvparams: the parameters which we need to check
6990
  @raise errors.OpPrereqError: if the parameters are not valid
6991

6992
  """
6993
  nodenames = _FilterVmNodes(lu, nodenames)
6994
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
6995
                                                  hvname,
6996
                                                  hvparams)
6997
  for node in nodenames:
6998
    info = hvinfo[node]
6999
    if info.offline:
7000
      continue
7001
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
7002

    
7003

    
7004
def _CheckOSParams(lu, required, nodenames, osname, osparams):
7005
  """OS parameters validation.
7006

7007
  @type lu: L{LogicalUnit}
7008
  @param lu: the logical unit for which we check
7009
  @type required: boolean
7010
  @param required: whether the validation should fail if the OS is not
7011
      found
7012
  @type nodenames: list
7013
  @param nodenames: the list of nodes on which we should check
7014
  @type osname: string
7015
  @param osname: the name of the hypervisor we should use
7016
  @type osparams: dict
7017
  @param osparams: the parameters which we need to check
7018
  @raise errors.OpPrereqError: if the parameters are not valid
7019

7020
  """
7021
  nodenames = _FilterVmNodes(lu, nodenames)
7022
  result = lu.rpc.call_os_validate(required, nodenames, osname,
7023
                                   [constants.OS_VALIDATE_PARAMETERS],
7024
                                   osparams)
7025
  for node, nres in result.items():
7026
    # we don't check for offline cases since this should be run only
7027
    # against the master node and/or an instance's nodes
7028
    nres.Raise("OS Parameters validation failed on node %s" % node)
7029
    if not nres.payload:
7030
      lu.LogInfo("OS %s not found on node %s, validation skipped",
7031
                 osname, node)
7032

    
7033

    
7034
class LUInstanceCreate(LogicalUnit):
7035
  """Create an instance.
7036

7037
  """
7038
  HPATH = "instance-add"
7039
  HTYPE = constants.HTYPE_INSTANCE
7040
  REQ_BGL = False
7041

    
7042
  def CheckArguments(self):
7043
    """Check arguments.
7044

7045
    """
7046
    # do not require name_check to ease forward/backward compatibility
7047
    # for tools
7048
    if self.op.no_install and self.op.start:
7049
      self.LogInfo("No-installation mode selected, disabling startup")
7050
      self.op.start = False
7051
    # validate/normalize the instance name
7052
    self.op.instance_name = \
7053
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
7054

    
7055
    if self.op.ip_check and not self.op.name_check:
7056
      # TODO: make the ip check more flexible and not depend on the name check
7057
      raise errors.OpPrereqError("Cannot do ip check without a name check",
7058
                                 errors.ECODE_INVAL)
7059

    
7060
    # check nics' parameter names
7061
    for nic in self.op.nics:
7062
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
7063

    
7064
    # check disks. parameter names and consistent adopt/no-adopt strategy
7065
    has_adopt = has_no_adopt = False
7066
    for disk in self.op.disks:
7067
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
7068
      if "adopt" in disk:
7069
        has_adopt = True
7070
      else:
7071
        has_no_adopt = True
7072
    if has_adopt and has_no_adopt:
7073
      raise errors.OpPrereqError("Either all disks are adopted or none is",
7074
                                 errors.ECODE_INVAL)
7075
    if has_adopt:
7076
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
7077
        raise errors.OpPrereqError("Disk adoption is not supported for the"
7078
                                   " '%s' disk template" %
7079
                                   self.op.disk_template,
7080
                                   errors.ECODE_INVAL)
7081
      if self.op.iallocator is not None:
7082
        raise errors.OpPrereqError("Disk adoption not allowed with an"
7083
                                   " iallocator script", errors.ECODE_INVAL)
7084
      if self.op.mode == constants.INSTANCE_IMPORT:
7085
        raise errors.OpPrereqError("Disk adoption not allowed for"
7086
                                   " instance import", errors.ECODE_INVAL)
7087
    else:
7088
      if self.op.disk_template in constants.DTS_MUST_ADOPT:
7089
        raise errors.OpPrereqError("Disk template %s requires disk adoption,"
7090
                                   " but no 'adopt' parameter given" %
7091
                                   self.op.disk_template,
7092
                                   errors.ECODE_INVAL)
7093

    
7094
    self.adopt_disks = has_adopt
7095

    
7096
    # instance name verification
7097
    if self.op.name_check:
7098
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
7099
      self.op.instance_name = self.hostname1.name
7100
      # used in CheckPrereq for ip ping check
7101
      self.check_ip = self.hostname1.ip
7102
    else:
7103
      self.check_ip = None
7104

    
7105
    # file storage checks
7106
    if (self.op.file_driver and
7107
        not self.op.file_driver in constants.FILE_DRIVER):
7108
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
7109
                                 self.op.file_driver, errors.ECODE_INVAL)
7110

    
7111
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
7112
      raise errors.OpPrereqError("File storage directory path not absolute",
7113
                                 errors.ECODE_INVAL)
7114

    
7115
    ### Node/iallocator related checks
7116
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
7117

    
7118
    if self.op.pnode is not None:
7119
      if self.op.disk_template in constants.DTS_NET_MIRROR:
7120
        if self.op.snode is None:
7121
          raise errors.OpPrereqError("The networked disk templates need"
7122
                                     " a mirror node", errors.ECODE_INVAL)
7123
      elif self.op.snode:
7124
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
7125
                        " template")
7126
        self.op.snode = None
7127

    
7128
    self._cds = _GetClusterDomainSecret()
7129

    
7130
    if self.op.mode == constants.INSTANCE_IMPORT:
7131
      # On import force_variant must be True, because if we forced it at
7132
      # initial install, our only chance when importing it back is that it
7133
      # works again!
7134
      self.op.force_variant = True
7135

    
7136
      if self.op.no_install:
7137
        self.LogInfo("No-installation mode has no effect during import")
7138

    
7139
    elif self.op.mode == constants.INSTANCE_CREATE:
7140
      if self.op.os_type is None:
7141
        raise errors.OpPrereqError("No guest OS specified",
7142
                                   errors.ECODE_INVAL)
7143
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
7144
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
7145
                                   " installation" % self.op.os_type,
7146
                                   errors.ECODE_STATE)
7147
      if self.op.disk_template is None:
7148
        raise errors.OpPrereqError("No disk template specified",
7149
                                   errors.ECODE_INVAL)
7150

    
7151
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7152
      # Check handshake to ensure both clusters have the same domain secret
7153
      src_handshake = self.op.source_handshake
7154
      if not src_handshake:
7155
        raise errors.OpPrereqError("Missing source handshake",
7156
                                   errors.ECODE_INVAL)
7157

    
7158
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
7159
                                                           src_handshake)
7160
      if errmsg:
7161
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
7162
                                   errors.ECODE_INVAL)
7163

    
7164
      # Load and check source CA
7165
      self.source_x509_ca_pem = self.op.source_x509_ca
7166
      if not self.source_x509_ca_pem:
7167
        raise errors.OpPrereqError("Missing source X509 CA",
7168
                                   errors.ECODE_INVAL)
7169

    
7170
      try:
7171
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
7172
                                                    self._cds)
7173
      except OpenSSL.crypto.Error, err:
7174
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
7175
                                   (err, ), errors.ECODE_INVAL)
7176

    
7177
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
7178
      if errcode is not None:
7179
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
7180
                                   errors.ECODE_INVAL)
7181

    
7182
      self.source_x509_ca = cert
7183

    
7184
      src_instance_name = self.op.source_instance_name
7185
      if not src_instance_name:
7186
        raise errors.OpPrereqError("Missing source instance name",
7187
                                   errors.ECODE_INVAL)
7188

    
7189
      self.source_instance_name = \
7190
          netutils.GetHostname(name=src_instance_name).name
7191

    
7192
    else:
7193
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
7194
                                 self.op.mode, errors.ECODE_INVAL)
7195

    
7196
  def ExpandNames(self):
7197
    """ExpandNames for CreateInstance.
7198

7199
    Figure out the right locks for instance creation.
7200

7201
    """
7202
    self.needed_locks = {}
7203

    
7204
    instance_name = self.op.instance_name
7205
    # this is just a preventive check, but someone might still add this
7206
    # instance in the meantime, and creation will fail at lock-add time
7207
    if instance_name in self.cfg.GetInstanceList():
7208
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7209
                                 instance_name, errors.ECODE_EXISTS)
7210

    
7211
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
7212

    
7213
    if self.op.iallocator:
7214
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7215
    else:
7216
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
7217
      nodelist = [self.op.pnode]
7218
      if self.op.snode is not None:
7219
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
7220
        nodelist.append(self.op.snode)
7221
      self.needed_locks[locking.LEVEL_NODE] = nodelist
7222

    
7223
    # in case of import lock the source node too
7224
    if self.op.mode == constants.INSTANCE_IMPORT:
7225
      src_node = self.op.src_node
7226
      src_path = self.op.src_path
7227

    
7228
      if src_path is None:
7229
        self.op.src_path = src_path = self.op.instance_name
7230

    
7231
      if src_node is None:
7232
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7233
        self.op.src_node = None
7234
        if os.path.isabs(src_path):
7235
          raise errors.OpPrereqError("Importing an instance from an absolute"
7236
                                     " path requires a source node option.",
7237
                                     errors.ECODE_INVAL)
7238
      else:
7239
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
7240
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
7241
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
7242
        if not os.path.isabs(src_path):
7243
          self.op.src_path = src_path = \
7244
            utils.PathJoin(constants.EXPORT_DIR, src_path)
7245

    
7246
  def _RunAllocator(self):
7247
    """Run the allocator based on input opcode.
7248

7249
    """
7250
    nics = [n.ToDict() for n in self.nics]
7251
    ial = IAllocator(self.cfg, self.rpc,
7252
                     mode=constants.IALLOCATOR_MODE_ALLOC,
7253
                     name=self.op.instance_name,
7254
                     disk_template=self.op.disk_template,
7255
                     tags=[],
7256
                     os=self.op.os_type,
7257
                     vcpus=self.be_full[constants.BE_VCPUS],
7258
                     mem_size=self.be_full[constants.BE_MEMORY],
7259
                     disks=self.disks,
7260
                     nics=nics,
7261
                     hypervisor=self.op.hypervisor,
7262
                     )
7263

    
7264
    ial.Run(self.op.iallocator)
7265

    
7266
    if not ial.success:
7267
      raise errors.OpPrereqError("Can't compute nodes using"
7268
                                 " iallocator '%s': %s" %
7269
                                 (self.op.iallocator, ial.info),
7270
                                 errors.ECODE_NORES)
7271
    if len(ial.result) != ial.required_nodes:
7272
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
7273
                                 " of nodes (%s), required %s" %
7274
                                 (self.op.iallocator, len(ial.result),
7275
                                  ial.required_nodes), errors.ECODE_FAULT)
7276
    self.op.pnode = ial.result[0]
7277
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
7278
                 self.op.instance_name, self.op.iallocator,
7279
                 utils.CommaJoin(ial.result))
7280
    if ial.required_nodes == 2:
7281
      self.op.snode = ial.result[1]
7282

    
7283
  def BuildHooksEnv(self):
7284
    """Build hooks env.
7285

7286
    This runs on master, primary and secondary nodes of the instance.
7287

7288
    """
7289
    env = {
7290
      "ADD_MODE": self.op.mode,
7291
      }
7292
    if self.op.mode == constants.INSTANCE_IMPORT:
7293
      env["SRC_NODE"] = self.op.src_node
7294
      env["SRC_PATH"] = self.op.src_path
7295
      env["SRC_IMAGES"] = self.src_images
7296

    
7297
    env.update(_BuildInstanceHookEnv(
7298
      name=self.op.instance_name,
7299
      primary_node=self.op.pnode,
7300
      secondary_nodes=self.secondaries,
7301
      status=self.op.start,
7302
      os_type=self.op.os_type,
7303
      memory=self.be_full[constants.BE_MEMORY],
7304
      vcpus=self.be_full[constants.BE_VCPUS],
7305
      nics=_NICListToTuple(self, self.nics),
7306
      disk_template=self.op.disk_template,
7307
      disks=[(d["size"], d["mode"]) for d in self.disks],
7308
      bep=self.be_full,
7309
      hvp=self.hv_full,
7310
      hypervisor_name=self.op.hypervisor,
7311
    ))
7312

    
7313
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
7314
          self.secondaries)
7315
    return env, nl, nl
7316

    
7317
  def _ReadExportInfo(self):
7318
    """Reads the export information from disk.
7319

7320
    It will override the opcode source node and path with the actual
7321
    information, if these two were not specified before.
7322

7323
    @return: the export information
7324

7325
    """
7326
    assert self.op.mode == constants.INSTANCE_IMPORT
7327

    
7328
    src_node = self.op.src_node
7329
    src_path = self.op.src_path
7330

    
7331
    if src_node is None:
7332
      locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
7333
      exp_list = self.rpc.call_export_list(locked_nodes)
7334
      found = False
7335
      for node in exp_list:
7336
        if exp_list[node].fail_msg:
7337
          continue
7338
        if src_path in exp_list[node].payload:
7339
          found = True
7340
          self.op.src_node = src_node = node
7341
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
7342
                                                       src_path)
7343
          break
7344
      if not found:
7345
        raise errors.OpPrereqError("No export found for relative path %s" %
7346
                                    src_path, errors.ECODE_INVAL)
7347

    
7348
    _CheckNodeOnline(self, src_node)
7349
    result = self.rpc.call_export_info(src_node, src_path)
7350
    result.Raise("No export or invalid export found in dir %s" % src_path)
7351

    
7352
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
7353
    if not export_info.has_section(constants.INISECT_EXP):
7354
      raise errors.ProgrammerError("Corrupted export config",
7355
                                   errors.ECODE_ENVIRON)
7356

    
7357
    ei_version = export_info.get(constants.INISECT_EXP, "version")
7358
    if (int(ei_version) != constants.EXPORT_VERSION):
7359
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
7360
                                 (ei_version, constants.EXPORT_VERSION),
7361
                                 errors.ECODE_ENVIRON)
7362
    return export_info
7363

    
7364
  def _ReadExportParams(self, einfo):
7365
    """Use export parameters as defaults.
7366

7367
    In case the opcode doesn't specify (as in override) some instance
7368
    parameters, then try to use them from the export information, if
7369
    that declares them.
7370

7371
    """
7372
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
7373

    
7374
    if self.op.disk_template is None:
7375
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
7376
        self.op.disk_template = einfo.get(constants.INISECT_INS,
7377
                                          "disk_template")
7378
      else:
7379
        raise errors.OpPrereqError("No disk template specified and the export"
7380
                                   " is missing the disk_template information",
7381
                                   errors.ECODE_INVAL)
7382

    
7383
    if not self.op.disks:
7384
      if einfo.has_option(constants.INISECT_INS, "disk_count"):
7385
        disks = []
7386
        # TODO: import the disk iv_name too
7387
        for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
7388
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
7389
          disks.append({"size": disk_sz})
7390
        self.op.disks = disks
7391
      else:
7392
        raise errors.OpPrereqError("No disk info specified and the export"
7393
                                   " is missing the disk information",
7394
                                   errors.ECODE_INVAL)
7395

    
7396
    if (not self.op.nics and
7397
        einfo.has_option(constants.INISECT_INS, "nic_count")):
7398
      nics = []
7399
      for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
7400
        ndict = {}
7401
        for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
7402
          v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
7403
          ndict[name] = v
7404
        nics.append(ndict)
7405
      self.op.nics = nics
7406

    
7407
    if (self.op.hypervisor is None and
7408
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
7409
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
7410
    if einfo.has_section(constants.INISECT_HYP):
7411
      # use the export parameters but do not override the ones
7412
      # specified by the user
7413
      for name, value in einfo.items(constants.INISECT_HYP):
7414
        if name not in self.op.hvparams:
7415
          self.op.hvparams[name] = value
7416

    
7417
    if einfo.has_section(constants.INISECT_BEP):
7418
      # use the parameters, without overriding
7419
      for name, value in einfo.items(constants.INISECT_BEP):
7420
        if name not in self.op.beparams:
7421
          self.op.beparams[name] = value
7422
    else:
7423
      # try to read the parameters old style, from the main section
7424
      for name in constants.BES_PARAMETERS:
7425
        if (name not in self.op.beparams and
7426
            einfo.has_option(constants.INISECT_INS, name)):
7427
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
7428

    
7429
    if einfo.has_section(constants.INISECT_OSP):
7430
      # use the parameters, without overriding
7431
      for name, value in einfo.items(constants.INISECT_OSP):
7432
        if name not in self.op.osparams:
7433
          self.op.osparams[name] = value
7434

    
7435
  def _RevertToDefaults(self, cluster):
7436
    """Revert the instance parameters to the default values.
7437

7438
    """
7439
    # hvparams
7440
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
7441
    for name in self.op.hvparams.keys():
7442
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
7443
        del self.op.hvparams[name]
7444
    # beparams
7445
    be_defs = cluster.SimpleFillBE({})
7446
    for name in self.op.beparams.keys():
7447
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
7448
        del self.op.beparams[name]
7449
    # nic params
7450
    nic_defs = cluster.SimpleFillNIC({})
7451
    for nic in self.op.nics:
7452
      for name in constants.NICS_PARAMETERS:
7453
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
7454
          del nic[name]
7455
    # osparams
7456
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
7457
    for name in self.op.osparams.keys():
7458
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
7459
        del self.op.osparams[name]
7460

    
7461
  def CheckPrereq(self):
7462
    """Check prerequisites.
7463

7464
    """
7465
    if self.op.mode == constants.INSTANCE_IMPORT:
7466
      export_info = self._ReadExportInfo()
7467
      self._ReadExportParams(export_info)
7468

    
7469
    if (not self.cfg.GetVGName() and
7470
        self.op.disk_template not in constants.DTS_NOT_LVM):
7471
      raise errors.OpPrereqError("Cluster does not support lvm-based"
7472
                                 " instances", errors.ECODE_STATE)
7473

    
7474
    if self.op.hypervisor is None:
7475
      self.op.hypervisor = self.cfg.GetHypervisorType()
7476

    
7477
    cluster = self.cfg.GetClusterInfo()
7478
    enabled_hvs = cluster.enabled_hypervisors
7479
    if self.op.hypervisor not in enabled_hvs:
7480
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
7481
                                 " cluster (%s)" % (self.op.hypervisor,
7482
                                  ",".join(enabled_hvs)),
7483
                                 errors.ECODE_STATE)
7484

    
7485
    # check hypervisor parameter syntax (locally)
7486
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
7487
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
7488
                                      self.op.hvparams)
7489
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
7490
    hv_type.CheckParameterSyntax(filled_hvp)
7491
    self.hv_full = filled_hvp
7492
    # check that we don't specify global parameters on an instance
7493
    _CheckGlobalHvParams(self.op.hvparams)
7494

    
7495
    # fill and remember the beparams dict
7496
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
7497
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
7498

    
7499
    # build os parameters
7500
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
7501

    
7502
    # now that hvp/bep are in final format, let's reset to defaults,
7503
    # if told to do so
7504
    if self.op.identify_defaults:
7505
      self._RevertToDefaults(cluster)
7506

    
7507
    # NIC buildup
7508
    self.nics = []
7509
    for idx, nic in enumerate(self.op.nics):
7510
      nic_mode_req = nic.get("mode", None)
7511
      nic_mode = nic_mode_req
7512
      if nic_mode is None:
7513
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
7514

    
7515
      # in routed mode, for the first nic, the default ip is 'auto'
7516
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
7517
        default_ip_mode = constants.VALUE_AUTO
7518
      else:
7519
        default_ip_mode = constants.VALUE_NONE
7520

    
7521
      # ip validity checks
7522
      ip = nic.get("ip", default_ip_mode)
7523
      if ip is None or ip.lower() == constants.VALUE_NONE:
7524
        nic_ip = None
7525
      elif ip.lower() == constants.VALUE_AUTO:
7526
        if not self.op.name_check:
7527
          raise errors.OpPrereqError("IP address set to auto but name checks"
7528
                                     " have been skipped",
7529
                                     errors.ECODE_INVAL)
7530
        nic_ip = self.hostname1.ip
7531
      else:
7532
        if not netutils.IPAddress.IsValid(ip):
7533
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
7534
                                     errors.ECODE_INVAL)
7535
        nic_ip = ip
7536

    
7537
      # TODO: check the ip address for uniqueness
7538
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
7539
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
7540
                                   errors.ECODE_INVAL)
7541

    
7542
      # MAC address verification
7543
      mac = nic.get("mac", constants.VALUE_AUTO)
7544
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7545
        mac = utils.NormalizeAndValidateMac(mac)
7546

    
7547
        try:
7548
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
7549
        except errors.ReservationError:
7550
          raise errors.OpPrereqError("MAC address %s already in use"
7551
                                     " in cluster" % mac,
7552
                                     errors.ECODE_NOTUNIQUE)
7553

    
7554
      # bridge verification
7555
      bridge = nic.get("bridge", None)
7556
      link = nic.get("link", None)
7557
      if bridge and link:
7558
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
7559
                                   " at the same time", errors.ECODE_INVAL)
7560
      elif bridge and nic_mode == constants.NIC_MODE_ROUTED:
7561
        raise errors.OpPrereqError("Cannot pass 'bridge' on a routed nic",
7562
                                   errors.ECODE_INVAL)
7563
      elif bridge:
7564
        link = bridge
7565

    
7566
      nicparams = {}
7567
      if nic_mode_req:
7568
        nicparams[constants.NIC_MODE] = nic_mode_req
7569
      if link:
7570
        nicparams[constants.NIC_LINK] = link
7571

    
7572
      check_params = cluster.SimpleFillNIC(nicparams)
7573
      objects.NIC.CheckParameterSyntax(check_params)
7574
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
7575

    
7576
    # disk checks/pre-build
7577
    self.disks = []
7578
    for disk in self.op.disks:
7579
      mode = disk.get("mode", constants.DISK_RDWR)
7580
      if mode not in constants.DISK_ACCESS_SET:
7581
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
7582
                                   mode, errors.ECODE_INVAL)
7583
      size = disk.get("size", None)
7584
      if size is None:
7585
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
7586
      try:
7587
        size = int(size)
7588
      except (TypeError, ValueError):
7589
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
7590
                                   errors.ECODE_INVAL)
7591
      vg = disk.get("vg", self.cfg.GetVGName())
7592
      new_disk = {"size": size, "mode": mode, "vg": vg}
7593
      if "adopt" in disk:
7594
        new_disk["adopt"] = disk["adopt"]
7595
      self.disks.append(new_disk)
7596

    
7597
    if self.op.mode == constants.INSTANCE_IMPORT:
7598

    
7599
      # Check that the new instance doesn't have less disks than the export
7600
      instance_disks = len(self.disks)
7601
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
7602
      if instance_disks < export_disks:
7603
        raise errors.OpPrereqError("Not enough disks to import."
7604
                                   " (instance: %d, export: %d)" %
7605
                                   (instance_disks, export_disks),
7606
                                   errors.ECODE_INVAL)
7607

    
7608
      disk_images = []
7609
      for idx in range(export_disks):
7610
        option = 'disk%d_dump' % idx
7611
        if export_info.has_option(constants.INISECT_INS, option):
7612
          # FIXME: are the old os-es, disk sizes, etc. useful?
7613
          export_name = export_info.get(constants.INISECT_INS, option)
7614
          image = utils.PathJoin(self.op.src_path, export_name)
7615
          disk_images.append(image)
7616
        else:
7617
          disk_images.append(False)
7618

    
7619
      self.src_images = disk_images
7620

    
7621
      old_name = export_info.get(constants.INISECT_INS, 'name')
7622
      try:
7623
        exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
7624
      except (TypeError, ValueError), err:
7625
        raise errors.OpPrereqError("Invalid export file, nic_count is not"
7626
                                   " an integer: %s" % str(err),
7627
                                   errors.ECODE_STATE)
7628
      if self.op.instance_name == old_name:
7629
        for idx, nic in enumerate(self.nics):
7630
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
7631
            nic_mac_ini = 'nic%d_mac' % idx
7632
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
7633

    
7634
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
7635

    
7636
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
7637
    if self.op.ip_check:
7638
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
7639
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
7640
                                   (self.check_ip, self.op.instance_name),
7641
                                   errors.ECODE_NOTUNIQUE)
7642

    
7643
    #### mac address generation
7644
    # By generating here the mac address both the allocator and the hooks get
7645
    # the real final mac address rather than the 'auto' or 'generate' value.
7646
    # There is a race condition between the generation and the instance object
7647
    # creation, which means that we know the mac is valid now, but we're not
7648
    # sure it will be when we actually add the instance. If things go bad
7649
    # adding the instance will abort because of a duplicate mac, and the
7650
    # creation job will fail.
7651
    for nic in self.nics:
7652
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7653
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
7654

    
7655
    #### allocator run
7656

    
7657
    if self.op.iallocator is not None:
7658
      self._RunAllocator()
7659

    
7660
    #### node related checks
7661

    
7662
    # check primary node
7663
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
7664
    assert self.pnode is not None, \
7665
      "Cannot retrieve locked node %s" % self.op.pnode
7666
    if pnode.offline:
7667
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
7668
                                 pnode.name, errors.ECODE_STATE)
7669
    if pnode.drained:
7670
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
7671
                                 pnode.name, errors.ECODE_STATE)
7672
    if not pnode.vm_capable:
7673
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
7674
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
7675

    
7676
    self.secondaries = []
7677

    
7678
    # mirror node verification
7679
    if self.op.disk_template in constants.DTS_NET_MIRROR:
7680
      if self.op.snode == pnode.name:
7681
        raise errors.OpPrereqError("The secondary node cannot be the"
7682
                                   " primary node.", errors.ECODE_INVAL)
7683
      _CheckNodeOnline(self, self.op.snode)
7684
      _CheckNodeNotDrained(self, self.op.snode)
7685
      _CheckNodeVmCapable(self, self.op.snode)
7686
      self.secondaries.append(self.op.snode)
7687

    
7688
    nodenames = [pnode.name] + self.secondaries
7689

    
7690
    if not self.adopt_disks:
7691
      # Check lv size requirements, if not adopting
7692
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
7693
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
7694

    
7695
    elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
7696
      all_lvs = set([i["vg"] + "/" + i["adopt"] for i in self.disks])
7697
      if len(all_lvs) != len(self.disks):
7698
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
7699
                                   errors.ECODE_INVAL)
7700
      for lv_name in all_lvs:
7701
        try:
7702
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
7703
          # to ReserveLV uses the same syntax
7704
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
7705
        except errors.ReservationError:
7706
          raise errors.OpPrereqError("LV named %s used by another instance" %
7707
                                     lv_name, errors.ECODE_NOTUNIQUE)
7708

    
7709
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
7710
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
7711

    
7712
      node_lvs = self.rpc.call_lv_list([pnode.name],
7713
                                       vg_names.payload.keys())[pnode.name]
7714
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
7715
      node_lvs = node_lvs.payload
7716

    
7717
      delta = all_lvs.difference(node_lvs.keys())
7718
      if delta:
7719
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
7720
                                   utils.CommaJoin(delta),
7721
                                   errors.ECODE_INVAL)
7722
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
7723
      if online_lvs:
7724
        raise errors.OpPrereqError("Online logical volumes found, cannot"
7725
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
7726
                                   errors.ECODE_STATE)
7727
      # update the size of disk based on what is found
7728
      for dsk in self.disks:
7729
        dsk["size"] = int(float(node_lvs[dsk["vg"] + "/" + dsk["adopt"]][0]))
7730

    
7731
    elif self.op.disk_template == constants.DT_BLOCK:
7732
      # Normalize and de-duplicate device paths
7733
      all_disks = set([os.path.abspath(i["adopt"]) for i in self.disks])
7734
      if len(all_disks) != len(self.disks):
7735
        raise errors.OpPrereqError("Duplicate disk names given for adoption",
7736
                                   errors.ECODE_INVAL)
7737
      baddisks = [d for d in all_disks
7738
                  if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
7739
      if baddisks:
7740
        raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
7741
                                   " cannot be adopted" %
7742
                                   (", ".join(baddisks),
7743
                                    constants.ADOPTABLE_BLOCKDEV_ROOT),
7744
                                   errors.ECODE_INVAL)
7745

    
7746
      node_disks = self.rpc.call_bdev_sizes([pnode.name],
7747
                                            list(all_disks))[pnode.name]
7748
      node_disks.Raise("Cannot get block device information from node %s" %
7749
                       pnode.name)
7750
      node_disks = node_disks.payload
7751
      delta = all_disks.difference(node_disks.keys())
7752
      if delta:
7753
        raise errors.OpPrereqError("Missing block device(s): %s" %
7754
                                   utils.CommaJoin(delta),
7755
                                   errors.ECODE_INVAL)
7756
      for dsk in self.disks:
7757
        dsk["size"] = int(float(node_disks[dsk["adopt"]]))
7758

    
7759
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
7760

    
7761
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
7762
    # check OS parameters (remotely)
7763
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
7764

    
7765
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
7766

    
7767
    # memory check on primary node
7768
    if self.op.start:
7769
      _CheckNodeFreeMemory(self, self.pnode.name,
7770
                           "creating instance %s" % self.op.instance_name,
7771
                           self.be_full[constants.BE_MEMORY],
7772
                           self.op.hypervisor)
7773

    
7774
    self.dry_run_result = list(nodenames)
7775

    
7776
  def Exec(self, feedback_fn):
7777
    """Create and add the instance to the cluster.
7778

7779
    """
7780
    instance = self.op.instance_name
7781
    pnode_name = self.pnode.name
7782

    
7783
    ht_kind = self.op.hypervisor
7784
    if ht_kind in constants.HTS_REQ_PORT:
7785
      network_port = self.cfg.AllocatePort()
7786
    else:
7787
      network_port = None
7788

    
7789
    if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE:
7790
      # this is needed because os.path.join does not accept None arguments
7791
      if self.op.file_storage_dir is None:
7792
        string_file_storage_dir = ""
7793
      else:
7794
        string_file_storage_dir = self.op.file_storage_dir
7795

    
7796
      # build the full file storage dir path
7797
      if self.op.disk_template == constants.DT_SHARED_FILE:
7798
        get_fsd_fn = self.cfg.GetSharedFileStorageDir
7799
      else:
7800
        get_fsd_fn = self.cfg.GetFileStorageDir
7801

    
7802
      file_storage_dir = utils.PathJoin(get_fsd_fn(),
7803
                                        string_file_storage_dir, instance)
7804
    else:
7805
      file_storage_dir = ""
7806

    
7807
    disks = _GenerateDiskTemplate(self,
7808
                                  self.op.disk_template,
7809
                                  instance, pnode_name,
7810
                                  self.secondaries,
7811
                                  self.disks,
7812
                                  file_storage_dir,
7813
                                  self.op.file_driver,
7814
                                  0,
7815
                                  feedback_fn)
7816

    
7817
    iobj = objects.Instance(name=instance, os=self.op.os_type,
7818
                            primary_node=pnode_name,
7819
                            nics=self.nics, disks=disks,
7820
                            disk_template=self.op.disk_template,
7821
                            admin_up=False,
7822
                            network_port=network_port,
7823
                            beparams=self.op.beparams,
7824
                            hvparams=self.op.hvparams,
7825
                            hypervisor=self.op.hypervisor,
7826
                            osparams=self.op.osparams,
7827
                            )
7828

    
7829
    if self.adopt_disks:
7830
      if self.op.disk_template == constants.DT_PLAIN:
7831
        # rename LVs to the newly-generated names; we need to construct
7832
        # 'fake' LV disks with the old data, plus the new unique_id
7833
        tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
7834
        rename_to = []
7835
        for t_dsk, a_dsk in zip (tmp_disks, self.disks):
7836
          rename_to.append(t_dsk.logical_id)
7837
          t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk["adopt"])
7838
          self.cfg.SetDiskID(t_dsk, pnode_name)
7839
        result = self.rpc.call_blockdev_rename(pnode_name,
7840
                                               zip(tmp_disks, rename_to))
7841
        result.Raise("Failed to rename adoped LVs")
7842
    else:
7843
      feedback_fn("* creating instance disks...")
7844
      try:
7845
        _CreateDisks(self, iobj)
7846
      except errors.OpExecError:
7847
        self.LogWarning("Device creation failed, reverting...")
7848
        try:
7849
          _RemoveDisks(self, iobj)
7850
        finally:
7851
          self.cfg.ReleaseDRBDMinors(instance)
7852
          raise
7853

    
7854
      if self.cfg.GetClusterInfo().prealloc_wipe_disks:
7855
        feedback_fn("* wiping instance disks...")
7856
        try:
7857
          _WipeDisks(self, iobj)
7858
        except errors.OpExecError:
7859
          self.LogWarning("Device wiping failed, reverting...")
7860
          try:
7861
            _RemoveDisks(self, iobj)
7862
          finally:
7863
            self.cfg.ReleaseDRBDMinors(instance)
7864
            raise
7865

    
7866
    feedback_fn("adding instance %s to cluster config" % instance)
7867

    
7868
    self.cfg.AddInstance(iobj, self.proc.GetECId())
7869

    
7870
    # Declare that we don't want to remove the instance lock anymore, as we've
7871
    # added the instance to the config
7872
    del self.remove_locks[locking.LEVEL_INSTANCE]
7873
    # Unlock all the nodes
7874
    if self.op.mode == constants.INSTANCE_IMPORT:
7875
      nodes_keep = [self.op.src_node]
7876
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
7877
                       if node != self.op.src_node]
7878
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
7879
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
7880
    else:
7881
      self.context.glm.release(locking.LEVEL_NODE)
7882
      del self.acquired_locks[locking.LEVEL_NODE]
7883

    
7884
    if self.op.wait_for_sync:
7885
      disk_abort = not _WaitForSync(self, iobj)
7886
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
7887
      # make sure the disks are not degraded (still sync-ing is ok)
7888
      time.sleep(15)
7889
      feedback_fn("* checking mirrors status")
7890
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
7891
    else:
7892
      disk_abort = False
7893

    
7894
    if disk_abort:
7895
      _RemoveDisks(self, iobj)
7896
      self.cfg.RemoveInstance(iobj.name)
7897
      # Make sure the instance lock gets removed
7898
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
7899
      raise errors.OpExecError("There are some degraded disks for"
7900
                               " this instance")
7901

    
7902
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
7903
      if self.op.mode == constants.INSTANCE_CREATE:
7904
        if not self.op.no_install:
7905
          feedback_fn("* running the instance OS create scripts...")
7906
          # FIXME: pass debug option from opcode to backend
7907
          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
7908
                                                 self.op.debug_level)
7909
          result.Raise("Could not add os for instance %s"
7910
                       " on node %s" % (instance, pnode_name))
7911

    
7912
      elif self.op.mode == constants.INSTANCE_IMPORT:
7913
        feedback_fn("* running the instance OS import scripts...")
7914

    
7915
        transfers = []
7916

    
7917
        for idx, image in enumerate(self.src_images):
7918
          if not image:
7919
            continue
7920

    
7921
          # FIXME: pass debug option from opcode to backend
7922
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
7923
                                             constants.IEIO_FILE, (image, ),
7924
                                             constants.IEIO_SCRIPT,
7925
                                             (iobj.disks[idx], idx),
7926
                                             None)
7927
          transfers.append(dt)
7928

    
7929
        import_result = \
7930
          masterd.instance.TransferInstanceData(self, feedback_fn,
7931
                                                self.op.src_node, pnode_name,
7932
                                                self.pnode.secondary_ip,
7933
                                                iobj, transfers)
7934
        if not compat.all(import_result):
7935
          self.LogWarning("Some disks for instance %s on node %s were not"
7936
                          " imported successfully" % (instance, pnode_name))
7937

    
7938
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7939
        feedback_fn("* preparing remote import...")
7940
        # The source cluster will stop the instance before attempting to make a
7941
        # connection. In some cases stopping an instance can take a long time,
7942
        # hence the shutdown timeout is added to the connection timeout.
7943
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
7944
                           self.op.source_shutdown_timeout)
7945
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
7946

    
7947
        assert iobj.primary_node == self.pnode.name
7948
        disk_results = \
7949
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
7950
                                        self.source_x509_ca,
7951
                                        self._cds, timeouts)
7952
        if not compat.all(disk_results):
7953
          # TODO: Should the instance still be started, even if some disks
7954
          # failed to import (valid for local imports, too)?
7955
          self.LogWarning("Some disks for instance %s on node %s were not"
7956
                          " imported successfully" % (instance, pnode_name))
7957

    
7958
        # Run rename script on newly imported instance
7959
        assert iobj.name == instance
7960
        feedback_fn("Running rename script for %s" % instance)
7961
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
7962
                                                   self.source_instance_name,
7963
                                                   self.op.debug_level)
7964
        if result.fail_msg:
7965
          self.LogWarning("Failed to run rename script for %s on node"
7966
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
7967

    
7968
      else:
7969
        # also checked in the prereq part
7970
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
7971
                                     % self.op.mode)
7972

    
7973
    if self.op.start:
7974
      iobj.admin_up = True
7975
      self.cfg.Update(iobj, feedback_fn)
7976
      logging.info("Starting instance %s on node %s", instance, pnode_name)
7977
      feedback_fn("* starting instance...")
7978
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
7979
      result.Raise("Could not start instance")
7980

    
7981
    return list(iobj.all_nodes)
7982

    
7983

    
7984
class LUInstanceConsole(NoHooksLU):
7985
  """Connect to an instance's console.
7986

7987
  This is somewhat special in that it returns the command line that
7988
  you need to run on the master node in order to connect to the
7989
  console.
7990

7991
  """
7992
  REQ_BGL = False
7993

    
7994
  def ExpandNames(self):
7995
    self._ExpandAndLockInstance()
7996

    
7997
  def CheckPrereq(self):
7998
    """Check prerequisites.
7999

8000
    This checks that the instance is in the cluster.
8001

8002
    """
8003
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8004
    assert self.instance is not None, \
8005
      "Cannot retrieve locked instance %s" % self.op.instance_name
8006
    _CheckNodeOnline(self, self.instance.primary_node)
8007

    
8008
  def Exec(self, feedback_fn):
8009
    """Connect to the console of an instance
8010

8011
    """
8012
    instance = self.instance
8013
    node = instance.primary_node
8014

    
8015
    node_insts = self.rpc.call_instance_list([node],
8016
                                             [instance.hypervisor])[node]
8017
    node_insts.Raise("Can't get node information from %s" % node)
8018

    
8019
    if instance.name not in node_insts.payload:
8020
      if instance.admin_up:
8021
        state = "ERROR_down"
8022
      else:
8023
        state = "ADMIN_down"
8024
      raise errors.OpExecError("Instance %s is not running (state %s)" %
8025
                               (instance.name, state))
8026

    
8027
    logging.debug("Connecting to console of %s on %s", instance.name, node)
8028

    
8029
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
8030

    
8031

    
8032
def _GetInstanceConsole(cluster, instance):
8033
  """Returns console information for an instance.
8034

8035
  @type cluster: L{objects.Cluster}
8036
  @type instance: L{objects.Instance}
8037
  @rtype: dict
8038

8039
  """
8040
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
8041
  # beparams and hvparams are passed separately, to avoid editing the
8042
  # instance and then saving the defaults in the instance itself.
8043
  hvparams = cluster.FillHV(instance)
8044
  beparams = cluster.FillBE(instance)
8045
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
8046

    
8047
  assert console.instance == instance.name
8048
  assert console.Validate()
8049

    
8050
  return console.ToDict()
8051

    
8052

    
8053
class LUInstanceReplaceDisks(LogicalUnit):
8054
  """Replace the disks of an instance.
8055

8056
  """
8057
  HPATH = "mirrors-replace"
8058
  HTYPE = constants.HTYPE_INSTANCE
8059
  REQ_BGL = False
8060

    
8061
  def CheckArguments(self):
8062
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
8063
                                  self.op.iallocator)
8064

    
8065
  def ExpandNames(self):
8066
    self._ExpandAndLockInstance()
8067

    
8068
    if self.op.iallocator is not None:
8069
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8070

    
8071
    elif self.op.remote_node is not None:
8072
      remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8073
      self.op.remote_node = remote_node
8074

    
8075
      # Warning: do not remove the locking of the new secondary here
8076
      # unless DRBD8.AddChildren is changed to work in parallel;
8077
      # currently it doesn't since parallel invocations of
8078
      # FindUnusedMinor will conflict
8079
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
8080
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
8081

    
8082
    else:
8083
      self.needed_locks[locking.LEVEL_NODE] = []
8084
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8085

    
8086
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
8087
                                   self.op.iallocator, self.op.remote_node,
8088
                                   self.op.disks, False, self.op.early_release)
8089

    
8090
    self.tasklets = [self.replacer]
8091

    
8092
  def DeclareLocks(self, level):
8093
    # If we're not already locking all nodes in the set we have to declare the
8094
    # instance's primary/secondary nodes.
8095
    if (level == locking.LEVEL_NODE and
8096
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
8097
      self._LockInstancesNodes()
8098

    
8099
  def BuildHooksEnv(self):
8100
    """Build hooks env.
8101

8102
    This runs on the master, the primary and all the secondaries.
8103

8104
    """
8105
    instance = self.replacer.instance
8106
    env = {
8107
      "MODE": self.op.mode,
8108
      "NEW_SECONDARY": self.op.remote_node,
8109
      "OLD_SECONDARY": instance.secondary_nodes[0],
8110
      }
8111
    env.update(_BuildInstanceHookEnvByObject(self, instance))
8112
    nl = [
8113
      self.cfg.GetMasterNode(),
8114
      instance.primary_node,
8115
      ]
8116
    if self.op.remote_node is not None:
8117
      nl.append(self.op.remote_node)
8118
    return env, nl, nl
8119

    
8120

    
8121
class TLReplaceDisks(Tasklet):
8122
  """Replaces disks for an instance.
8123

8124
  Note: Locking is not within the scope of this class.
8125

8126
  """
8127
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
8128
               disks, delay_iallocator, early_release):
8129
    """Initializes this class.
8130

8131
    """
8132
    Tasklet.__init__(self, lu)
8133

    
8134
    # Parameters
8135
    self.instance_name = instance_name
8136
    self.mode = mode
8137
    self.iallocator_name = iallocator_name
8138
    self.remote_node = remote_node
8139
    self.disks = disks
8140
    self.delay_iallocator = delay_iallocator
8141
    self.early_release = early_release
8142

    
8143
    # Runtime data
8144
    self.instance = None
8145
    self.new_node = None
8146
    self.target_node = None
8147
    self.other_node = None
8148
    self.remote_node_info = None
8149
    self.node_secondary_ip = None
8150

    
8151
  @staticmethod
8152
  def CheckArguments(mode, remote_node, iallocator):
8153
    """Helper function for users of this class.
8154

8155
    """
8156
    # check for valid parameter combination
8157
    if mode == constants.REPLACE_DISK_CHG:
8158
      if remote_node is None and iallocator is None:
8159
        raise errors.OpPrereqError("When changing the secondary either an"
8160
                                   " iallocator script must be used or the"
8161
                                   " new node given", errors.ECODE_INVAL)
8162

    
8163
      if remote_node is not None and iallocator is not None:
8164
        raise errors.OpPrereqError("Give either the iallocator or the new"
8165
                                   " secondary, not both", errors.ECODE_INVAL)
8166

    
8167
    elif remote_node is not None or iallocator is not None:
8168
      # Not replacing the secondary
8169
      raise errors.OpPrereqError("The iallocator and new node options can"
8170
                                 " only be used when changing the"
8171
                                 " secondary node", errors.ECODE_INVAL)
8172

    
8173
  @staticmethod
8174
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
8175
    """Compute a new secondary node using an IAllocator.
8176

8177
    """
8178
    ial = IAllocator(lu.cfg, lu.rpc,
8179
                     mode=constants.IALLOCATOR_MODE_RELOC,
8180
                     name=instance_name,
8181
                     relocate_from=relocate_from)
8182

    
8183
    ial.Run(iallocator_name)
8184

    
8185
    if not ial.success:
8186
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
8187
                                 " %s" % (iallocator_name, ial.info),
8188
                                 errors.ECODE_NORES)
8189

    
8190
    if len(ial.result) != ial.required_nodes:
8191
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8192
                                 " of nodes (%s), required %s" %
8193
                                 (iallocator_name,
8194
                                  len(ial.result), ial.required_nodes),
8195
                                 errors.ECODE_FAULT)
8196

    
8197
    remote_node_name = ial.result[0]
8198

    
8199
    lu.LogInfo("Selected new secondary for instance '%s': %s",
8200
               instance_name, remote_node_name)
8201

    
8202
    return remote_node_name
8203

    
8204
  def _FindFaultyDisks(self, node_name):
8205
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
8206
                                    node_name, True)
8207

    
8208
  def CheckPrereq(self):
8209
    """Check prerequisites.
8210

8211
    This checks that the instance is in the cluster.
8212

8213
    """
8214
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
8215
    assert instance is not None, \
8216
      "Cannot retrieve locked instance %s" % self.instance_name
8217

    
8218
    if instance.disk_template != constants.DT_DRBD8:
8219
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
8220
                                 " instances", errors.ECODE_INVAL)
8221

    
8222
    if len(instance.secondary_nodes) != 1:
8223
      raise errors.OpPrereqError("The instance has a strange layout,"
8224
                                 " expected one secondary but found %d" %
8225
                                 len(instance.secondary_nodes),
8226
                                 errors.ECODE_FAULT)
8227

    
8228
    if not self.delay_iallocator:
8229
      self._CheckPrereq2()
8230

    
8231
  def _CheckPrereq2(self):
8232
    """Check prerequisites, second part.
8233

8234
    This function should always be part of CheckPrereq. It was separated and is
8235
    now called from Exec because during node evacuation iallocator was only
8236
    called with an unmodified cluster model, not taking planned changes into
8237
    account.
8238

8239
    """
8240
    instance = self.instance
8241
    secondary_node = instance.secondary_nodes[0]
8242

    
8243
    if self.iallocator_name is None:
8244
      remote_node = self.remote_node
8245
    else:
8246
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
8247
                                       instance.name, instance.secondary_nodes)
8248

    
8249
    if remote_node is not None:
8250
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
8251
      assert self.remote_node_info is not None, \
8252
        "Cannot retrieve locked node %s" % remote_node
8253
    else:
8254
      self.remote_node_info = None
8255

    
8256
    if remote_node == self.instance.primary_node:
8257
      raise errors.OpPrereqError("The specified node is the primary node of"
8258
                                 " the instance.", errors.ECODE_INVAL)
8259

    
8260
    if remote_node == secondary_node:
8261
      raise errors.OpPrereqError("The specified node is already the"
8262
                                 " secondary node of the instance.",
8263
                                 errors.ECODE_INVAL)
8264

    
8265
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
8266
                                    constants.REPLACE_DISK_CHG):
8267
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
8268
                                 errors.ECODE_INVAL)
8269

    
8270
    if self.mode == constants.REPLACE_DISK_AUTO:
8271
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
8272
      faulty_secondary = self._FindFaultyDisks(secondary_node)
8273

    
8274
      if faulty_primary and faulty_secondary:
8275
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
8276
                                   " one node and can not be repaired"
8277
                                   " automatically" % self.instance_name,
8278
                                   errors.ECODE_STATE)
8279

    
8280
      if faulty_primary:
8281
        self.disks = faulty_primary
8282
        self.target_node = instance.primary_node
8283
        self.other_node = secondary_node
8284
        check_nodes = [self.target_node, self.other_node]
8285
      elif faulty_secondary:
8286
        self.disks = faulty_secondary
8287
        self.target_node = secondary_node
8288
        self.other_node = instance.primary_node
8289
        check_nodes = [self.target_node, self.other_node]
8290
      else:
8291
        self.disks = []
8292
        check_nodes = []
8293

    
8294
    else:
8295
      # Non-automatic modes
8296
      if self.mode == constants.REPLACE_DISK_PRI:
8297
        self.target_node = instance.primary_node
8298
        self.other_node = secondary_node
8299
        check_nodes = [self.target_node, self.other_node]
8300

    
8301
      elif self.mode == constants.REPLACE_DISK_SEC:
8302
        self.target_node = secondary_node
8303
        self.other_node = instance.primary_node
8304
        check_nodes = [self.target_node, self.other_node]
8305

    
8306
      elif self.mode == constants.REPLACE_DISK_CHG:
8307
        self.new_node = remote_node
8308
        self.other_node = instance.primary_node
8309
        self.target_node = secondary_node
8310
        check_nodes = [self.new_node, self.other_node]
8311

    
8312
        _CheckNodeNotDrained(self.lu, remote_node)
8313
        _CheckNodeVmCapable(self.lu, remote_node)
8314

    
8315
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
8316
        assert old_node_info is not None
8317
        if old_node_info.offline and not self.early_release:
8318
          # doesn't make sense to delay the release
8319
          self.early_release = True
8320
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
8321
                          " early-release mode", secondary_node)
8322

    
8323
      else:
8324
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
8325
                                     self.mode)
8326

    
8327
      # If not specified all disks should be replaced
8328
      if not self.disks:
8329
        self.disks = range(len(self.instance.disks))
8330

    
8331
    for node in check_nodes:
8332
      _CheckNodeOnline(self.lu, node)
8333

    
8334
    # Check whether disks are valid
8335
    for disk_idx in self.disks:
8336
      instance.FindDisk(disk_idx)
8337

    
8338
    # Get secondary node IP addresses
8339
    node_2nd_ip = {}
8340

    
8341
    for node_name in [self.target_node, self.other_node, self.new_node]:
8342
      if node_name is not None:
8343
        node_2nd_ip[node_name] = self.cfg.GetNodeInfo(node_name).secondary_ip
8344

    
8345
    self.node_secondary_ip = node_2nd_ip
8346

    
8347
  def Exec(self, feedback_fn):
8348
    """Execute disk replacement.
8349

8350
    This dispatches the disk replacement to the appropriate handler.
8351

8352
    """
8353
    if self.delay_iallocator:
8354
      self._CheckPrereq2()
8355

    
8356
    if not self.disks:
8357
      feedback_fn("No disks need replacement")
8358
      return
8359

    
8360
    feedback_fn("Replacing disk(s) %s for %s" %
8361
                (utils.CommaJoin(self.disks), self.instance.name))
8362

    
8363
    activate_disks = (not self.instance.admin_up)
8364

    
8365
    # Activate the instance disks if we're replacing them on a down instance
8366
    if activate_disks:
8367
      _StartInstanceDisks(self.lu, self.instance, True)
8368

    
8369
    try:
8370
      # Should we replace the secondary node?
8371
      if self.new_node is not None:
8372
        fn = self._ExecDrbd8Secondary
8373
      else:
8374
        fn = self._ExecDrbd8DiskOnly
8375

    
8376
      return fn(feedback_fn)
8377

    
8378
    finally:
8379
      # Deactivate the instance disks if we're replacing them on a
8380
      # down instance
8381
      if activate_disks:
8382
        _SafeShutdownInstanceDisks(self.lu, self.instance)
8383

    
8384
  def _CheckVolumeGroup(self, nodes):
8385
    self.lu.LogInfo("Checking volume groups")
8386

    
8387
    vgname = self.cfg.GetVGName()
8388

    
8389
    # Make sure volume group exists on all involved nodes
8390
    results = self.rpc.call_vg_list(nodes)
8391
    if not results:
8392
      raise errors.OpExecError("Can't list volume groups on the nodes")
8393

    
8394
    for node in nodes:
8395
      res = results[node]
8396
      res.Raise("Error checking node %s" % node)
8397
      if vgname not in res.payload:
8398
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
8399
                                 (vgname, node))
8400

    
8401
  def _CheckDisksExistence(self, nodes):
8402
    # Check disk existence
8403
    for idx, dev in enumerate(self.instance.disks):
8404
      if idx not in self.disks:
8405
        continue
8406

    
8407
      for node in nodes:
8408
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
8409
        self.cfg.SetDiskID(dev, node)
8410

    
8411
        result = self.rpc.call_blockdev_find(node, dev)
8412

    
8413
        msg = result.fail_msg
8414
        if msg or not result.payload:
8415
          if not msg:
8416
            msg = "disk not found"
8417
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
8418
                                   (idx, node, msg))
8419

    
8420
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
8421
    for idx, dev in enumerate(self.instance.disks):
8422
      if idx not in self.disks:
8423
        continue
8424

    
8425
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
8426
                      (idx, node_name))
8427

    
8428
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
8429
                                   ldisk=ldisk):
8430
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
8431
                                 " replace disks for instance %s" %
8432
                                 (node_name, self.instance.name))
8433

    
8434
  def _CreateNewStorage(self, node_name):
8435
    vgname = self.cfg.GetVGName()
8436
    iv_names = {}
8437

    
8438
    for idx, dev in enumerate(self.instance.disks):
8439
      if idx not in self.disks:
8440
        continue
8441

    
8442
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
8443

    
8444
      self.cfg.SetDiskID(dev, node_name)
8445

    
8446
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
8447
      names = _GenerateUniqueNames(self.lu, lv_names)
8448

    
8449
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
8450
                             logical_id=(vgname, names[0]))
8451
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
8452
                             logical_id=(vgname, names[1]))
8453

    
8454
      new_lvs = [lv_data, lv_meta]
8455
      old_lvs = dev.children
8456
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
8457

    
8458
      # we pass force_create=True to force the LVM creation
8459
      for new_lv in new_lvs:
8460
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
8461
                        _GetInstanceInfoText(self.instance), False)
8462

    
8463
    return iv_names
8464

    
8465
  def _CheckDevices(self, node_name, iv_names):
8466
    for name, (dev, _, _) in iv_names.iteritems():
8467
      self.cfg.SetDiskID(dev, node_name)
8468

    
8469
      result = self.rpc.call_blockdev_find(node_name, dev)
8470

    
8471
      msg = result.fail_msg
8472
      if msg or not result.payload:
8473
        if not msg:
8474
          msg = "disk not found"
8475
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
8476
                                 (name, msg))
8477

    
8478
      if result.payload.is_degraded:
8479
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
8480

    
8481
  def _RemoveOldStorage(self, node_name, iv_names):
8482
    for name, (_, old_lvs, _) in iv_names.iteritems():
8483
      self.lu.LogInfo("Remove logical volumes for %s" % name)
8484

    
8485
      for lv in old_lvs:
8486
        self.cfg.SetDiskID(lv, node_name)
8487

    
8488
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
8489
        if msg:
8490
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
8491
                             hint="remove unused LVs manually")
8492

    
8493
  def _ReleaseNodeLock(self, node_name):
8494
    """Releases the lock for a given node."""
8495
    self.lu.context.glm.release(locking.LEVEL_NODE, node_name)
8496

    
8497
  def _ExecDrbd8DiskOnly(self, feedback_fn):
8498
    """Replace a disk on the primary or secondary for DRBD 8.
8499

8500
    The algorithm for replace is quite complicated:
8501

8502
      1. for each disk to be replaced:
8503

8504
        1. create new LVs on the target node with unique names
8505
        1. detach old LVs from the drbd device
8506
        1. rename old LVs to name_replaced.<time_t>
8507
        1. rename new LVs to old LVs
8508
        1. attach the new LVs (with the old names now) to the drbd device
8509

8510
      1. wait for sync across all devices
8511

8512
      1. for each modified disk:
8513

8514
        1. remove old LVs (which have the name name_replaces.<time_t>)
8515

8516
    Failures are not very well handled.
8517

8518
    """
8519
    steps_total = 6
8520

    
8521
    # Step: check device activation
8522
    self.lu.LogStep(1, steps_total, "Check device existence")
8523
    self._CheckDisksExistence([self.other_node, self.target_node])
8524
    self._CheckVolumeGroup([self.target_node, self.other_node])
8525

    
8526
    # Step: check other node consistency
8527
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8528
    self._CheckDisksConsistency(self.other_node,
8529
                                self.other_node == self.instance.primary_node,
8530
                                False)
8531

    
8532
    # Step: create new storage
8533
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8534
    iv_names = self._CreateNewStorage(self.target_node)
8535

    
8536
    # Step: for each lv, detach+rename*2+attach
8537
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8538
    for dev, old_lvs, new_lvs in iv_names.itervalues():
8539
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
8540

    
8541
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
8542
                                                     old_lvs)
8543
      result.Raise("Can't detach drbd from local storage on node"
8544
                   " %s for device %s" % (self.target_node, dev.iv_name))
8545
      #dev.children = []
8546
      #cfg.Update(instance)
8547

    
8548
      # ok, we created the new LVs, so now we know we have the needed
8549
      # storage; as such, we proceed on the target node to rename
8550
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
8551
      # using the assumption that logical_id == physical_id (which in
8552
      # turn is the unique_id on that node)
8553

    
8554
      # FIXME(iustin): use a better name for the replaced LVs
8555
      temp_suffix = int(time.time())
8556
      ren_fn = lambda d, suff: (d.physical_id[0],
8557
                                d.physical_id[1] + "_replaced-%s" % suff)
8558

    
8559
      # Build the rename list based on what LVs exist on the node
8560
      rename_old_to_new = []
8561
      for to_ren in old_lvs:
8562
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
8563
        if not result.fail_msg and result.payload:
8564
          # device exists
8565
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
8566

    
8567
      self.lu.LogInfo("Renaming the old LVs on the target node")
8568
      result = self.rpc.call_blockdev_rename(self.target_node,
8569
                                             rename_old_to_new)
8570
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
8571

    
8572
      # Now we rename the new LVs to the old LVs
8573
      self.lu.LogInfo("Renaming the new LVs on the target node")
8574
      rename_new_to_old = [(new, old.physical_id)
8575
                           for old, new in zip(old_lvs, new_lvs)]
8576
      result = self.rpc.call_blockdev_rename(self.target_node,
8577
                                             rename_new_to_old)
8578
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
8579

    
8580
      for old, new in zip(old_lvs, new_lvs):
8581
        new.logical_id = old.logical_id
8582
        self.cfg.SetDiskID(new, self.target_node)
8583

    
8584
      for disk in old_lvs:
8585
        disk.logical_id = ren_fn(disk, temp_suffix)
8586
        self.cfg.SetDiskID(disk, self.target_node)
8587

    
8588
      # Now that the new lvs have the old name, we can add them to the device
8589
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
8590
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
8591
                                                  new_lvs)
8592
      msg = result.fail_msg
8593
      if msg:
8594
        for new_lv in new_lvs:
8595
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
8596
                                               new_lv).fail_msg
8597
          if msg2:
8598
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
8599
                               hint=("cleanup manually the unused logical"
8600
                                     "volumes"))
8601
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
8602

    
8603
      dev.children = new_lvs
8604

    
8605
      self.cfg.Update(self.instance, feedback_fn)
8606

    
8607
    cstep = 5
8608
    if self.early_release:
8609
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8610
      cstep += 1
8611
      self._RemoveOldStorage(self.target_node, iv_names)
8612
      # WARNING: we release both node locks here, do not do other RPCs
8613
      # than WaitForSync to the primary node
8614
      self._ReleaseNodeLock([self.target_node, self.other_node])
8615

    
8616
    # Wait for sync
8617
    # This can fail as the old devices are degraded and _WaitForSync
8618
    # does a combined result over all disks, so we don't check its return value
8619
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8620
    cstep += 1
8621
    _WaitForSync(self.lu, self.instance)
8622

    
8623
    # Check all devices manually
8624
    self._CheckDevices(self.instance.primary_node, iv_names)
8625

    
8626
    # Step: remove old storage
8627
    if not self.early_release:
8628
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8629
      cstep += 1
8630
      self._RemoveOldStorage(self.target_node, iv_names)
8631

    
8632
  def _ExecDrbd8Secondary(self, feedback_fn):
8633
    """Replace the secondary node for DRBD 8.
8634

8635
    The algorithm for replace is quite complicated:
8636
      - for all disks of the instance:
8637
        - create new LVs on the new node with same names
8638
        - shutdown the drbd device on the old secondary
8639
        - disconnect the drbd network on the primary
8640
        - create the drbd device on the new secondary
8641
        - network attach the drbd on the primary, using an artifice:
8642
          the drbd code for Attach() will connect to the network if it
8643
          finds a device which is connected to the good local disks but
8644
          not network enabled
8645
      - wait for sync across all devices
8646
      - remove all disks from the old secondary
8647

8648
    Failures are not very well handled.
8649

8650
    """
8651
    steps_total = 6
8652

    
8653
    # Step: check device activation
8654
    self.lu.LogStep(1, steps_total, "Check device existence")
8655
    self._CheckDisksExistence([self.instance.primary_node])
8656
    self._CheckVolumeGroup([self.instance.primary_node])
8657

    
8658
    # Step: check other node consistency
8659
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8660
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
8661

    
8662
    # Step: create new storage
8663
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8664
    for idx, dev in enumerate(self.instance.disks):
8665
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
8666
                      (self.new_node, idx))
8667
      # we pass force_create=True to force LVM creation
8668
      for new_lv in dev.children:
8669
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
8670
                        _GetInstanceInfoText(self.instance), False)
8671

    
8672
    # Step 4: dbrd minors and drbd setups changes
8673
    # after this, we must manually remove the drbd minors on both the
8674
    # error and the success paths
8675
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8676
    minors = self.cfg.AllocateDRBDMinor([self.new_node
8677
                                         for dev in self.instance.disks],
8678
                                        self.instance.name)
8679
    logging.debug("Allocated minors %r", minors)
8680

    
8681
    iv_names = {}
8682
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
8683
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
8684
                      (self.new_node, idx))
8685
      # create new devices on new_node; note that we create two IDs:
8686
      # one without port, so the drbd will be activated without
8687
      # networking information on the new node at this stage, and one
8688
      # with network, for the latter activation in step 4
8689
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
8690
      if self.instance.primary_node == o_node1:
8691
        p_minor = o_minor1
8692
      else:
8693
        assert self.instance.primary_node == o_node2, "Three-node instance?"
8694
        p_minor = o_minor2
8695

    
8696
      new_alone_id = (self.instance.primary_node, self.new_node, None,
8697
                      p_minor, new_minor, o_secret)
8698
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
8699
                    p_minor, new_minor, o_secret)
8700

    
8701
      iv_names[idx] = (dev, dev.children, new_net_id)
8702
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
8703
                    new_net_id)
8704
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
8705
                              logical_id=new_alone_id,
8706
                              children=dev.children,
8707
                              size=dev.size)
8708
      try:
8709
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
8710
                              _GetInstanceInfoText(self.instance), False)
8711
      except errors.GenericError:
8712
        self.cfg.ReleaseDRBDMinors(self.instance.name)
8713
        raise
8714

    
8715
    # We have new devices, shutdown the drbd on the old secondary
8716
    for idx, dev in enumerate(self.instance.disks):
8717
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
8718
      self.cfg.SetDiskID(dev, self.target_node)
8719
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
8720
      if msg:
8721
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
8722
                           "node: %s" % (idx, msg),
8723
                           hint=("Please cleanup this device manually as"
8724
                                 " soon as possible"))
8725

    
8726
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
8727
    result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
8728
                                               self.node_secondary_ip,
8729
                                               self.instance.disks)\
8730
                                              [self.instance.primary_node]
8731

    
8732
    msg = result.fail_msg
8733
    if msg:
8734
      # detaches didn't succeed (unlikely)
8735
      self.cfg.ReleaseDRBDMinors(self.instance.name)
8736
      raise errors.OpExecError("Can't detach the disks from the network on"
8737
                               " old node: %s" % (msg,))
8738

    
8739
    # if we managed to detach at least one, we update all the disks of
8740
    # the instance to point to the new secondary
8741
    self.lu.LogInfo("Updating instance configuration")
8742
    for dev, _, new_logical_id in iv_names.itervalues():
8743
      dev.logical_id = new_logical_id
8744
      self.cfg.SetDiskID(dev, self.instance.primary_node)
8745

    
8746
    self.cfg.Update(self.instance, feedback_fn)
8747

    
8748
    # and now perform the drbd attach
8749
    self.lu.LogInfo("Attaching primary drbds to new secondary"
8750
                    " (standalone => connected)")
8751
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
8752
                                            self.new_node],
8753
                                           self.node_secondary_ip,
8754
                                           self.instance.disks,
8755
                                           self.instance.name,
8756
                                           False)
8757
    for to_node, to_result in result.items():
8758
      msg = to_result.fail_msg
8759
      if msg:
8760
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
8761
                           to_node, msg,
8762
                           hint=("please do a gnt-instance info to see the"
8763
                                 " status of disks"))
8764
    cstep = 5
8765
    if self.early_release:
8766
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8767
      cstep += 1
8768
      self._RemoveOldStorage(self.target_node, iv_names)
8769
      # WARNING: we release all node locks here, do not do other RPCs
8770
      # than WaitForSync to the primary node
8771
      self._ReleaseNodeLock([self.instance.primary_node,
8772
                             self.target_node,
8773
                             self.new_node])
8774

    
8775
    # Wait for sync
8776
    # This can fail as the old devices are degraded and _WaitForSync
8777
    # does a combined result over all disks, so we don't check its return value
8778
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8779
    cstep += 1
8780
    _WaitForSync(self.lu, self.instance)
8781

    
8782
    # Check all devices manually
8783
    self._CheckDevices(self.instance.primary_node, iv_names)
8784

    
8785
    # Step: remove old storage
8786
    if not self.early_release:
8787
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8788
      self._RemoveOldStorage(self.target_node, iv_names)
8789

    
8790

    
8791
class LURepairNodeStorage(NoHooksLU):
8792
  """Repairs the volume group on a node.
8793

8794
  """
8795
  REQ_BGL = False
8796

    
8797
  def CheckArguments(self):
8798
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
8799

    
8800
    storage_type = self.op.storage_type
8801

    
8802
    if (constants.SO_FIX_CONSISTENCY not in
8803
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
8804
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
8805
                                 " repaired" % storage_type,
8806
                                 errors.ECODE_INVAL)
8807

    
8808
  def ExpandNames(self):
8809
    self.needed_locks = {
8810
      locking.LEVEL_NODE: [self.op.node_name],
8811
      }
8812

    
8813
  def _CheckFaultyDisks(self, instance, node_name):
8814
    """Ensure faulty disks abort the opcode or at least warn."""
8815
    try:
8816
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
8817
                                  node_name, True):
8818
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
8819
                                   " node '%s'" % (instance.name, node_name),
8820
                                   errors.ECODE_STATE)
8821
    except errors.OpPrereqError, err:
8822
      if self.op.ignore_consistency:
8823
        self.proc.LogWarning(str(err.args[0]))
8824
      else:
8825
        raise
8826

    
8827
  def CheckPrereq(self):
8828
    """Check prerequisites.
8829

8830
    """
8831
    # Check whether any instance on this node has faulty disks
8832
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
8833
      if not inst.admin_up:
8834
        continue
8835
      check_nodes = set(inst.all_nodes)
8836
      check_nodes.discard(self.op.node_name)
8837
      for inst_node_name in check_nodes:
8838
        self._CheckFaultyDisks(inst, inst_node_name)
8839

    
8840
  def Exec(self, feedback_fn):
8841
    feedback_fn("Repairing storage unit '%s' on %s ..." %
8842
                (self.op.name, self.op.node_name))
8843

    
8844
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
8845
    result = self.rpc.call_storage_execute(self.op.node_name,
8846
                                           self.op.storage_type, st_args,
8847
                                           self.op.name,
8848
                                           constants.SO_FIX_CONSISTENCY)
8849
    result.Raise("Failed to repair storage unit '%s' on %s" %
8850
                 (self.op.name, self.op.node_name))
8851

    
8852

    
8853
class LUNodeEvacStrategy(NoHooksLU):
8854
  """Computes the node evacuation strategy.
8855

8856
  """
8857
  REQ_BGL = False
8858

    
8859
  def CheckArguments(self):
8860
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
8861

    
8862
  def ExpandNames(self):
8863
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
8864
    self.needed_locks = locks = {}
8865
    if self.op.remote_node is None:
8866
      locks[locking.LEVEL_NODE] = locking.ALL_SET
8867
    else:
8868
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8869
      locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
8870

    
8871
  def Exec(self, feedback_fn):
8872
    if self.op.remote_node is not None:
8873
      instances = []
8874
      for node in self.op.nodes:
8875
        instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
8876
      result = []
8877
      for i in instances:
8878
        if i.primary_node == self.op.remote_node:
8879
          raise errors.OpPrereqError("Node %s is the primary node of"
8880
                                     " instance %s, cannot use it as"
8881
                                     " secondary" %
8882
                                     (self.op.remote_node, i.name),
8883
                                     errors.ECODE_INVAL)
8884
        result.append([i.name, self.op.remote_node])
8885
    else:
8886
      ial = IAllocator(self.cfg, self.rpc,
8887
                       mode=constants.IALLOCATOR_MODE_MEVAC,
8888
                       evac_nodes=self.op.nodes)
8889
      ial.Run(self.op.iallocator, validate=True)
8890
      if not ial.success:
8891
        raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
8892
                                 errors.ECODE_NORES)
8893
      result = ial.result
8894
    return result
8895

    
8896

    
8897
class LUInstanceGrowDisk(LogicalUnit):
8898
  """Grow a disk of an instance.
8899

8900
  """
8901
  HPATH = "disk-grow"
8902
  HTYPE = constants.HTYPE_INSTANCE
8903
  REQ_BGL = False
8904

    
8905
  def ExpandNames(self):
8906
    self._ExpandAndLockInstance()
8907
    self.needed_locks[locking.LEVEL_NODE] = []
8908
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8909

    
8910
  def DeclareLocks(self, level):
8911
    if level == locking.LEVEL_NODE:
8912
      self._LockInstancesNodes()
8913

    
8914
  def BuildHooksEnv(self):
8915
    """Build hooks env.
8916

8917
    This runs on the master, the primary and all the secondaries.
8918

8919
    """
8920
    env = {
8921
      "DISK": self.op.disk,
8922
      "AMOUNT": self.op.amount,
8923
      }
8924
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
8925
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
8926
    return env, nl, nl
8927

    
8928
  def CheckPrereq(self):
8929
    """Check prerequisites.
8930

8931
    This checks that the instance is in the cluster.
8932

8933
    """
8934
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8935
    assert instance is not None, \
8936
      "Cannot retrieve locked instance %s" % self.op.instance_name
8937
    nodenames = list(instance.all_nodes)
8938
    for node in nodenames:
8939
      _CheckNodeOnline(self, node)
8940

    
8941
    self.instance = instance
8942

    
8943
    if instance.disk_template not in constants.DTS_GROWABLE:
8944
      raise errors.OpPrereqError("Instance's disk layout does not support"
8945
                                 " growing.", errors.ECODE_INVAL)
8946

    
8947
    self.disk = instance.FindDisk(self.op.disk)
8948

    
8949
    if instance.disk_template not in (constants.DT_FILE,
8950
                                      constants.DT_SHARED_FILE):
8951
      # TODO: check the free disk space for file, when that feature will be
8952
      # supported
8953
      _CheckNodesFreeDiskPerVG(self, nodenames,
8954
                               self.disk.ComputeGrowth(self.op.amount))
8955

    
8956
  def Exec(self, feedback_fn):
8957
    """Execute disk grow.
8958

8959
    """
8960
    instance = self.instance
8961
    disk = self.disk
8962

    
8963
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
8964
    if not disks_ok:
8965
      raise errors.OpExecError("Cannot activate block device to grow")
8966

    
8967
    for node in instance.all_nodes:
8968
      self.cfg.SetDiskID(disk, node)
8969
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
8970
      result.Raise("Grow request failed to node %s" % node)
8971

    
8972
      # TODO: Rewrite code to work properly
8973
      # DRBD goes into sync mode for a short amount of time after executing the
8974
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
8975
      # calling "resize" in sync mode fails. Sleeping for a short amount of
8976
      # time is a work-around.
8977
      time.sleep(5)
8978

    
8979
    disk.RecordGrow(self.op.amount)
8980
    self.cfg.Update(instance, feedback_fn)
8981
    if self.op.wait_for_sync:
8982
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
8983
      if disk_abort:
8984
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
8985
                             " status.\nPlease check the instance.")
8986
      if not instance.admin_up:
8987
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
8988
    elif not instance.admin_up:
8989
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
8990
                           " not supposed to be running because no wait for"
8991
                           " sync mode was requested.")
8992

    
8993

    
8994
class LUInstanceQueryData(NoHooksLU):
8995
  """Query runtime instance data.
8996

8997
  """
8998
  REQ_BGL = False
8999

    
9000
  def ExpandNames(self):
9001
    self.needed_locks = {}
9002
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
9003

    
9004
    if self.op.instances:
9005
      self.wanted_names = []
9006
      for name in self.op.instances:
9007
        full_name = _ExpandInstanceName(self.cfg, name)
9008
        self.wanted_names.append(full_name)
9009
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
9010
    else:
9011
      self.wanted_names = None
9012
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
9013

    
9014
    self.needed_locks[locking.LEVEL_NODE] = []
9015
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9016

    
9017
  def DeclareLocks(self, level):
9018
    if level == locking.LEVEL_NODE:
9019
      self._LockInstancesNodes()
9020

    
9021
  def CheckPrereq(self):
9022
    """Check prerequisites.
9023

9024
    This only checks the optional instance list against the existing names.
9025

9026
    """
9027
    if self.wanted_names is None:
9028
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
9029

    
9030
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
9031
                             in self.wanted_names]
9032

    
9033
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
9034
    """Returns the status of a block device
9035

9036
    """
9037
    if self.op.static or not node:
9038
      return None
9039

    
9040
    self.cfg.SetDiskID(dev, node)
9041

    
9042
    result = self.rpc.call_blockdev_find(node, dev)
9043
    if result.offline:
9044
      return None
9045

    
9046
    result.Raise("Can't compute disk status for %s" % instance_name)
9047

    
9048
    status = result.payload
9049
    if status is None:
9050
      return None
9051

    
9052
    return (status.dev_path, status.major, status.minor,
9053
            status.sync_percent, status.estimated_time,
9054
            status.is_degraded, status.ldisk_status)
9055

    
9056
  def _ComputeDiskStatus(self, instance, snode, dev):
9057
    """Compute block device status.
9058

9059
    """
9060
    if dev.dev_type in constants.LDS_DRBD:
9061
      # we change the snode then (otherwise we use the one passed in)
9062
      if dev.logical_id[0] == instance.primary_node:
9063
        snode = dev.logical_id[1]
9064
      else:
9065
        snode = dev.logical_id[0]
9066

    
9067
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
9068
                                              instance.name, dev)
9069
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
9070

    
9071
    if dev.children:
9072
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
9073
                      for child in dev.children]
9074
    else:
9075
      dev_children = []
9076

    
9077
    data = {
9078
      "iv_name": dev.iv_name,
9079
      "dev_type": dev.dev_type,
9080
      "logical_id": dev.logical_id,
9081
      "physical_id": dev.physical_id,
9082
      "pstatus": dev_pstatus,
9083
      "sstatus": dev_sstatus,
9084
      "children": dev_children,
9085
      "mode": dev.mode,
9086
      "size": dev.size,
9087
      }
9088

    
9089
    return data
9090

    
9091
  def Exec(self, feedback_fn):
9092
    """Gather and return data"""
9093
    result = {}
9094

    
9095
    cluster = self.cfg.GetClusterInfo()
9096

    
9097
    for instance in self.wanted_instances:
9098
      if not self.op.static:
9099
        remote_info = self.rpc.call_instance_info(instance.primary_node,
9100
                                                  instance.name,
9101
                                                  instance.hypervisor)
9102
        remote_info.Raise("Error checking node %s" % instance.primary_node)
9103
        remote_info = remote_info.payload
9104
        if remote_info and "state" in remote_info:
9105
          remote_state = "up"
9106
        else:
9107
          remote_state = "down"
9108
      else:
9109
        remote_state = None
9110
      if instance.admin_up:
9111
        config_state = "up"
9112
      else:
9113
        config_state = "down"
9114

    
9115
      disks = [self._ComputeDiskStatus(instance, None, device)
9116
               for device in instance.disks]
9117

    
9118
      idict = {
9119
        "name": instance.name,
9120
        "config_state": config_state,
9121
        "run_state": remote_state,
9122
        "pnode": instance.primary_node,
9123
        "snodes": instance.secondary_nodes,
9124
        "os": instance.os,
9125
        # this happens to be the same format used for hooks
9126
        "nics": _NICListToTuple(self, instance.nics),
9127
        "disk_template": instance.disk_template,
9128
        "disks": disks,
9129
        "hypervisor": instance.hypervisor,
9130
        "network_port": instance.network_port,
9131
        "hv_instance": instance.hvparams,
9132
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
9133
        "be_instance": instance.beparams,
9134
        "be_actual": cluster.FillBE(instance),
9135
        "os_instance": instance.osparams,
9136
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
9137
        "serial_no": instance.serial_no,
9138
        "mtime": instance.mtime,
9139
        "ctime": instance.ctime,
9140
        "uuid": instance.uuid,
9141
        }
9142

    
9143
      result[instance.name] = idict
9144

    
9145
    return result
9146

    
9147

    
9148
class LUInstanceSetParams(LogicalUnit):
9149
  """Modifies an instances's parameters.
9150

9151
  """
9152
  HPATH = "instance-modify"
9153
  HTYPE = constants.HTYPE_INSTANCE
9154
  REQ_BGL = False
9155

    
9156
  def CheckArguments(self):
9157
    if not (self.op.nics or self.op.disks or self.op.disk_template or
9158
            self.op.hvparams or self.op.beparams or self.op.os_name):
9159
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
9160

    
9161
    if self.op.hvparams:
9162
      _CheckGlobalHvParams(self.op.hvparams)
9163

    
9164
    # Disk validation
9165
    disk_addremove = 0
9166
    for disk_op, disk_dict in self.op.disks:
9167
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
9168
      if disk_op == constants.DDM_REMOVE:
9169
        disk_addremove += 1
9170
        continue
9171
      elif disk_op == constants.DDM_ADD:
9172
        disk_addremove += 1
9173
      else:
9174
        if not isinstance(disk_op, int):
9175
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
9176
        if not isinstance(disk_dict, dict):
9177
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
9178
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9179

    
9180
      if disk_op == constants.DDM_ADD:
9181
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
9182
        if mode not in constants.DISK_ACCESS_SET:
9183
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
9184
                                     errors.ECODE_INVAL)
9185
        size = disk_dict.get('size', None)
9186
        if size is None:
9187
          raise errors.OpPrereqError("Required disk parameter size missing",
9188
                                     errors.ECODE_INVAL)
9189
        try:
9190
          size = int(size)
9191
        except (TypeError, ValueError), err:
9192
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
9193
                                     str(err), errors.ECODE_INVAL)
9194
        disk_dict['size'] = size
9195
      else:
9196
        # modification of disk
9197
        if 'size' in disk_dict:
9198
          raise errors.OpPrereqError("Disk size change not possible, use"
9199
                                     " grow-disk", errors.ECODE_INVAL)
9200

    
9201
    if disk_addremove > 1:
9202
      raise errors.OpPrereqError("Only one disk add or remove operation"
9203
                                 " supported at a time", errors.ECODE_INVAL)
9204

    
9205
    if self.op.disks and self.op.disk_template is not None:
9206
      raise errors.OpPrereqError("Disk template conversion and other disk"
9207
                                 " changes not supported at the same time",
9208
                                 errors.ECODE_INVAL)
9209

    
9210
    if (self.op.disk_template and
9211
        self.op.disk_template in constants.DTS_NET_MIRROR and
9212
        self.op.remote_node is None):
9213
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
9214
                                 " one requires specifying a secondary node",
9215
                                 errors.ECODE_INVAL)
9216

    
9217
    # NIC validation
9218
    nic_addremove = 0
9219
    for nic_op, nic_dict in self.op.nics:
9220
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
9221
      if nic_op == constants.DDM_REMOVE:
9222
        nic_addremove += 1
9223
        continue
9224
      elif nic_op == constants.DDM_ADD:
9225
        nic_addremove += 1
9226
      else:
9227
        if not isinstance(nic_op, int):
9228
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
9229
        if not isinstance(nic_dict, dict):
9230
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
9231
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9232

    
9233
      # nic_dict should be a dict
9234
      nic_ip = nic_dict.get('ip', None)
9235
      if nic_ip is not None:
9236
        if nic_ip.lower() == constants.VALUE_NONE:
9237
          nic_dict['ip'] = None
9238
        else:
9239
          if not netutils.IPAddress.IsValid(nic_ip):
9240
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
9241
                                       errors.ECODE_INVAL)
9242

    
9243
      nic_bridge = nic_dict.get('bridge', None)
9244
      nic_link = nic_dict.get('link', None)
9245
      if nic_bridge and nic_link:
9246
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
9247
                                   " at the same time", errors.ECODE_INVAL)
9248
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
9249
        nic_dict['bridge'] = None
9250
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
9251
        nic_dict['link'] = None
9252

    
9253
      if nic_op == constants.DDM_ADD:
9254
        nic_mac = nic_dict.get('mac', None)
9255
        if nic_mac is None:
9256
          nic_dict['mac'] = constants.VALUE_AUTO
9257

    
9258
      if 'mac' in nic_dict:
9259
        nic_mac = nic_dict['mac']
9260
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9261
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
9262

    
9263
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
9264
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
9265
                                     " modifying an existing nic",
9266
                                     errors.ECODE_INVAL)
9267

    
9268
    if nic_addremove > 1:
9269
      raise errors.OpPrereqError("Only one NIC add or remove operation"
9270
                                 " supported at a time", errors.ECODE_INVAL)
9271

    
9272
  def ExpandNames(self):
9273
    self._ExpandAndLockInstance()
9274
    self.needed_locks[locking.LEVEL_NODE] = []
9275
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9276

    
9277
  def DeclareLocks(self, level):
9278
    if level == locking.LEVEL_NODE:
9279
      self._LockInstancesNodes()
9280
      if self.op.disk_template and self.op.remote_node:
9281
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9282
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
9283

    
9284
  def BuildHooksEnv(self):
9285
    """Build hooks env.
9286

9287
    This runs on the master, primary and secondaries.
9288

9289
    """
9290
    args = dict()
9291
    if constants.BE_MEMORY in self.be_new:
9292
      args['memory'] = self.be_new[constants.BE_MEMORY]
9293
    if constants.BE_VCPUS in self.be_new:
9294
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
9295
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
9296
    # information at all.
9297
    if self.op.nics:
9298
      args['nics'] = []
9299
      nic_override = dict(self.op.nics)
9300
      for idx, nic in enumerate(self.instance.nics):
9301
        if idx in nic_override:
9302
          this_nic_override = nic_override[idx]
9303
        else:
9304
          this_nic_override = {}
9305
        if 'ip' in this_nic_override:
9306
          ip = this_nic_override['ip']
9307
        else:
9308
          ip = nic.ip
9309
        if 'mac' in this_nic_override:
9310
          mac = this_nic_override['mac']
9311
        else:
9312
          mac = nic.mac
9313
        if idx in self.nic_pnew:
9314
          nicparams = self.nic_pnew[idx]
9315
        else:
9316
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
9317
        mode = nicparams[constants.NIC_MODE]
9318
        link = nicparams[constants.NIC_LINK]
9319
        args['nics'].append((ip, mac, mode, link))
9320
      if constants.DDM_ADD in nic_override:
9321
        ip = nic_override[constants.DDM_ADD].get('ip', None)
9322
        mac = nic_override[constants.DDM_ADD]['mac']
9323
        nicparams = self.nic_pnew[constants.DDM_ADD]
9324
        mode = nicparams[constants.NIC_MODE]
9325
        link = nicparams[constants.NIC_LINK]
9326
        args['nics'].append((ip, mac, mode, link))
9327
      elif constants.DDM_REMOVE in nic_override:
9328
        del args['nics'][-1]
9329

    
9330
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
9331
    if self.op.disk_template:
9332
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
9333
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
9334
    return env, nl, nl
9335

    
9336
  def CheckPrereq(self):
9337
    """Check prerequisites.
9338

9339
    This only checks the instance list against the existing names.
9340

9341
    """
9342
    # checking the new params on the primary/secondary nodes
9343

    
9344
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9345
    cluster = self.cluster = self.cfg.GetClusterInfo()
9346
    assert self.instance is not None, \
9347
      "Cannot retrieve locked instance %s" % self.op.instance_name
9348
    pnode = instance.primary_node
9349
    nodelist = list(instance.all_nodes)
9350

    
9351
    # OS change
9352
    if self.op.os_name and not self.op.force:
9353
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
9354
                      self.op.force_variant)
9355
      instance_os = self.op.os_name
9356
    else:
9357
      instance_os = instance.os
9358

    
9359
    if self.op.disk_template:
9360
      if instance.disk_template == self.op.disk_template:
9361
        raise errors.OpPrereqError("Instance already has disk template %s" %
9362
                                   instance.disk_template, errors.ECODE_INVAL)
9363

    
9364
      if (instance.disk_template,
9365
          self.op.disk_template) not in self._DISK_CONVERSIONS:
9366
        raise errors.OpPrereqError("Unsupported disk template conversion from"
9367
                                   " %s to %s" % (instance.disk_template,
9368
                                                  self.op.disk_template),
9369
                                   errors.ECODE_INVAL)
9370
      _CheckInstanceDown(self, instance, "cannot change disk template")
9371
      if self.op.disk_template in constants.DTS_NET_MIRROR:
9372
        if self.op.remote_node == pnode:
9373
          raise errors.OpPrereqError("Given new secondary node %s is the same"
9374
                                     " as the primary node of the instance" %
9375
                                     self.op.remote_node, errors.ECODE_STATE)
9376
        _CheckNodeOnline(self, self.op.remote_node)
9377
        _CheckNodeNotDrained(self, self.op.remote_node)
9378
        # FIXME: here we assume that the old instance type is DT_PLAIN
9379
        assert instance.disk_template == constants.DT_PLAIN
9380
        disks = [{"size": d.size, "vg": d.logical_id[0]}
9381
                 for d in instance.disks]
9382
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
9383
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
9384

    
9385
    # hvparams processing
9386
    if self.op.hvparams:
9387
      hv_type = instance.hypervisor
9388
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
9389
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
9390
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
9391

    
9392
      # local check
9393
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
9394
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
9395
      self.hv_new = hv_new # the new actual values
9396
      self.hv_inst = i_hvdict # the new dict (without defaults)
9397
    else:
9398
      self.hv_new = self.hv_inst = {}
9399

    
9400
    # beparams processing
9401
    if self.op.beparams:
9402
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
9403
                                   use_none=True)
9404
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
9405
      be_new = cluster.SimpleFillBE(i_bedict)
9406
      self.be_new = be_new # the new actual values
9407
      self.be_inst = i_bedict # the new dict (without defaults)
9408
    else:
9409
      self.be_new = self.be_inst = {}
9410

    
9411
    # osparams processing
9412
    if self.op.osparams:
9413
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
9414
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
9415
      self.os_inst = i_osdict # the new dict (without defaults)
9416
    else:
9417
      self.os_inst = {}
9418

    
9419
    self.warn = []
9420

    
9421
    if constants.BE_MEMORY in self.op.beparams and not self.op.force:
9422
      mem_check_list = [pnode]
9423
      if be_new[constants.BE_AUTO_BALANCE]:
9424
        # either we changed auto_balance to yes or it was from before
9425
        mem_check_list.extend(instance.secondary_nodes)
9426
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
9427
                                                  instance.hypervisor)
9428
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
9429
                                         instance.hypervisor)
9430
      pninfo = nodeinfo[pnode]
9431
      msg = pninfo.fail_msg
9432
      if msg:
9433
        # Assume the primary node is unreachable and go ahead
9434
        self.warn.append("Can't get info from primary node %s: %s" %
9435
                         (pnode,  msg))
9436
      elif not isinstance(pninfo.payload.get('memory_free', None), int):
9437
        self.warn.append("Node data from primary node %s doesn't contain"
9438
                         " free memory information" % pnode)
9439
      elif instance_info.fail_msg:
9440
        self.warn.append("Can't get instance runtime information: %s" %
9441
                        instance_info.fail_msg)
9442
      else:
9443
        if instance_info.payload:
9444
          current_mem = int(instance_info.payload['memory'])
9445
        else:
9446
          # Assume instance not running
9447
          # (there is a slight race condition here, but it's not very probable,
9448
          # and we have no other way to check)
9449
          current_mem = 0
9450
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
9451
                    pninfo.payload['memory_free'])
9452
        if miss_mem > 0:
9453
          raise errors.OpPrereqError("This change will prevent the instance"
9454
                                     " from starting, due to %d MB of memory"
9455
                                     " missing on its primary node" % miss_mem,
9456
                                     errors.ECODE_NORES)
9457

    
9458
      if be_new[constants.BE_AUTO_BALANCE]:
9459
        for node, nres in nodeinfo.items():
9460
          if node not in instance.secondary_nodes:
9461
            continue
9462
          msg = nres.fail_msg
9463
          if msg:
9464
            self.warn.append("Can't get info from secondary node %s: %s" %
9465
                             (node, msg))
9466
          elif not isinstance(nres.payload.get('memory_free', None), int):
9467
            self.warn.append("Secondary node %s didn't return free"
9468
                             " memory information" % node)
9469
          elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
9470
            self.warn.append("Not enough memory to failover instance to"
9471
                             " secondary node %s" % node)
9472

    
9473
    # NIC processing
9474
    self.nic_pnew = {}
9475
    self.nic_pinst = {}
9476
    for nic_op, nic_dict in self.op.nics:
9477
      if nic_op == constants.DDM_REMOVE:
9478
        if not instance.nics:
9479
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
9480
                                     errors.ECODE_INVAL)
9481
        continue
9482
      if nic_op != constants.DDM_ADD:
9483
        # an existing nic
9484
        if not instance.nics:
9485
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
9486
                                     " no NICs" % nic_op,
9487
                                     errors.ECODE_INVAL)
9488
        if nic_op < 0 or nic_op >= len(instance.nics):
9489
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
9490
                                     " are 0 to %d" %
9491
                                     (nic_op, len(instance.nics) - 1),
9492
                                     errors.ECODE_INVAL)
9493
        old_nic_params = instance.nics[nic_op].nicparams
9494
        old_nic_ip = instance.nics[nic_op].ip
9495
      else:
9496
        old_nic_params = {}
9497
        old_nic_ip = None
9498

    
9499
      update_params_dict = dict([(key, nic_dict[key])
9500
                                 for key in constants.NICS_PARAMETERS
9501
                                 if key in nic_dict])
9502

    
9503
      if 'bridge' in nic_dict:
9504
        update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
9505

    
9506
      new_nic_params = _GetUpdatedParams(old_nic_params,
9507
                                         update_params_dict)
9508
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
9509
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
9510
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
9511
      self.nic_pinst[nic_op] = new_nic_params
9512
      self.nic_pnew[nic_op] = new_filled_nic_params
9513
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
9514

    
9515
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
9516
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
9517
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
9518
        if msg:
9519
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
9520
          if self.op.force:
9521
            self.warn.append(msg)
9522
          else:
9523
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
9524
      if new_nic_mode == constants.NIC_MODE_ROUTED:
9525
        if 'ip' in nic_dict:
9526
          nic_ip = nic_dict['ip']
9527
        else:
9528
          nic_ip = old_nic_ip
9529
        if nic_ip is None:
9530
          raise errors.OpPrereqError('Cannot set the nic ip to None'
9531
                                     ' on a routed nic', errors.ECODE_INVAL)
9532
      if 'mac' in nic_dict:
9533
        nic_mac = nic_dict['mac']
9534
        if nic_mac is None:
9535
          raise errors.OpPrereqError('Cannot set the nic mac to None',
9536
                                     errors.ECODE_INVAL)
9537
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9538
          # otherwise generate the mac
9539
          nic_dict['mac'] = self.cfg.GenerateMAC(self.proc.GetECId())
9540
        else:
9541
          # or validate/reserve the current one
9542
          try:
9543
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
9544
          except errors.ReservationError:
9545
            raise errors.OpPrereqError("MAC address %s already in use"
9546
                                       " in cluster" % nic_mac,
9547
                                       errors.ECODE_NOTUNIQUE)
9548

    
9549
    # DISK processing
9550
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
9551
      raise errors.OpPrereqError("Disk operations not supported for"
9552
                                 " diskless instances",
9553
                                 errors.ECODE_INVAL)
9554
    for disk_op, _ in self.op.disks:
9555
      if disk_op == constants.DDM_REMOVE:
9556
        if len(instance.disks) == 1:
9557
          raise errors.OpPrereqError("Cannot remove the last disk of"
9558
                                     " an instance", errors.ECODE_INVAL)
9559
        _CheckInstanceDown(self, instance, "cannot remove disks")
9560

    
9561
      if (disk_op == constants.DDM_ADD and
9562
          len(instance.disks) >= constants.MAX_DISKS):
9563
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
9564
                                   " add more" % constants.MAX_DISKS,
9565
                                   errors.ECODE_STATE)
9566
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
9567
        # an existing disk
9568
        if disk_op < 0 or disk_op >= len(instance.disks):
9569
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
9570
                                     " are 0 to %d" %
9571
                                     (disk_op, len(instance.disks)),
9572
                                     errors.ECODE_INVAL)
9573

    
9574
    return
9575

    
9576
  def _ConvertPlainToDrbd(self, feedback_fn):
9577
    """Converts an instance from plain to drbd.
9578

9579
    """
9580
    feedback_fn("Converting template to drbd")
9581
    instance = self.instance
9582
    pnode = instance.primary_node
9583
    snode = self.op.remote_node
9584

    
9585
    # create a fake disk info for _GenerateDiskTemplate
9586
    disk_info = [{"size": d.size, "mode": d.mode} for d in instance.disks]
9587
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
9588
                                      instance.name, pnode, [snode],
9589
                                      disk_info, None, None, 0, feedback_fn)
9590
    info = _GetInstanceInfoText(instance)
9591
    feedback_fn("Creating aditional volumes...")
9592
    # first, create the missing data and meta devices
9593
    for disk in new_disks:
9594
      # unfortunately this is... not too nice
9595
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
9596
                            info, True)
9597
      for child in disk.children:
9598
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
9599
    # at this stage, all new LVs have been created, we can rename the
9600
    # old ones
9601
    feedback_fn("Renaming original volumes...")
9602
    rename_list = [(o, n.children[0].logical_id)
9603
                   for (o, n) in zip(instance.disks, new_disks)]
9604
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
9605
    result.Raise("Failed to rename original LVs")
9606

    
9607
    feedback_fn("Initializing DRBD devices...")
9608
    # all child devices are in place, we can now create the DRBD devices
9609
    for disk in new_disks:
9610
      for node in [pnode, snode]:
9611
        f_create = node == pnode
9612
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
9613

    
9614
    # at this point, the instance has been modified
9615
    instance.disk_template = constants.DT_DRBD8
9616
    instance.disks = new_disks
9617
    self.cfg.Update(instance, feedback_fn)
9618

    
9619
    # disks are created, waiting for sync
9620
    disk_abort = not _WaitForSync(self, instance)
9621
    if disk_abort:
9622
      raise errors.OpExecError("There are some degraded disks for"
9623
                               " this instance, please cleanup manually")
9624

    
9625
  def _ConvertDrbdToPlain(self, feedback_fn):
9626
    """Converts an instance from drbd to plain.
9627

9628
    """
9629
    instance = self.instance
9630
    assert len(instance.secondary_nodes) == 1
9631
    pnode = instance.primary_node
9632
    snode = instance.secondary_nodes[0]
9633
    feedback_fn("Converting template to plain")
9634

    
9635
    old_disks = instance.disks
9636
    new_disks = [d.children[0] for d in old_disks]
9637

    
9638
    # copy over size and mode
9639
    for parent, child in zip(old_disks, new_disks):
9640
      child.size = parent.size
9641
      child.mode = parent.mode
9642

    
9643
    # update instance structure
9644
    instance.disks = new_disks
9645
    instance.disk_template = constants.DT_PLAIN
9646
    self.cfg.Update(instance, feedback_fn)
9647

    
9648
    feedback_fn("Removing volumes on the secondary node...")
9649
    for disk in old_disks:
9650
      self.cfg.SetDiskID(disk, snode)
9651
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
9652
      if msg:
9653
        self.LogWarning("Could not remove block device %s on node %s,"
9654
                        " continuing anyway: %s", disk.iv_name, snode, msg)
9655

    
9656
    feedback_fn("Removing unneeded volumes on the primary node...")
9657
    for idx, disk in enumerate(old_disks):
9658
      meta = disk.children[1]
9659
      self.cfg.SetDiskID(meta, pnode)
9660
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
9661
      if msg:
9662
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
9663
                        " continuing anyway: %s", idx, pnode, msg)
9664

    
9665
  def Exec(self, feedback_fn):
9666
    """Modifies an instance.
9667

9668
    All parameters take effect only at the next restart of the instance.
9669

9670
    """
9671
    # Process here the warnings from CheckPrereq, as we don't have a
9672
    # feedback_fn there.
9673
    for warn in self.warn:
9674
      feedback_fn("WARNING: %s" % warn)
9675

    
9676
    result = []
9677
    instance = self.instance
9678
    # disk changes
9679
    for disk_op, disk_dict in self.op.disks:
9680
      if disk_op == constants.DDM_REMOVE:
9681
        # remove the last disk
9682
        device = instance.disks.pop()
9683
        device_idx = len(instance.disks)
9684
        for node, disk in device.ComputeNodeTree(instance.primary_node):
9685
          self.cfg.SetDiskID(disk, node)
9686
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
9687
          if msg:
9688
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
9689
                            " continuing anyway", device_idx, node, msg)
9690
        result.append(("disk/%d" % device_idx, "remove"))
9691
      elif disk_op == constants.DDM_ADD:
9692
        # add a new disk
9693
        if instance.disk_template in (constants.DT_FILE,
9694
                                        constants.DT_SHARED_FILE):
9695
          file_driver, file_path = instance.disks[0].logical_id
9696
          file_path = os.path.dirname(file_path)
9697
        else:
9698
          file_driver = file_path = None
9699
        disk_idx_base = len(instance.disks)
9700
        new_disk = _GenerateDiskTemplate(self,
9701
                                         instance.disk_template,
9702
                                         instance.name, instance.primary_node,
9703
                                         instance.secondary_nodes,
9704
                                         [disk_dict],
9705
                                         file_path,
9706
                                         file_driver,
9707
                                         disk_idx_base, feedback_fn)[0]
9708
        instance.disks.append(new_disk)
9709
        info = _GetInstanceInfoText(instance)
9710

    
9711
        logging.info("Creating volume %s for instance %s",
9712
                     new_disk.iv_name, instance.name)
9713
        # Note: this needs to be kept in sync with _CreateDisks
9714
        #HARDCODE
9715
        for node in instance.all_nodes:
9716
          f_create = node == instance.primary_node
9717
          try:
9718
            _CreateBlockDev(self, node, instance, new_disk,
9719
                            f_create, info, f_create)
9720
          except errors.OpExecError, err:
9721
            self.LogWarning("Failed to create volume %s (%s) on"
9722
                            " node %s: %s",
9723
                            new_disk.iv_name, new_disk, node, err)
9724
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
9725
                       (new_disk.size, new_disk.mode)))
9726
      else:
9727
        # change a given disk
9728
        instance.disks[disk_op].mode = disk_dict['mode']
9729
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
9730

    
9731
    if self.op.disk_template:
9732
      r_shut = _ShutdownInstanceDisks(self, instance)
9733
      if not r_shut:
9734
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
9735
                                 " proceed with disk template conversion")
9736
      mode = (instance.disk_template, self.op.disk_template)
9737
      try:
9738
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
9739
      except:
9740
        self.cfg.ReleaseDRBDMinors(instance.name)
9741
        raise
9742
      result.append(("disk_template", self.op.disk_template))
9743

    
9744
    # NIC changes
9745
    for nic_op, nic_dict in self.op.nics:
9746
      if nic_op == constants.DDM_REMOVE:
9747
        # remove the last nic
9748
        del instance.nics[-1]
9749
        result.append(("nic.%d" % len(instance.nics), "remove"))
9750
      elif nic_op == constants.DDM_ADD:
9751
        # mac and bridge should be set, by now
9752
        mac = nic_dict['mac']
9753
        ip = nic_dict.get('ip', None)
9754
        nicparams = self.nic_pinst[constants.DDM_ADD]
9755
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
9756
        instance.nics.append(new_nic)
9757
        result.append(("nic.%d" % (len(instance.nics) - 1),
9758
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
9759
                       (new_nic.mac, new_nic.ip,
9760
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
9761
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
9762
                       )))
9763
      else:
9764
        for key in 'mac', 'ip':
9765
          if key in nic_dict:
9766
            setattr(instance.nics[nic_op], key, nic_dict[key])
9767
        if nic_op in self.nic_pinst:
9768
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
9769
        for key, val in nic_dict.iteritems():
9770
          result.append(("nic.%s/%d" % (key, nic_op), val))
9771

    
9772
    # hvparams changes
9773
    if self.op.hvparams:
9774
      instance.hvparams = self.hv_inst
9775
      for key, val in self.op.hvparams.iteritems():
9776
        result.append(("hv/%s" % key, val))
9777

    
9778
    # beparams changes
9779
    if self.op.beparams:
9780
      instance.beparams = self.be_inst
9781
      for key, val in self.op.beparams.iteritems():
9782
        result.append(("be/%s" % key, val))
9783

    
9784
    # OS change
9785
    if self.op.os_name:
9786
      instance.os = self.op.os_name
9787

    
9788
    # osparams changes
9789
    if self.op.osparams:
9790
      instance.osparams = self.os_inst
9791
      for key, val in self.op.osparams.iteritems():
9792
        result.append(("os/%s" % key, val))
9793

    
9794
    self.cfg.Update(instance, feedback_fn)
9795

    
9796
    return result
9797

    
9798
  _DISK_CONVERSIONS = {
9799
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
9800
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
9801
    }
9802

    
9803

    
9804
class LUBackupQuery(NoHooksLU):
9805
  """Query the exports list
9806

9807
  """
9808
  REQ_BGL = False
9809

    
9810
  def ExpandNames(self):
9811
    self.needed_locks = {}
9812
    self.share_locks[locking.LEVEL_NODE] = 1
9813
    if not self.op.nodes:
9814
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9815
    else:
9816
      self.needed_locks[locking.LEVEL_NODE] = \
9817
        _GetWantedNodes(self, self.op.nodes)
9818

    
9819
  def Exec(self, feedback_fn):
9820
    """Compute the list of all the exported system images.
9821

9822
    @rtype: dict
9823
    @return: a dictionary with the structure node->(export-list)
9824
        where export-list is a list of the instances exported on
9825
        that node.
9826

9827
    """
9828
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
9829
    rpcresult = self.rpc.call_export_list(self.nodes)
9830
    result = {}
9831
    for node in rpcresult:
9832
      if rpcresult[node].fail_msg:
9833
        result[node] = False
9834
      else:
9835
        result[node] = rpcresult[node].payload
9836

    
9837
    return result
9838

    
9839

    
9840
class LUBackupPrepare(NoHooksLU):
9841
  """Prepares an instance for an export and returns useful information.
9842

9843
  """
9844
  REQ_BGL = False
9845

    
9846
  def ExpandNames(self):
9847
    self._ExpandAndLockInstance()
9848

    
9849
  def CheckPrereq(self):
9850
    """Check prerequisites.
9851

9852
    """
9853
    instance_name = self.op.instance_name
9854

    
9855
    self.instance = self.cfg.GetInstanceInfo(instance_name)
9856
    assert self.instance is not None, \
9857
          "Cannot retrieve locked instance %s" % self.op.instance_name
9858
    _CheckNodeOnline(self, self.instance.primary_node)
9859

    
9860
    self._cds = _GetClusterDomainSecret()
9861

    
9862
  def Exec(self, feedback_fn):
9863
    """Prepares an instance for an export.
9864

9865
    """
9866
    instance = self.instance
9867

    
9868
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
9869
      salt = utils.GenerateSecret(8)
9870

    
9871
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
9872
      result = self.rpc.call_x509_cert_create(instance.primary_node,
9873
                                              constants.RIE_CERT_VALIDITY)
9874
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
9875

    
9876
      (name, cert_pem) = result.payload
9877

    
9878
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
9879
                                             cert_pem)
9880

    
9881
      return {
9882
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
9883
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
9884
                          salt),
9885
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
9886
        }
9887

    
9888
    return None
9889

    
9890

    
9891
class LUBackupExport(LogicalUnit):
9892
  """Export an instance to an image in the cluster.
9893

9894
  """
9895
  HPATH = "instance-export"
9896
  HTYPE = constants.HTYPE_INSTANCE
9897
  REQ_BGL = False
9898

    
9899
  def CheckArguments(self):
9900
    """Check the arguments.
9901

9902
    """
9903
    self.x509_key_name = self.op.x509_key_name
9904
    self.dest_x509_ca_pem = self.op.destination_x509_ca
9905

    
9906
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
9907
      if not self.x509_key_name:
9908
        raise errors.OpPrereqError("Missing X509 key name for encryption",
9909
                                   errors.ECODE_INVAL)
9910

    
9911
      if not self.dest_x509_ca_pem:
9912
        raise errors.OpPrereqError("Missing destination X509 CA",
9913
                                   errors.ECODE_INVAL)
9914

    
9915
  def ExpandNames(self):
9916
    self._ExpandAndLockInstance()
9917

    
9918
    # Lock all nodes for local exports
9919
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9920
      # FIXME: lock only instance primary and destination node
9921
      #
9922
      # Sad but true, for now we have do lock all nodes, as we don't know where
9923
      # the previous export might be, and in this LU we search for it and
9924
      # remove it from its current node. In the future we could fix this by:
9925
      #  - making a tasklet to search (share-lock all), then create the
9926
      #    new one, then one to remove, after
9927
      #  - removing the removal operation altogether
9928
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9929

    
9930
  def DeclareLocks(self, level):
9931
    """Last minute lock declaration."""
9932
    # All nodes are locked anyway, so nothing to do here.
9933

    
9934
  def BuildHooksEnv(self):
9935
    """Build hooks env.
9936

9937
    This will run on the master, primary node and target node.
9938

9939
    """
9940
    env = {
9941
      "EXPORT_MODE": self.op.mode,
9942
      "EXPORT_NODE": self.op.target_node,
9943
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
9944
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
9945
      # TODO: Generic function for boolean env variables
9946
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
9947
      }
9948

    
9949
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
9950

    
9951
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
9952

    
9953
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9954
      nl.append(self.op.target_node)
9955

    
9956
    return env, nl, nl
9957

    
9958
  def CheckPrereq(self):
9959
    """Check prerequisites.
9960

9961
    This checks that the instance and node names are valid.
9962

9963
    """
9964
    instance_name = self.op.instance_name
9965

    
9966
    self.instance = self.cfg.GetInstanceInfo(instance_name)
9967
    assert self.instance is not None, \
9968
          "Cannot retrieve locked instance %s" % self.op.instance_name
9969
    _CheckNodeOnline(self, self.instance.primary_node)
9970

    
9971
    if (self.op.remove_instance and self.instance.admin_up and
9972
        not self.op.shutdown):
9973
      raise errors.OpPrereqError("Can not remove instance without shutting it"
9974
                                 " down before")
9975

    
9976
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9977
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
9978
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
9979
      assert self.dst_node is not None
9980

    
9981
      _CheckNodeOnline(self, self.dst_node.name)
9982
      _CheckNodeNotDrained(self, self.dst_node.name)
9983

    
9984
      self._cds = None
9985
      self.dest_disk_info = None
9986
      self.dest_x509_ca = None
9987

    
9988
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
9989
      self.dst_node = None
9990

    
9991
      if len(self.op.target_node) != len(self.instance.disks):
9992
        raise errors.OpPrereqError(("Received destination information for %s"
9993
                                    " disks, but instance %s has %s disks") %
9994
                                   (len(self.op.target_node), instance_name,
9995
                                    len(self.instance.disks)),
9996
                                   errors.ECODE_INVAL)
9997

    
9998
      cds = _GetClusterDomainSecret()
9999

    
10000
      # Check X509 key name
10001
      try:
10002
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
10003
      except (TypeError, ValueError), err:
10004
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
10005

    
10006
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
10007
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
10008
                                   errors.ECODE_INVAL)
10009

    
10010
      # Load and verify CA
10011
      try:
10012
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
10013
      except OpenSSL.crypto.Error, err:
10014
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
10015
                                   (err, ), errors.ECODE_INVAL)
10016

    
10017
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
10018
      if errcode is not None:
10019
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
10020
                                   (msg, ), errors.ECODE_INVAL)
10021

    
10022
      self.dest_x509_ca = cert
10023

    
10024
      # Verify target information
10025
      disk_info = []
10026
      for idx, disk_data in enumerate(self.op.target_node):
10027
        try:
10028
          (host, port, magic) = \
10029
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
10030
        except errors.GenericError, err:
10031
          raise errors.OpPrereqError("Target info for disk %s: %s" %
10032
                                     (idx, err), errors.ECODE_INVAL)
10033

    
10034
        disk_info.append((host, port, magic))
10035

    
10036
      assert len(disk_info) == len(self.op.target_node)
10037
      self.dest_disk_info = disk_info
10038

    
10039
    else:
10040
      raise errors.ProgrammerError("Unhandled export mode %r" %
10041
                                   self.op.mode)
10042

    
10043
    # instance disk type verification
10044
    # TODO: Implement export support for file-based disks
10045
    for disk in self.instance.disks:
10046
      if disk.dev_type == constants.LD_FILE:
10047
        raise errors.OpPrereqError("Export not supported for instances with"
10048
                                   " file-based disks", errors.ECODE_INVAL)
10049

    
10050
  def _CleanupExports(self, feedback_fn):
10051
    """Removes exports of current instance from all other nodes.
10052

10053
    If an instance in a cluster with nodes A..D was exported to node C, its
10054
    exports will be removed from the nodes A, B and D.
10055

10056
    """
10057
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
10058

    
10059
    nodelist = self.cfg.GetNodeList()
10060
    nodelist.remove(self.dst_node.name)
10061

    
10062
    # on one-node clusters nodelist will be empty after the removal
10063
    # if we proceed the backup would be removed because OpBackupQuery
10064
    # substitutes an empty list with the full cluster node list.
10065
    iname = self.instance.name
10066
    if nodelist:
10067
      feedback_fn("Removing old exports for instance %s" % iname)
10068
      exportlist = self.rpc.call_export_list(nodelist)
10069
      for node in exportlist:
10070
        if exportlist[node].fail_msg:
10071
          continue
10072
        if iname in exportlist[node].payload:
10073
          msg = self.rpc.call_export_remove(node, iname).fail_msg
10074
          if msg:
10075
            self.LogWarning("Could not remove older export for instance %s"
10076
                            " on node %s: %s", iname, node, msg)
10077

    
10078
  def Exec(self, feedback_fn):
10079
    """Export an instance to an image in the cluster.
10080

10081
    """
10082
    assert self.op.mode in constants.EXPORT_MODES
10083

    
10084
    instance = self.instance
10085
    src_node = instance.primary_node
10086

    
10087
    if self.op.shutdown:
10088
      # shutdown the instance, but not the disks
10089
      feedback_fn("Shutting down instance %s" % instance.name)
10090
      result = self.rpc.call_instance_shutdown(src_node, instance,
10091
                                               self.op.shutdown_timeout)
10092
      # TODO: Maybe ignore failures if ignore_remove_failures is set
10093
      result.Raise("Could not shutdown instance %s on"
10094
                   " node %s" % (instance.name, src_node))
10095

    
10096
    # set the disks ID correctly since call_instance_start needs the
10097
    # correct drbd minor to create the symlinks
10098
    for disk in instance.disks:
10099
      self.cfg.SetDiskID(disk, src_node)
10100

    
10101
    activate_disks = (not instance.admin_up)
10102

    
10103
    if activate_disks:
10104
      # Activate the instance disks if we'exporting a stopped instance
10105
      feedback_fn("Activating disks for %s" % instance.name)
10106
      _StartInstanceDisks(self, instance, None)
10107

    
10108
    try:
10109
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
10110
                                                     instance)
10111

    
10112
      helper.CreateSnapshots()
10113
      try:
10114
        if (self.op.shutdown and instance.admin_up and
10115
            not self.op.remove_instance):
10116
          assert not activate_disks
10117
          feedback_fn("Starting instance %s" % instance.name)
10118
          result = self.rpc.call_instance_start(src_node, instance, None, None)
10119
          msg = result.fail_msg
10120
          if msg:
10121
            feedback_fn("Failed to start instance: %s" % msg)
10122
            _ShutdownInstanceDisks(self, instance)
10123
            raise errors.OpExecError("Could not start instance: %s" % msg)
10124

    
10125
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
10126
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
10127
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10128
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
10129
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
10130

    
10131
          (key_name, _, _) = self.x509_key_name
10132

    
10133
          dest_ca_pem = \
10134
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
10135
                                            self.dest_x509_ca)
10136

    
10137
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
10138
                                                     key_name, dest_ca_pem,
10139
                                                     timeouts)
10140
      finally:
10141
        helper.Cleanup()
10142

    
10143
      # Check for backwards compatibility
10144
      assert len(dresults) == len(instance.disks)
10145
      assert compat.all(isinstance(i, bool) for i in dresults), \
10146
             "Not all results are boolean: %r" % dresults
10147

    
10148
    finally:
10149
      if activate_disks:
10150
        feedback_fn("Deactivating disks for %s" % instance.name)
10151
        _ShutdownInstanceDisks(self, instance)
10152

    
10153
    if not (compat.all(dresults) and fin_resu):
10154
      failures = []
10155
      if not fin_resu:
10156
        failures.append("export finalization")
10157
      if not compat.all(dresults):
10158
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
10159
                               if not dsk)
10160
        failures.append("disk export: disk(s) %s" % fdsk)
10161

    
10162
      raise errors.OpExecError("Export failed, errors in %s" %
10163
                               utils.CommaJoin(failures))
10164

    
10165
    # At this point, the export was successful, we can cleanup/finish
10166

    
10167
    # Remove instance if requested
10168
    if self.op.remove_instance:
10169
      feedback_fn("Removing instance %s" % instance.name)
10170
      _RemoveInstance(self, feedback_fn, instance,
10171
                      self.op.ignore_remove_failures)
10172

    
10173
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10174
      self._CleanupExports(feedback_fn)
10175

    
10176
    return fin_resu, dresults
10177

    
10178

    
10179
class LUBackupRemove(NoHooksLU):
10180
  """Remove exports related to the named instance.
10181

10182
  """
10183
  REQ_BGL = False
10184

    
10185
  def ExpandNames(self):
10186
    self.needed_locks = {}
10187
    # We need all nodes to be locked in order for RemoveExport to work, but we
10188
    # don't need to lock the instance itself, as nothing will happen to it (and
10189
    # we can remove exports also for a removed instance)
10190
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10191

    
10192
  def Exec(self, feedback_fn):
10193
    """Remove any export.
10194

10195
    """
10196
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
10197
    # If the instance was not found we'll try with the name that was passed in.
10198
    # This will only work if it was an FQDN, though.
10199
    fqdn_warn = False
10200
    if not instance_name:
10201
      fqdn_warn = True
10202
      instance_name = self.op.instance_name
10203

    
10204
    locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
10205
    exportlist = self.rpc.call_export_list(locked_nodes)
10206
    found = False
10207
    for node in exportlist:
10208
      msg = exportlist[node].fail_msg
10209
      if msg:
10210
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
10211
        continue
10212
      if instance_name in exportlist[node].payload:
10213
        found = True
10214
        result = self.rpc.call_export_remove(node, instance_name)
10215
        msg = result.fail_msg
10216
        if msg:
10217
          logging.error("Could not remove export for instance %s"
10218
                        " on node %s: %s", instance_name, node, msg)
10219

    
10220
    if fqdn_warn and not found:
10221
      feedback_fn("Export not found. If trying to remove an export belonging"
10222
                  " to a deleted instance please use its Fully Qualified"
10223
                  " Domain Name.")
10224

    
10225

    
10226
class LUGroupAdd(LogicalUnit):
10227
  """Logical unit for creating node groups.
10228

10229
  """
10230
  HPATH = "group-add"
10231
  HTYPE = constants.HTYPE_GROUP
10232
  REQ_BGL = False
10233

    
10234
  def ExpandNames(self):
10235
    # We need the new group's UUID here so that we can create and acquire the
10236
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
10237
    # that it should not check whether the UUID exists in the configuration.
10238
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
10239
    self.needed_locks = {}
10240
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10241

    
10242
  def CheckPrereq(self):
10243
    """Check prerequisites.
10244

10245
    This checks that the given group name is not an existing node group
10246
    already.
10247

10248
    """
10249
    try:
10250
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10251
    except errors.OpPrereqError:
10252
      pass
10253
    else:
10254
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
10255
                                 " node group (UUID: %s)" %
10256
                                 (self.op.group_name, existing_uuid),
10257
                                 errors.ECODE_EXISTS)
10258

    
10259
    if self.op.ndparams:
10260
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10261

    
10262
  def BuildHooksEnv(self):
10263
    """Build hooks env.
10264

10265
    """
10266
    env = {
10267
      "GROUP_NAME": self.op.group_name,
10268
      }
10269
    mn = self.cfg.GetMasterNode()
10270
    return env, [mn], [mn]
10271

    
10272
  def Exec(self, feedback_fn):
10273
    """Add the node group to the cluster.
10274

10275
    """
10276
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
10277
                                  uuid=self.group_uuid,
10278
                                  alloc_policy=self.op.alloc_policy,
10279
                                  ndparams=self.op.ndparams)
10280

    
10281
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
10282
    del self.remove_locks[locking.LEVEL_NODEGROUP]
10283

    
10284

    
10285
class LUGroupAssignNodes(NoHooksLU):
10286
  """Logical unit for assigning nodes to groups.
10287

10288
  """
10289
  REQ_BGL = False
10290

    
10291
  def ExpandNames(self):
10292
    # These raise errors.OpPrereqError on their own:
10293
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10294
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
10295

    
10296
    # We want to lock all the affected nodes and groups. We have readily
10297
    # available the list of nodes, and the *destination* group. To gather the
10298
    # list of "source" groups, we need to fetch node information.
10299
    self.node_data = self.cfg.GetAllNodesInfo()
10300
    affected_groups = set(self.node_data[node].group for node in self.op.nodes)
10301
    affected_groups.add(self.group_uuid)
10302

    
10303
    self.needed_locks = {
10304
      locking.LEVEL_NODEGROUP: list(affected_groups),
10305
      locking.LEVEL_NODE: self.op.nodes,
10306
      }
10307

    
10308
  def CheckPrereq(self):
10309
    """Check prerequisites.
10310

10311
    """
10312
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10313
    instance_data = self.cfg.GetAllInstancesInfo()
10314

    
10315
    if self.group is None:
10316
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10317
                               (self.op.group_name, self.group_uuid))
10318

    
10319
    (new_splits, previous_splits) = \
10320
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
10321
                                             for node in self.op.nodes],
10322
                                            self.node_data, instance_data)
10323

    
10324
    if new_splits:
10325
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
10326

    
10327
      if not self.op.force:
10328
        raise errors.OpExecError("The following instances get split by this"
10329
                                 " change and --force was not given: %s" %
10330
                                 fmt_new_splits)
10331
      else:
10332
        self.LogWarning("This operation will split the following instances: %s",
10333
                        fmt_new_splits)
10334

    
10335
        if previous_splits:
10336
          self.LogWarning("In addition, these already-split instances continue"
10337
                          " to be spit across groups: %s",
10338
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
10339

    
10340
  def Exec(self, feedback_fn):
10341
    """Assign nodes to a new group.
10342

10343
    """
10344
    for node in self.op.nodes:
10345
      self.node_data[node].group = self.group_uuid
10346

    
10347
    self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
10348

    
10349
  @staticmethod
10350
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
10351
    """Check for split instances after a node assignment.
10352

10353
    This method considers a series of node assignments as an atomic operation,
10354
    and returns information about split instances after applying the set of
10355
    changes.
10356

10357
    In particular, it returns information about newly split instances, and
10358
    instances that were already split, and remain so after the change.
10359

10360
    Only instances whose disk template is listed in constants.DTS_NET_MIRROR are
10361
    considered.
10362

10363
    @type changes: list of (node_name, new_group_uuid) pairs.
10364
    @param changes: list of node assignments to consider.
10365
    @param node_data: a dict with data for all nodes
10366
    @param instance_data: a dict with all instances to consider
10367
    @rtype: a two-tuple
10368
    @return: a list of instances that were previously okay and result split as a
10369
      consequence of this change, and a list of instances that were previously
10370
      split and this change does not fix.
10371

10372
    """
10373
    changed_nodes = dict((node, group) for node, group in changes
10374
                         if node_data[node].group != group)
10375

    
10376
    all_split_instances = set()
10377
    previously_split_instances = set()
10378

    
10379
    def InstanceNodes(instance):
10380
      return [instance.primary_node] + list(instance.secondary_nodes)
10381

    
10382
    for inst in instance_data.values():
10383
      if inst.disk_template not in constants.DTS_NET_MIRROR:
10384
        continue
10385

    
10386
      instance_nodes = InstanceNodes(inst)
10387

    
10388
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
10389
        previously_split_instances.add(inst.name)
10390

    
10391
      if len(set(changed_nodes.get(node, node_data[node].group)
10392
                 for node in instance_nodes)) > 1:
10393
        all_split_instances.add(inst.name)
10394

    
10395
    return (list(all_split_instances - previously_split_instances),
10396
            list(previously_split_instances & all_split_instances))
10397

    
10398

    
10399
class _GroupQuery(_QueryBase):
10400

    
10401
  FIELDS = query.GROUP_FIELDS
10402

    
10403
  def ExpandNames(self, lu):
10404
    lu.needed_locks = {}
10405

    
10406
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
10407
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
10408

    
10409
    if not self.names:
10410
      self.wanted = [name_to_uuid[name]
10411
                     for name in utils.NiceSort(name_to_uuid.keys())]
10412
    else:
10413
      # Accept names to be either names or UUIDs.
10414
      missing = []
10415
      self.wanted = []
10416
      all_uuid = frozenset(self._all_groups.keys())
10417

    
10418
      for name in self.names:
10419
        if name in all_uuid:
10420
          self.wanted.append(name)
10421
        elif name in name_to_uuid:
10422
          self.wanted.append(name_to_uuid[name])
10423
        else:
10424
          missing.append(name)
10425

    
10426
      if missing:
10427
        raise errors.OpPrereqError("Some groups do not exist: %s" % missing,
10428
                                   errors.ECODE_NOENT)
10429

    
10430
  def DeclareLocks(self, lu, level):
10431
    pass
10432

    
10433
  def _GetQueryData(self, lu):
10434
    """Computes the list of node groups and their attributes.
10435

10436
    """
10437
    do_nodes = query.GQ_NODE in self.requested_data
10438
    do_instances = query.GQ_INST in self.requested_data
10439

    
10440
    group_to_nodes = None
10441
    group_to_instances = None
10442

    
10443
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
10444
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
10445
    # latter GetAllInstancesInfo() is not enough, for we have to go through
10446
    # instance->node. Hence, we will need to process nodes even if we only need
10447
    # instance information.
10448
    if do_nodes or do_instances:
10449
      all_nodes = lu.cfg.GetAllNodesInfo()
10450
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
10451
      node_to_group = {}
10452

    
10453
      for node in all_nodes.values():
10454
        if node.group in group_to_nodes:
10455
          group_to_nodes[node.group].append(node.name)
10456
          node_to_group[node.name] = node.group
10457

    
10458
      if do_instances:
10459
        all_instances = lu.cfg.GetAllInstancesInfo()
10460
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
10461

    
10462
        for instance in all_instances.values():
10463
          node = instance.primary_node
10464
          if node in node_to_group:
10465
            group_to_instances[node_to_group[node]].append(instance.name)
10466

    
10467
        if not do_nodes:
10468
          # Do not pass on node information if it was not requested.
10469
          group_to_nodes = None
10470

    
10471
    return query.GroupQueryData([self._all_groups[uuid]
10472
                                 for uuid in self.wanted],
10473
                                group_to_nodes, group_to_instances)
10474

    
10475

    
10476
class LUGroupQuery(NoHooksLU):
10477
  """Logical unit for querying node groups.
10478

10479
  """
10480
  REQ_BGL = False
10481

    
10482
  def CheckArguments(self):
10483
    self.gq = _GroupQuery(self.op.names, self.op.output_fields, False)
10484

    
10485
  def ExpandNames(self):
10486
    self.gq.ExpandNames(self)
10487

    
10488
  def Exec(self, feedback_fn):
10489
    return self.gq.OldStyleQuery(self)
10490

    
10491

    
10492
class LUGroupSetParams(LogicalUnit):
10493
  """Modifies the parameters of a node group.
10494

10495
  """
10496
  HPATH = "group-modify"
10497
  HTYPE = constants.HTYPE_GROUP
10498
  REQ_BGL = False
10499

    
10500
  def CheckArguments(self):
10501
    all_changes = [
10502
      self.op.ndparams,
10503
      self.op.alloc_policy,
10504
      ]
10505

    
10506
    if all_changes.count(None) == len(all_changes):
10507
      raise errors.OpPrereqError("Please pass at least one modification",
10508
                                 errors.ECODE_INVAL)
10509

    
10510
  def ExpandNames(self):
10511
    # This raises errors.OpPrereqError on its own:
10512
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10513

    
10514
    self.needed_locks = {
10515
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10516
      }
10517

    
10518
  def CheckPrereq(self):
10519
    """Check prerequisites.
10520

10521
    """
10522
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10523

    
10524
    if self.group is None:
10525
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10526
                               (self.op.group_name, self.group_uuid))
10527

    
10528
    if self.op.ndparams:
10529
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
10530
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10531
      self.new_ndparams = new_ndparams
10532

    
10533
  def BuildHooksEnv(self):
10534
    """Build hooks env.
10535

10536
    """
10537
    env = {
10538
      "GROUP_NAME": self.op.group_name,
10539
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
10540
      }
10541
    mn = self.cfg.GetMasterNode()
10542
    return env, [mn], [mn]
10543

    
10544
  def Exec(self, feedback_fn):
10545
    """Modifies the node group.
10546

10547
    """
10548
    result = []
10549

    
10550
    if self.op.ndparams:
10551
      self.group.ndparams = self.new_ndparams
10552
      result.append(("ndparams", str(self.group.ndparams)))
10553

    
10554
    if self.op.alloc_policy:
10555
      self.group.alloc_policy = self.op.alloc_policy
10556

    
10557
    self.cfg.Update(self.group, feedback_fn)
10558
    return result
10559

    
10560

    
10561

    
10562
class LUGroupRemove(LogicalUnit):
10563
  HPATH = "group-remove"
10564
  HTYPE = constants.HTYPE_GROUP
10565
  REQ_BGL = False
10566

    
10567
  def ExpandNames(self):
10568
    # This will raises errors.OpPrereqError on its own:
10569
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10570
    self.needed_locks = {
10571
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10572
      }
10573

    
10574
  def CheckPrereq(self):
10575
    """Check prerequisites.
10576

10577
    This checks that the given group name exists as a node group, that is
10578
    empty (i.e., contains no nodes), and that is not the last group of the
10579
    cluster.
10580

10581
    """
10582
    # Verify that the group is empty.
10583
    group_nodes = [node.name
10584
                   for node in self.cfg.GetAllNodesInfo().values()
10585
                   if node.group == self.group_uuid]
10586

    
10587
    if group_nodes:
10588
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
10589
                                 " nodes: %s" %
10590
                                 (self.op.group_name,
10591
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
10592
                                 errors.ECODE_STATE)
10593

    
10594
    # Verify the cluster would not be left group-less.
10595
    if len(self.cfg.GetNodeGroupList()) == 1:
10596
      raise errors.OpPrereqError("Group '%s' is the only group,"
10597
                                 " cannot be removed" %
10598
                                 self.op.group_name,
10599
                                 errors.ECODE_STATE)
10600

    
10601
  def BuildHooksEnv(self):
10602
    """Build hooks env.
10603

10604
    """
10605
    env = {
10606
      "GROUP_NAME": self.op.group_name,
10607
      }
10608
    mn = self.cfg.GetMasterNode()
10609
    return env, [mn], [mn]
10610

    
10611
  def Exec(self, feedback_fn):
10612
    """Remove the node group.
10613

10614
    """
10615
    try:
10616
      self.cfg.RemoveNodeGroup(self.group_uuid)
10617
    except errors.ConfigurationError:
10618
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
10619
                               (self.op.group_name, self.group_uuid))
10620

    
10621
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10622

    
10623

    
10624
class LUGroupRename(LogicalUnit):
10625
  HPATH = "group-rename"
10626
  HTYPE = constants.HTYPE_GROUP
10627
  REQ_BGL = False
10628

    
10629
  def ExpandNames(self):
10630
    # This raises errors.OpPrereqError on its own:
10631
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.old_name)
10632

    
10633
    self.needed_locks = {
10634
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10635
      }
10636

    
10637
  def CheckPrereq(self):
10638
    """Check prerequisites.
10639

10640
    This checks that the given old_name exists as a node group, and that
10641
    new_name doesn't.
10642

10643
    """
10644
    try:
10645
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
10646
    except errors.OpPrereqError:
10647
      pass
10648
    else:
10649
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
10650
                                 " node group (UUID: %s)" %
10651
                                 (self.op.new_name, new_name_uuid),
10652
                                 errors.ECODE_EXISTS)
10653

    
10654
  def BuildHooksEnv(self):
10655
    """Build hooks env.
10656

10657
    """
10658
    env = {
10659
      "OLD_NAME": self.op.old_name,
10660
      "NEW_NAME": self.op.new_name,
10661
      }
10662

    
10663
    mn = self.cfg.GetMasterNode()
10664
    all_nodes = self.cfg.GetAllNodesInfo()
10665
    run_nodes = [mn]
10666
    all_nodes.pop(mn, None)
10667

    
10668
    for node in all_nodes.values():
10669
      if node.group == self.group_uuid:
10670
        run_nodes.append(node.name)
10671

    
10672
    return env, run_nodes, run_nodes
10673

    
10674
  def Exec(self, feedback_fn):
10675
    """Rename the node group.
10676

10677
    """
10678
    group = self.cfg.GetNodeGroup(self.group_uuid)
10679

    
10680
    if group is None:
10681
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10682
                               (self.op.old_name, self.group_uuid))
10683

    
10684
    group.name = self.op.new_name
10685
    self.cfg.Update(group, feedback_fn)
10686

    
10687
    return self.op.new_name
10688

    
10689

    
10690
class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
10691
  """Generic tags LU.
10692

10693
  This is an abstract class which is the parent of all the other tags LUs.
10694

10695
  """
10696

    
10697
  def ExpandNames(self):
10698
    self.needed_locks = {}
10699
    if self.op.kind == constants.TAG_NODE:
10700
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
10701
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
10702
    elif self.op.kind == constants.TAG_INSTANCE:
10703
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
10704
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
10705

    
10706
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
10707
    # not possible to acquire the BGL based on opcode parameters)
10708

    
10709
  def CheckPrereq(self):
10710
    """Check prerequisites.
10711

10712
    """
10713
    if self.op.kind == constants.TAG_CLUSTER:
10714
      self.target = self.cfg.GetClusterInfo()
10715
    elif self.op.kind == constants.TAG_NODE:
10716
      self.target = self.cfg.GetNodeInfo(self.op.name)
10717
    elif self.op.kind == constants.TAG_INSTANCE:
10718
      self.target = self.cfg.GetInstanceInfo(self.op.name)
10719
    else:
10720
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
10721
                                 str(self.op.kind), errors.ECODE_INVAL)
10722

    
10723

    
10724
class LUTagsGet(TagsLU):
10725
  """Returns the tags of a given object.
10726

10727
  """
10728
  REQ_BGL = False
10729

    
10730
  def ExpandNames(self):
10731
    TagsLU.ExpandNames(self)
10732

    
10733
    # Share locks as this is only a read operation
10734
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
10735

    
10736
  def Exec(self, feedback_fn):
10737
    """Returns the tag list.
10738

10739
    """
10740
    return list(self.target.GetTags())
10741

    
10742

    
10743
class LUTagsSearch(NoHooksLU):
10744
  """Searches the tags for a given pattern.
10745

10746
  """
10747
  REQ_BGL = False
10748

    
10749
  def ExpandNames(self):
10750
    self.needed_locks = {}
10751

    
10752
  def CheckPrereq(self):
10753
    """Check prerequisites.
10754

10755
    This checks the pattern passed for validity by compiling it.
10756

10757
    """
10758
    try:
10759
      self.re = re.compile(self.op.pattern)
10760
    except re.error, err:
10761
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
10762
                                 (self.op.pattern, err), errors.ECODE_INVAL)
10763

    
10764
  def Exec(self, feedback_fn):
10765
    """Returns the tag list.
10766

10767
    """
10768
    cfg = self.cfg
10769
    tgts = [("/cluster", cfg.GetClusterInfo())]
10770
    ilist = cfg.GetAllInstancesInfo().values()
10771
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
10772
    nlist = cfg.GetAllNodesInfo().values()
10773
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
10774
    results = []
10775
    for path, target in tgts:
10776
      for tag in target.GetTags():
10777
        if self.re.search(tag):
10778
          results.append((path, tag))
10779
    return results
10780

    
10781

    
10782
class LUTagsSet(TagsLU):
10783
  """Sets a tag on a given object.
10784

10785
  """
10786
  REQ_BGL = False
10787

    
10788
  def CheckPrereq(self):
10789
    """Check prerequisites.
10790

10791
    This checks the type and length of the tag name and value.
10792

10793
    """
10794
    TagsLU.CheckPrereq(self)
10795
    for tag in self.op.tags:
10796
      objects.TaggableObject.ValidateTag(tag)
10797

    
10798
  def Exec(self, feedback_fn):
10799
    """Sets the tag.
10800

10801
    """
10802
    try:
10803
      for tag in self.op.tags:
10804
        self.target.AddTag(tag)
10805
    except errors.TagError, err:
10806
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
10807
    self.cfg.Update(self.target, feedback_fn)
10808

    
10809

    
10810
class LUTagsDel(TagsLU):
10811
  """Delete a list of tags from a given object.
10812

10813
  """
10814
  REQ_BGL = False
10815

    
10816
  def CheckPrereq(self):
10817
    """Check prerequisites.
10818

10819
    This checks that we have the given tag.
10820

10821
    """
10822
    TagsLU.CheckPrereq(self)
10823
    for tag in self.op.tags:
10824
      objects.TaggableObject.ValidateTag(tag)
10825
    del_tags = frozenset(self.op.tags)
10826
    cur_tags = self.target.GetTags()
10827

    
10828
    diff_tags = del_tags - cur_tags
10829
    if diff_tags:
10830
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
10831
      raise errors.OpPrereqError("Tag(s) %s not found" %
10832
                                 (utils.CommaJoin(diff_names), ),
10833
                                 errors.ECODE_NOENT)
10834

    
10835
  def Exec(self, feedback_fn):
10836
    """Remove the tag from the object.
10837

10838
    """
10839
    for tag in self.op.tags:
10840
      self.target.RemoveTag(tag)
10841
    self.cfg.Update(self.target, feedback_fn)
10842

    
10843

    
10844
class LUTestDelay(NoHooksLU):
10845
  """Sleep for a specified amount of time.
10846

10847
  This LU sleeps on the master and/or nodes for a specified amount of
10848
  time.
10849

10850
  """
10851
  REQ_BGL = False
10852

    
10853
  def ExpandNames(self):
10854
    """Expand names and set required locks.
10855

10856
    This expands the node list, if any.
10857

10858
    """
10859
    self.needed_locks = {}
10860
    if self.op.on_nodes:
10861
      # _GetWantedNodes can be used here, but is not always appropriate to use
10862
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
10863
      # more information.
10864
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
10865
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
10866

    
10867
  def _TestDelay(self):
10868
    """Do the actual sleep.
10869

10870
    """
10871
    if self.op.on_master:
10872
      if not utils.TestDelay(self.op.duration):
10873
        raise errors.OpExecError("Error during master delay test")
10874
    if self.op.on_nodes:
10875
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
10876
      for node, node_result in result.items():
10877
        node_result.Raise("Failure during rpc call to node %s" % node)
10878

    
10879
  def Exec(self, feedback_fn):
10880
    """Execute the test delay opcode, with the wanted repetitions.
10881

10882
    """
10883
    if self.op.repeat == 0:
10884
      self._TestDelay()
10885
    else:
10886
      top_value = self.op.repeat - 1
10887
      for i in range(self.op.repeat):
10888
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
10889
        self._TestDelay()
10890

    
10891

    
10892
class LUTestJqueue(NoHooksLU):
10893
  """Utility LU to test some aspects of the job queue.
10894

10895
  """
10896
  REQ_BGL = False
10897

    
10898
  # Must be lower than default timeout for WaitForJobChange to see whether it
10899
  # notices changed jobs
10900
  _CLIENT_CONNECT_TIMEOUT = 20.0
10901
  _CLIENT_CONFIRM_TIMEOUT = 60.0
10902

    
10903
  @classmethod
10904
  def _NotifyUsingSocket(cls, cb, errcls):
10905
    """Opens a Unix socket and waits for another program to connect.
10906

10907
    @type cb: callable
10908
    @param cb: Callback to send socket name to client
10909
    @type errcls: class
10910
    @param errcls: Exception class to use for errors
10911

10912
    """
10913
    # Using a temporary directory as there's no easy way to create temporary
10914
    # sockets without writing a custom loop around tempfile.mktemp and
10915
    # socket.bind
10916
    tmpdir = tempfile.mkdtemp()
10917
    try:
10918
      tmpsock = utils.PathJoin(tmpdir, "sock")
10919

    
10920
      logging.debug("Creating temporary socket at %s", tmpsock)
10921
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
10922
      try:
10923
        sock.bind(tmpsock)
10924
        sock.listen(1)
10925

    
10926
        # Send details to client
10927
        cb(tmpsock)
10928

    
10929
        # Wait for client to connect before continuing
10930
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
10931
        try:
10932
          (conn, _) = sock.accept()
10933
        except socket.error, err:
10934
          raise errcls("Client didn't connect in time (%s)" % err)
10935
      finally:
10936
        sock.close()
10937
    finally:
10938
      # Remove as soon as client is connected
10939
      shutil.rmtree(tmpdir)
10940

    
10941
    # Wait for client to close
10942
    try:
10943
      try:
10944
        # pylint: disable-msg=E1101
10945
        # Instance of '_socketobject' has no ... member
10946
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
10947
        conn.recv(1)
10948
      except socket.error, err:
10949
        raise errcls("Client failed to confirm notification (%s)" % err)
10950
    finally:
10951
      conn.close()
10952

    
10953
  def _SendNotification(self, test, arg, sockname):
10954
    """Sends a notification to the client.
10955

10956
    @type test: string
10957
    @param test: Test name
10958
    @param arg: Test argument (depends on test)
10959
    @type sockname: string
10960
    @param sockname: Socket path
10961

10962
    """
10963
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
10964

    
10965
  def _Notify(self, prereq, test, arg):
10966
    """Notifies the client of a test.
10967

10968
    @type prereq: bool
10969
    @param prereq: Whether this is a prereq-phase test
10970
    @type test: string
10971
    @param test: Test name
10972
    @param arg: Test argument (depends on test)
10973

10974
    """
10975
    if prereq:
10976
      errcls = errors.OpPrereqError
10977
    else:
10978
      errcls = errors.OpExecError
10979

    
10980
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
10981
                                                  test, arg),
10982
                                   errcls)
10983

    
10984
  def CheckArguments(self):
10985
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
10986
    self.expandnames_calls = 0
10987

    
10988
  def ExpandNames(self):
10989
    checkargs_calls = getattr(self, "checkargs_calls", 0)
10990
    if checkargs_calls < 1:
10991
      raise errors.ProgrammerError("CheckArguments was not called")
10992

    
10993
    self.expandnames_calls += 1
10994

    
10995
    if self.op.notify_waitlock:
10996
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
10997

    
10998
    self.LogInfo("Expanding names")
10999

    
11000
    # Get lock on master node (just to get a lock, not for a particular reason)
11001
    self.needed_locks = {
11002
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
11003
      }
11004

    
11005
  def Exec(self, feedback_fn):
11006
    if self.expandnames_calls < 1:
11007
      raise errors.ProgrammerError("ExpandNames was not called")
11008

    
11009
    if self.op.notify_exec:
11010
      self._Notify(False, constants.JQT_EXEC, None)
11011

    
11012
    self.LogInfo("Executing")
11013

    
11014
    if self.op.log_messages:
11015
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
11016
      for idx, msg in enumerate(self.op.log_messages):
11017
        self.LogInfo("Sending log message %s", idx + 1)
11018
        feedback_fn(constants.JQT_MSGPREFIX + msg)
11019
        # Report how many test messages have been sent
11020
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
11021

    
11022
    if self.op.fail:
11023
      raise errors.OpExecError("Opcode failure was requested")
11024

    
11025
    return True
11026

    
11027

    
11028
class IAllocator(object):
11029
  """IAllocator framework.
11030

11031
  An IAllocator instance has three sets of attributes:
11032
    - cfg that is needed to query the cluster
11033
    - input data (all members of the _KEYS class attribute are required)
11034
    - four buffer attributes (in|out_data|text), that represent the
11035
      input (to the external script) in text and data structure format,
11036
      and the output from it, again in two formats
11037
    - the result variables from the script (success, info, nodes) for
11038
      easy usage
11039

11040
  """
11041
  # pylint: disable-msg=R0902
11042
  # lots of instance attributes
11043
  _ALLO_KEYS = [
11044
    "name", "mem_size", "disks", "disk_template",
11045
    "os", "tags", "nics", "vcpus", "hypervisor",
11046
    ]
11047
  _RELO_KEYS = [
11048
    "name", "relocate_from",
11049
    ]
11050
  _EVAC_KEYS = [
11051
    "evac_nodes",
11052
    ]
11053

    
11054
  def __init__(self, cfg, rpc, mode, **kwargs):
11055
    self.cfg = cfg
11056
    self.rpc = rpc
11057
    # init buffer variables
11058
    self.in_text = self.out_text = self.in_data = self.out_data = None
11059
    # init all input fields so that pylint is happy
11060
    self.mode = mode
11061
    self.mem_size = self.disks = self.disk_template = None
11062
    self.os = self.tags = self.nics = self.vcpus = None
11063
    self.hypervisor = None
11064
    self.relocate_from = None
11065
    self.name = None
11066
    self.evac_nodes = None
11067
    # computed fields
11068
    self.required_nodes = None
11069
    # init result fields
11070
    self.success = self.info = self.result = None
11071
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11072
      keyset = self._ALLO_KEYS
11073
      fn = self._AddNewInstance
11074
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11075
      keyset = self._RELO_KEYS
11076
      fn = self._AddRelocateInstance
11077
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11078
      keyset = self._EVAC_KEYS
11079
      fn = self._AddEvacuateNodes
11080
    else:
11081
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
11082
                                   " IAllocator" % self.mode)
11083
    for key in kwargs:
11084
      if key not in keyset:
11085
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
11086
                                     " IAllocator" % key)
11087
      setattr(self, key, kwargs[key])
11088

    
11089
    for key in keyset:
11090
      if key not in kwargs:
11091
        raise errors.ProgrammerError("Missing input parameter '%s' to"
11092
                                     " IAllocator" % key)
11093
    self._BuildInputData(fn)
11094

    
11095
  def _ComputeClusterData(self):
11096
    """Compute the generic allocator input data.
11097

11098
    This is the data that is independent of the actual operation.
11099

11100
    """
11101
    cfg = self.cfg
11102
    cluster_info = cfg.GetClusterInfo()
11103
    # cluster data
11104
    data = {
11105
      "version": constants.IALLOCATOR_VERSION,
11106
      "cluster_name": cfg.GetClusterName(),
11107
      "cluster_tags": list(cluster_info.GetTags()),
11108
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
11109
      # we don't have job IDs
11110
      }
11111
    ninfo = cfg.GetAllNodesInfo()
11112
    iinfo = cfg.GetAllInstancesInfo().values()
11113
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
11114

    
11115
    # node data
11116
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
11117

    
11118
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11119
      hypervisor_name = self.hypervisor
11120
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11121
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
11122
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11123
      hypervisor_name = cluster_info.enabled_hypervisors[0]
11124

    
11125
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
11126
                                        hypervisor_name)
11127
    node_iinfo = \
11128
      self.rpc.call_all_instances_info(node_list,
11129
                                       cluster_info.enabled_hypervisors)
11130

    
11131
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
11132

    
11133
    config_ndata = self._ComputeBasicNodeData(ninfo)
11134
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
11135
                                                 i_list, config_ndata)
11136
    assert len(data["nodes"]) == len(ninfo), \
11137
        "Incomplete node data computed"
11138

    
11139
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
11140

    
11141
    self.in_data = data
11142

    
11143
  @staticmethod
11144
  def _ComputeNodeGroupData(cfg):
11145
    """Compute node groups data.
11146

11147
    """
11148
    ng = {}
11149
    for guuid, gdata in cfg.GetAllNodeGroupsInfo().items():
11150
      ng[guuid] = {
11151
        "name": gdata.name,
11152
        "alloc_policy": gdata.alloc_policy,
11153
        }
11154
    return ng
11155

    
11156
  @staticmethod
11157
  def _ComputeBasicNodeData(node_cfg):
11158
    """Compute global node data.
11159

11160
    @rtype: dict
11161
    @returns: a dict of name: (node dict, node config)
11162

11163
    """
11164
    node_results = {}
11165
    for ninfo in node_cfg.values():
11166
      # fill in static (config-based) values
11167
      pnr = {
11168
        "tags": list(ninfo.GetTags()),
11169
        "primary_ip": ninfo.primary_ip,
11170
        "secondary_ip": ninfo.secondary_ip,
11171
        "offline": ninfo.offline,
11172
        "drained": ninfo.drained,
11173
        "master_candidate": ninfo.master_candidate,
11174
        "group": ninfo.group,
11175
        "master_capable": ninfo.master_capable,
11176
        "vm_capable": ninfo.vm_capable,
11177
        }
11178

    
11179
      node_results[ninfo.name] = pnr
11180

    
11181
    return node_results
11182

    
11183
  @staticmethod
11184
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
11185
                              node_results):
11186
    """Compute global node data.
11187

11188
    @param node_results: the basic node structures as filled from the config
11189

11190
    """
11191
    # make a copy of the current dict
11192
    node_results = dict(node_results)
11193
    for nname, nresult in node_data.items():
11194
      assert nname in node_results, "Missing basic data for node %s" % nname
11195
      ninfo = node_cfg[nname]
11196

    
11197
      if not (ninfo.offline or ninfo.drained):
11198
        nresult.Raise("Can't get data for node %s" % nname)
11199
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
11200
                                nname)
11201
        remote_info = nresult.payload
11202

    
11203
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
11204
                     'vg_size', 'vg_free', 'cpu_total']:
11205
          if attr not in remote_info:
11206
            raise errors.OpExecError("Node '%s' didn't return attribute"
11207
                                     " '%s'" % (nname, attr))
11208
          if not isinstance(remote_info[attr], int):
11209
            raise errors.OpExecError("Node '%s' returned invalid value"
11210
                                     " for '%s': %s" %
11211
                                     (nname, attr, remote_info[attr]))
11212
        # compute memory used by primary instances
11213
        i_p_mem = i_p_up_mem = 0
11214
        for iinfo, beinfo in i_list:
11215
          if iinfo.primary_node == nname:
11216
            i_p_mem += beinfo[constants.BE_MEMORY]
11217
            if iinfo.name not in node_iinfo[nname].payload:
11218
              i_used_mem = 0
11219
            else:
11220
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
11221
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
11222
            remote_info['memory_free'] -= max(0, i_mem_diff)
11223

    
11224
            if iinfo.admin_up:
11225
              i_p_up_mem += beinfo[constants.BE_MEMORY]
11226

    
11227
        # compute memory used by instances
11228
        pnr_dyn = {
11229
          "total_memory": remote_info['memory_total'],
11230
          "reserved_memory": remote_info['memory_dom0'],
11231
          "free_memory": remote_info['memory_free'],
11232
          "total_disk": remote_info['vg_size'],
11233
          "free_disk": remote_info['vg_free'],
11234
          "total_cpus": remote_info['cpu_total'],
11235
          "i_pri_memory": i_p_mem,
11236
          "i_pri_up_memory": i_p_up_mem,
11237
          }
11238
        pnr_dyn.update(node_results[nname])
11239
        node_results[nname] = pnr_dyn
11240

    
11241
    return node_results
11242

    
11243
  @staticmethod
11244
  def _ComputeInstanceData(cluster_info, i_list):
11245
    """Compute global instance data.
11246

11247
    """
11248
    instance_data = {}
11249
    for iinfo, beinfo in i_list:
11250
      nic_data = []
11251
      for nic in iinfo.nics:
11252
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
11253
        nic_dict = {"mac": nic.mac,
11254
                    "ip": nic.ip,
11255
                    "mode": filled_params[constants.NIC_MODE],
11256
                    "link": filled_params[constants.NIC_LINK],
11257
                   }
11258
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
11259
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
11260
        nic_data.append(nic_dict)
11261
      pir = {
11262
        "tags": list(iinfo.GetTags()),
11263
        "admin_up": iinfo.admin_up,
11264
        "vcpus": beinfo[constants.BE_VCPUS],
11265
        "memory": beinfo[constants.BE_MEMORY],
11266
        "os": iinfo.os,
11267
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
11268
        "nics": nic_data,
11269
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
11270
        "disk_template": iinfo.disk_template,
11271
        "hypervisor": iinfo.hypervisor,
11272
        }
11273
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
11274
                                                 pir["disks"])
11275
      instance_data[iinfo.name] = pir
11276

    
11277
    return instance_data
11278

    
11279
  def _AddNewInstance(self):
11280
    """Add new instance data to allocator structure.
11281

11282
    This in combination with _AllocatorGetClusterData will create the
11283
    correct structure needed as input for the allocator.
11284

11285
    The checks for the completeness of the opcode must have already been
11286
    done.
11287

11288
    """
11289
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
11290

    
11291
    if self.disk_template in constants.DTS_NET_MIRROR:
11292
      self.required_nodes = 2
11293
    else:
11294
      self.required_nodes = 1
11295
    request = {
11296
      "name": self.name,
11297
      "disk_template": self.disk_template,
11298
      "tags": self.tags,
11299
      "os": self.os,
11300
      "vcpus": self.vcpus,
11301
      "memory": self.mem_size,
11302
      "disks": self.disks,
11303
      "disk_space_total": disk_space,
11304
      "nics": self.nics,
11305
      "required_nodes": self.required_nodes,
11306
      }
11307
    return request
11308

    
11309
  def _AddRelocateInstance(self):
11310
    """Add relocate instance data to allocator structure.
11311

11312
    This in combination with _IAllocatorGetClusterData will create the
11313
    correct structure needed as input for the allocator.
11314

11315
    The checks for the completeness of the opcode must have already been
11316
    done.
11317

11318
    """
11319
    instance = self.cfg.GetInstanceInfo(self.name)
11320
    if instance is None:
11321
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
11322
                                   " IAllocator" % self.name)
11323

    
11324
    if instance.disk_template not in constants.DTS_MIRRORED:
11325
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
11326
                                 errors.ECODE_INVAL)
11327

    
11328
    if instance.disk_template in constants.DTS_NET_MIRROR and \
11329
        len(instance.secondary_nodes) != 1:
11330
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
11331
                                 errors.ECODE_STATE)
11332

    
11333
    self.required_nodes = 1
11334
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
11335
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
11336

    
11337
    request = {
11338
      "name": self.name,
11339
      "disk_space_total": disk_space,
11340
      "required_nodes": self.required_nodes,
11341
      "relocate_from": self.relocate_from,
11342
      }
11343
    return request
11344

    
11345
  def _AddEvacuateNodes(self):
11346
    """Add evacuate nodes data to allocator structure.
11347

11348
    """
11349
    request = {
11350
      "evac_nodes": self.evac_nodes
11351
      }
11352
    return request
11353

    
11354
  def _BuildInputData(self, fn):
11355
    """Build input data structures.
11356

11357
    """
11358
    self._ComputeClusterData()
11359

    
11360
    request = fn()
11361
    request["type"] = self.mode
11362
    self.in_data["request"] = request
11363

    
11364
    self.in_text = serializer.Dump(self.in_data)
11365

    
11366
  def Run(self, name, validate=True, call_fn=None):
11367
    """Run an instance allocator and return the results.
11368

11369
    """
11370
    if call_fn is None:
11371
      call_fn = self.rpc.call_iallocator_runner
11372

    
11373
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
11374
    result.Raise("Failure while running the iallocator script")
11375

    
11376
    self.out_text = result.payload
11377
    if validate:
11378
      self._ValidateResult()
11379

    
11380
  def _ValidateResult(self):
11381
    """Process the allocator results.
11382

11383
    This will process and if successful save the result in
11384
    self.out_data and the other parameters.
11385

11386
    """
11387
    try:
11388
      rdict = serializer.Load(self.out_text)
11389
    except Exception, err:
11390
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
11391

    
11392
    if not isinstance(rdict, dict):
11393
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
11394

    
11395
    # TODO: remove backwards compatiblity in later versions
11396
    if "nodes" in rdict and "result" not in rdict:
11397
      rdict["result"] = rdict["nodes"]
11398
      del rdict["nodes"]
11399

    
11400
    for key in "success", "info", "result":
11401
      if key not in rdict:
11402
        raise errors.OpExecError("Can't parse iallocator results:"
11403
                                 " missing key '%s'" % key)
11404
      setattr(self, key, rdict[key])
11405

    
11406
    if not isinstance(rdict["result"], list):
11407
      raise errors.OpExecError("Can't parse iallocator results: 'result' key"
11408
                               " is not a list")
11409
    self.out_data = rdict
11410

    
11411

    
11412
class LUTestAllocator(NoHooksLU):
11413
  """Run allocator tests.
11414

11415
  This LU runs the allocator tests
11416

11417
  """
11418
  def CheckPrereq(self):
11419
    """Check prerequisites.
11420

11421
    This checks the opcode parameters depending on the director and mode test.
11422

11423
    """
11424
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11425
      for attr in ["mem_size", "disks", "disk_template",
11426
                   "os", "tags", "nics", "vcpus"]:
11427
        if not hasattr(self.op, attr):
11428
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
11429
                                     attr, errors.ECODE_INVAL)
11430
      iname = self.cfg.ExpandInstanceName(self.op.name)
11431
      if iname is not None:
11432
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
11433
                                   iname, errors.ECODE_EXISTS)
11434
      if not isinstance(self.op.nics, list):
11435
        raise errors.OpPrereqError("Invalid parameter 'nics'",
11436
                                   errors.ECODE_INVAL)
11437
      if not isinstance(self.op.disks, list):
11438
        raise errors.OpPrereqError("Invalid parameter 'disks'",
11439
                                   errors.ECODE_INVAL)
11440
      for row in self.op.disks:
11441
        if (not isinstance(row, dict) or
11442
            "size" not in row or
11443
            not isinstance(row["size"], int) or
11444
            "mode" not in row or
11445
            row["mode"] not in ['r', 'w']):
11446
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
11447
                                     " parameter", errors.ECODE_INVAL)
11448
      if self.op.hypervisor is None:
11449
        self.op.hypervisor = self.cfg.GetHypervisorType()
11450
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11451
      fname = _ExpandInstanceName(self.cfg, self.op.name)
11452
      self.op.name = fname
11453
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
11454
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11455
      if not hasattr(self.op, "evac_nodes"):
11456
        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
11457
                                   " opcode input", errors.ECODE_INVAL)
11458
    else:
11459
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
11460
                                 self.op.mode, errors.ECODE_INVAL)
11461

    
11462
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
11463
      if self.op.allocator is None:
11464
        raise errors.OpPrereqError("Missing allocator name",
11465
                                   errors.ECODE_INVAL)
11466
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
11467
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
11468
                                 self.op.direction, errors.ECODE_INVAL)
11469

    
11470
  def Exec(self, feedback_fn):
11471
    """Run the allocator test.
11472

11473
    """
11474
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11475
      ial = IAllocator(self.cfg, self.rpc,
11476
                       mode=self.op.mode,
11477
                       name=self.op.name,
11478
                       mem_size=self.op.mem_size,
11479
                       disks=self.op.disks,
11480
                       disk_template=self.op.disk_template,
11481
                       os=self.op.os,
11482
                       tags=self.op.tags,
11483
                       nics=self.op.nics,
11484
                       vcpus=self.op.vcpus,
11485
                       hypervisor=self.op.hypervisor,
11486
                       )
11487
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11488
      ial = IAllocator(self.cfg, self.rpc,
11489
                       mode=self.op.mode,
11490
                       name=self.op.name,
11491
                       relocate_from=list(self.relocate_from),
11492
                       )
11493
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11494
      ial = IAllocator(self.cfg, self.rpc,
11495
                       mode=self.op.mode,
11496
                       evac_nodes=self.op.evac_nodes)
11497
    else:
11498
      raise errors.ProgrammerError("Uncatched mode %s in"
11499
                                   " LUTestAllocator.Exec", self.op.mode)
11500

    
11501
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
11502
      result = ial.in_text
11503
    else:
11504
      ial.Run(self.op.allocator, validate=False)
11505
      result = ial.out_text
11506
    return result
11507

    
11508

    
11509
#: Query type implementations
11510
_QUERY_IMPL = {
11511
  constants.QR_INSTANCE: _InstanceQuery,
11512
  constants.QR_NODE: _NodeQuery,
11513
  constants.QR_GROUP: _GroupQuery,
11514
  }
11515

    
11516

    
11517
def _GetQueryImplementation(name):
11518
  """Returns the implemtnation for a query type.
11519

11520
  @param name: Query type, must be one of L{constants.QR_OP_QUERY}
11521

11522
  """
11523
  try:
11524
    return _QUERY_IMPL[name]
11525
  except KeyError:
11526
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
11527
                               errors.ECODE_INVAL)