Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 9a833381

History | View | Annotate | Download (405.8 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 CheckArguments(self):
6060
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
6061

    
6062
  def ExpandNames(self):
6063
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6064

    
6065
    self.needed_locks = {}
6066

    
6067
    # Create tasklets for migrating instances for all instances on this node
6068
    names = []
6069
    tasklets = []
6070

    
6071
    self.lock_all_nodes = False
6072

    
6073
    for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name):
6074
      logging.debug("Migrating instance %s", inst.name)
6075
      names.append(inst.name)
6076

    
6077
      tasklets.append(TLMigrateInstance(self, inst.name, False,
6078
                                        self.op.iallocator, None))
6079

    
6080
      if inst.disk_template in constants.DTS_EXT_MIRROR:
6081
        # We need to lock all nodes, as the iallocator will choose the
6082
        # destination nodes afterwards
6083
        self.lock_all_nodes = True
6084

    
6085
    self.tasklets = tasklets
6086

    
6087
    # Declare node locks
6088
    if self.lock_all_nodes:
6089
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6090
    else:
6091
      self.needed_locks[locking.LEVEL_NODE] = [self.op.node_name]
6092
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6093

    
6094
    # Declare instance locks
6095
    self.needed_locks[locking.LEVEL_INSTANCE] = names
6096

    
6097
  def DeclareLocks(self, level):
6098
    if level == locking.LEVEL_NODE and not self.lock_all_nodes:
6099
      self._LockInstancesNodes()
6100

    
6101
  def BuildHooksEnv(self):
6102
    """Build hooks env.
6103

6104
    This runs on the master, the primary and all the secondaries.
6105

6106
    """
6107
    env = {
6108
      "NODE_NAME": self.op.node_name,
6109
      }
6110

    
6111
    nl = [self.cfg.GetMasterNode()]
6112

    
6113
    return (env, nl, nl)
6114

    
6115

    
6116
class TLMigrateInstance(Tasklet):
6117
  """Tasklet class for instance migration.
6118

6119
  @type live: boolean
6120
  @ivar live: whether the migration will be done live or non-live;
6121
      this variable is initalized only after CheckPrereq has run
6122

6123
  """
6124
  def __init__(self, lu, instance_name, cleanup,
6125
               iallocator=None, target_node=None):
6126
    """Initializes this class.
6127

6128
    """
6129
    Tasklet.__init__(self, lu)
6130

    
6131
    # Parameters
6132
    self.instance_name = instance_name
6133
    self.cleanup = cleanup
6134
    self.live = False # will be overridden later
6135
    self.iallocator = iallocator
6136
    self.target_node = target_node
6137

    
6138
  def CheckPrereq(self):
6139
    """Check prerequisites.
6140

6141
    This checks that the instance is in the cluster.
6142

6143
    """
6144
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
6145
    instance = self.cfg.GetInstanceInfo(instance_name)
6146
    assert instance is not None
6147
    self.instance = instance
6148

    
6149
    if instance.disk_template not in constants.DTS_MIRRORED:
6150
      raise errors.OpPrereqError("Instance's disk layout '%s' does not allow"
6151
                                 " migrations" % instance.disk_template,
6152
                                 errors.ECODE_STATE)
6153

    
6154
    if instance.disk_template in constants.DTS_EXT_MIRROR:
6155
      if [self.iallocator, self.target_node].count(None) != 1:
6156
        raise errors.OpPrereqError("Do not specify both, iallocator and"
6157
                                   " target node", errors.ECODE_INVAL)
6158

    
6159
      if self.iallocator:
6160
        self._RunAllocator()
6161

    
6162
      # self.target_node is already populated, either directly or by the
6163
      # iallocator run
6164
      target_node = self.target_node
6165

    
6166
      if len(self.lu.tasklets) == 1:
6167
        # It is safe to remove locks only when we're the only tasklet in the LU
6168
        nodes_keep = [instance.primary_node, self.target_node]
6169
        nodes_rel = [node for node in self.lu.acquired_locks[locking.LEVEL_NODE]
6170
                     if node not in nodes_keep]
6171
        self.lu.context.glm.release(locking.LEVEL_NODE, nodes_rel)
6172
        self.lu.acquired_locks[locking.LEVEL_NODE] = nodes_keep
6173

    
6174
    else:
6175
      secondary_nodes = instance.secondary_nodes
6176
      if not secondary_nodes:
6177
        raise errors.ConfigurationError("No secondary node but using"
6178
                                        " %s disk template" %
6179
                                        instance.disk_template)
6180
      target_node = secondary_nodes[0]
6181

    
6182
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
6183

    
6184
    # check memory requirements on the secondary node
6185
    _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
6186
                         instance.name, i_be[constants.BE_MEMORY],
6187
                         instance.hypervisor)
6188

    
6189
    # check bridge existance
6190
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
6191

    
6192
    if not self.cleanup:
6193
      _CheckNodeNotDrained(self.lu, target_node)
6194
      result = self.rpc.call_instance_migratable(instance.primary_node,
6195
                                                 instance)
6196
      result.Raise("Can't migrate, please use failover",
6197
                   prereq=True, ecode=errors.ECODE_STATE)
6198

    
6199

    
6200
  def _RunAllocator(self):
6201
    """Run the allocator based on input opcode.
6202

6203
    """
6204
    ial = IAllocator(self.cfg, self.rpc,
6205
                     mode=constants.IALLOCATOR_MODE_RELOC,
6206
                     name=self.instance_name,
6207
                     # TODO See why hail breaks with a single node below
6208
                     relocate_from=[self.instance.primary_node,
6209
                                    self.instance.primary_node],
6210
                     )
6211

    
6212
    ial.Run(self.iallocator)
6213

    
6214
    if not ial.success:
6215
      raise errors.OpPrereqError("Can't compute nodes using"
6216
                                 " iallocator '%s': %s" %
6217
                                 (self.iallocator, ial.info),
6218
                                 errors.ECODE_NORES)
6219
    if len(ial.result) != ial.required_nodes:
6220
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
6221
                                 " of nodes (%s), required %s" %
6222
                                 (self.iallocator, len(ial.result),
6223
                                  ial.required_nodes), errors.ECODE_FAULT)
6224
    self.target_node = ial.result[0]
6225
    self.lu.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
6226
                 self.instance_name, self.iallocator,
6227
                 utils.CommaJoin(ial.result))
6228

    
6229
    if self.lu.op.live is not None and self.lu.op.mode is not None:
6230
      raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
6231
                                 " parameters are accepted",
6232
                                 errors.ECODE_INVAL)
6233
    if self.lu.op.live is not None:
6234
      if self.lu.op.live:
6235
        self.lu.op.mode = constants.HT_MIGRATION_LIVE
6236
      else:
6237
        self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
6238
      # reset the 'live' parameter to None so that repeated
6239
      # invocations of CheckPrereq do not raise an exception
6240
      self.lu.op.live = None
6241
    elif self.lu.op.mode is None:
6242
      # read the default value from the hypervisor
6243
      i_hv = self.cfg.GetClusterInfo().FillHV(self.instance, skip_globals=False)
6244
      self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
6245

    
6246
    self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
6247

    
6248
  def _WaitUntilSync(self):
6249
    """Poll with custom rpc for disk sync.
6250

6251
    This uses our own step-based rpc call.
6252

6253
    """
6254
    self.feedback_fn("* wait until resync is done")
6255
    all_done = False
6256
    while not all_done:
6257
      all_done = True
6258
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
6259
                                            self.nodes_ip,
6260
                                            self.instance.disks)
6261
      min_percent = 100
6262
      for node, nres in result.items():
6263
        nres.Raise("Cannot resync disks on node %s" % node)
6264
        node_done, node_percent = nres.payload
6265
        all_done = all_done and node_done
6266
        if node_percent is not None:
6267
          min_percent = min(min_percent, node_percent)
6268
      if not all_done:
6269
        if min_percent < 100:
6270
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
6271
        time.sleep(2)
6272

    
6273
  def _EnsureSecondary(self, node):
6274
    """Demote a node to secondary.
6275

6276
    """
6277
    self.feedback_fn("* switching node %s to secondary mode" % node)
6278

    
6279
    for dev in self.instance.disks:
6280
      self.cfg.SetDiskID(dev, node)
6281

    
6282
    result = self.rpc.call_blockdev_close(node, self.instance.name,
6283
                                          self.instance.disks)
6284
    result.Raise("Cannot change disk to secondary on node %s" % node)
6285

    
6286
  def _GoStandalone(self):
6287
    """Disconnect from the network.
6288

6289
    """
6290
    self.feedback_fn("* changing into standalone mode")
6291
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
6292
                                               self.instance.disks)
6293
    for node, nres in result.items():
6294
      nres.Raise("Cannot disconnect disks node %s" % node)
6295

    
6296
  def _GoReconnect(self, multimaster):
6297
    """Reconnect to the network.
6298

6299
    """
6300
    if multimaster:
6301
      msg = "dual-master"
6302
    else:
6303
      msg = "single-master"
6304
    self.feedback_fn("* changing disks into %s mode" % msg)
6305
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
6306
                                           self.instance.disks,
6307
                                           self.instance.name, multimaster)
6308
    for node, nres in result.items():
6309
      nres.Raise("Cannot change disks config on node %s" % node)
6310

    
6311
  def _ExecCleanup(self):
6312
    """Try to cleanup after a failed migration.
6313

6314
    The cleanup is done by:
6315
      - check that the instance is running only on one node
6316
        (and update the config if needed)
6317
      - change disks on its secondary node to secondary
6318
      - wait until disks are fully synchronized
6319
      - disconnect from the network
6320
      - change disks into single-master mode
6321
      - wait again until disks are fully synchronized
6322

6323
    """
6324
    instance = self.instance
6325
    target_node = self.target_node
6326
    source_node = self.source_node
6327

    
6328
    # check running on only one node
6329
    self.feedback_fn("* checking where the instance actually runs"
6330
                     " (if this hangs, the hypervisor might be in"
6331
                     " a bad state)")
6332
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
6333
    for node, result in ins_l.items():
6334
      result.Raise("Can't contact node %s" % node)
6335

    
6336
    runningon_source = instance.name in ins_l[source_node].payload
6337
    runningon_target = instance.name in ins_l[target_node].payload
6338

    
6339
    if runningon_source and runningon_target:
6340
      raise errors.OpExecError("Instance seems to be running on two nodes,"
6341
                               " or the hypervisor is confused. You will have"
6342
                               " to ensure manually that it runs only on one"
6343
                               " and restart this operation.")
6344

    
6345
    if not (runningon_source or runningon_target):
6346
      raise errors.OpExecError("Instance does not seem to be running at all."
6347
                               " In this case, it's safer to repair by"
6348
                               " running 'gnt-instance stop' to ensure disk"
6349
                               " shutdown, and then restarting it.")
6350

    
6351
    if runningon_target:
6352
      # the migration has actually succeeded, we need to update the config
6353
      self.feedback_fn("* instance running on secondary node (%s),"
6354
                       " updating config" % target_node)
6355
      instance.primary_node = target_node
6356
      self.cfg.Update(instance, self.feedback_fn)
6357
      demoted_node = source_node
6358
    else:
6359
      self.feedback_fn("* instance confirmed to be running on its"
6360
                       " primary node (%s)" % source_node)
6361
      demoted_node = target_node
6362

    
6363
    if instance.disk_template in constants.DTS_NET_MIRROR:
6364
      self._EnsureSecondary(demoted_node)
6365
      try:
6366
        self._WaitUntilSync()
6367
      except errors.OpExecError:
6368
        # we ignore here errors, since if the device is standalone, it
6369
        # won't be able to sync
6370
        pass
6371
      self._GoStandalone()
6372
      self._GoReconnect(False)
6373
      self._WaitUntilSync()
6374

    
6375
    self.feedback_fn("* done")
6376

    
6377
  def _RevertDiskStatus(self):
6378
    """Try to revert the disk status after a failed migration.
6379

6380
    """
6381
    target_node = self.target_node
6382
    if self.instance.disk_template in constants.DTS_EXT_MIRROR:
6383
      return
6384

    
6385
    try:
6386
      self._EnsureSecondary(target_node)
6387
      self._GoStandalone()
6388
      self._GoReconnect(False)
6389
      self._WaitUntilSync()
6390
    except errors.OpExecError, err:
6391
      self.lu.LogWarning("Migration failed and I can't reconnect the"
6392
                         " drives: error '%s'\n"
6393
                         "Please look and recover the instance status" %
6394
                         str(err))
6395

    
6396
  def _AbortMigration(self):
6397
    """Call the hypervisor code to abort a started migration.
6398

6399
    """
6400
    instance = self.instance
6401
    target_node = self.target_node
6402
    migration_info = self.migration_info
6403

    
6404
    abort_result = self.rpc.call_finalize_migration(target_node,
6405
                                                    instance,
6406
                                                    migration_info,
6407
                                                    False)
6408
    abort_msg = abort_result.fail_msg
6409
    if abort_msg:
6410
      logging.error("Aborting migration failed on target node %s: %s",
6411
                    target_node, abort_msg)
6412
      # Don't raise an exception here, as we stil have to try to revert the
6413
      # disk status, even if this step failed.
6414

    
6415
  def _ExecMigration(self):
6416
    """Migrate an instance.
6417

6418
    The migrate is done by:
6419
      - change the disks into dual-master mode
6420
      - wait until disks are fully synchronized again
6421
      - migrate the instance
6422
      - change disks on the new secondary node (the old primary) to secondary
6423
      - wait until disks are fully synchronized
6424
      - change disks into single-master mode
6425

6426
    """
6427
    instance = self.instance
6428
    target_node = self.target_node
6429
    source_node = self.source_node
6430

    
6431
    self.feedback_fn("* checking disk consistency between source and target")
6432
    for dev in instance.disks:
6433
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
6434
        raise errors.OpExecError("Disk %s is degraded or not fully"
6435
                                 " synchronized on target node,"
6436
                                 " aborting migrate." % dev.iv_name)
6437

    
6438
    # First get the migration information from the remote node
6439
    result = self.rpc.call_migration_info(source_node, instance)
6440
    msg = result.fail_msg
6441
    if msg:
6442
      log_err = ("Failed fetching source migration information from %s: %s" %
6443
                 (source_node, msg))
6444
      logging.error(log_err)
6445
      raise errors.OpExecError(log_err)
6446

    
6447
    self.migration_info = migration_info = result.payload
6448

    
6449
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
6450
      # Then switch the disks to master/master mode
6451
      self._EnsureSecondary(target_node)
6452
      self._GoStandalone()
6453
      self._GoReconnect(True)
6454
      self._WaitUntilSync()
6455

    
6456
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
6457
    result = self.rpc.call_accept_instance(target_node,
6458
                                           instance,
6459
                                           migration_info,
6460
                                           self.nodes_ip[target_node])
6461

    
6462
    msg = result.fail_msg
6463
    if msg:
6464
      logging.error("Instance pre-migration failed, trying to revert"
6465
                    " disk status: %s", msg)
6466
      self.feedback_fn("Pre-migration failed, aborting")
6467
      self._AbortMigration()
6468
      self._RevertDiskStatus()
6469
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
6470
                               (instance.name, msg))
6471

    
6472
    self.feedback_fn("* migrating instance to %s" % target_node)
6473
    time.sleep(10)
6474
    result = self.rpc.call_instance_migrate(source_node, instance,
6475
                                            self.nodes_ip[target_node],
6476
                                            self.live)
6477
    msg = result.fail_msg
6478
    if msg:
6479
      logging.error("Instance migration failed, trying to revert"
6480
                    " disk status: %s", msg)
6481
      self.feedback_fn("Migration failed, aborting")
6482
      self._AbortMigration()
6483
      self._RevertDiskStatus()
6484
      raise errors.OpExecError("Could not migrate instance %s: %s" %
6485
                               (instance.name, msg))
6486
    time.sleep(10)
6487

    
6488
    instance.primary_node = target_node
6489
    # distribute new instance config to the other nodes
6490
    self.cfg.Update(instance, self.feedback_fn)
6491

    
6492
    result = self.rpc.call_finalize_migration(target_node,
6493
                                              instance,
6494
                                              migration_info,
6495
                                              True)
6496
    msg = result.fail_msg
6497
    if msg:
6498
      logging.error("Instance migration succeeded, but finalization failed:"
6499
                    " %s", msg)
6500
      raise errors.OpExecError("Could not finalize instance migration: %s" %
6501
                               msg)
6502

    
6503
    if self.instance.disk_template not in constants.DTS_EXT_MIRROR:
6504
      self._EnsureSecondary(source_node)
6505
      self._WaitUntilSync()
6506
      self._GoStandalone()
6507
      self._GoReconnect(False)
6508
      self._WaitUntilSync()
6509

    
6510
    self.feedback_fn("* done")
6511

    
6512
  def Exec(self, feedback_fn):
6513
    """Perform the migration.
6514

6515
    """
6516
    feedback_fn("Migrating instance %s" % self.instance.name)
6517

    
6518
    self.feedback_fn = feedback_fn
6519

    
6520
    self.source_node = self.instance.primary_node
6521

    
6522
    # FIXME: if we implement migrate-to-any in DRBD, this needs fixing
6523
    if self.instance.disk_template in constants.DTS_NET_MIRROR:
6524
      self.target_node = self.instance.secondary_nodes[0]
6525
      # Otherwise self.target_node has been populated either
6526
      # directly, or through an iallocator.
6527

    
6528
    self.all_nodes = [self.source_node, self.target_node]
6529
    self.nodes_ip = {
6530
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
6531
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
6532
      }
6533

    
6534
    if self.cleanup:
6535
      return self._ExecCleanup()
6536
    else:
6537
      return self._ExecMigration()
6538

    
6539

    
6540
def _CreateBlockDev(lu, node, instance, device, force_create,
6541
                    info, force_open):
6542
  """Create a tree of block devices on a given node.
6543

6544
  If this device type has to be created on secondaries, create it and
6545
  all its children.
6546

6547
  If not, just recurse to children keeping the same 'force' value.
6548

6549
  @param lu: the lu on whose behalf we execute
6550
  @param node: the node on which to create the device
6551
  @type instance: L{objects.Instance}
6552
  @param instance: the instance which owns the device
6553
  @type device: L{objects.Disk}
6554
  @param device: the device to create
6555
  @type force_create: boolean
6556
  @param force_create: whether to force creation of this device; this
6557
      will be change to True whenever we find a device which has
6558
      CreateOnSecondary() attribute
6559
  @param info: the extra 'metadata' we should attach to the device
6560
      (this will be represented as a LVM tag)
6561
  @type force_open: boolean
6562
  @param force_open: this parameter will be passes to the
6563
      L{backend.BlockdevCreate} function where it specifies
6564
      whether we run on primary or not, and it affects both
6565
      the child assembly and the device own Open() execution
6566

6567
  """
6568
  if device.CreateOnSecondary():
6569
    force_create = True
6570

    
6571
  if device.children:
6572
    for child in device.children:
6573
      _CreateBlockDev(lu, node, instance, child, force_create,
6574
                      info, force_open)
6575

    
6576
  if not force_create:
6577
    return
6578

    
6579
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
6580

    
6581

    
6582
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
6583
  """Create a single block device on a given node.
6584

6585
  This will not recurse over children of the device, so they must be
6586
  created in advance.
6587

6588
  @param lu: the lu on whose behalf we execute
6589
  @param node: the node on which to create the device
6590
  @type instance: L{objects.Instance}
6591
  @param instance: the instance which owns the device
6592
  @type device: L{objects.Disk}
6593
  @param device: the device to create
6594
  @param info: the extra 'metadata' we should attach to the device
6595
      (this will be represented as a LVM tag)
6596
  @type force_open: boolean
6597
  @param force_open: this parameter will be passes to the
6598
      L{backend.BlockdevCreate} function where it specifies
6599
      whether we run on primary or not, and it affects both
6600
      the child assembly and the device own Open() execution
6601

6602
  """
6603
  lu.cfg.SetDiskID(device, node)
6604
  result = lu.rpc.call_blockdev_create(node, device, device.size,
6605
                                       instance.name, force_open, info)
6606
  result.Raise("Can't create block device %s on"
6607
               " node %s for instance %s" % (device, node, instance.name))
6608
  if device.physical_id is None:
6609
    device.physical_id = result.payload
6610

    
6611

    
6612
def _GenerateUniqueNames(lu, exts):
6613
  """Generate a suitable LV name.
6614

6615
  This will generate a logical volume name for the given instance.
6616

6617
  """
6618
  results = []
6619
  for val in exts:
6620
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
6621
    results.append("%s%s" % (new_id, val))
6622
  return results
6623

    
6624

    
6625
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgname, names, iv_name,
6626
                         p_minor, s_minor):
6627
  """Generate a drbd8 device complete with its children.
6628

6629
  """
6630
  port = lu.cfg.AllocatePort()
6631
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
6632
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
6633
                          logical_id=(vgname, names[0]))
6634
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
6635
                          logical_id=(vgname, names[1]))
6636
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
6637
                          logical_id=(primary, secondary, port,
6638
                                      p_minor, s_minor,
6639
                                      shared_secret),
6640
                          children=[dev_data, dev_meta],
6641
                          iv_name=iv_name)
6642
  return drbd_dev
6643

    
6644

    
6645
def _GenerateDiskTemplate(lu, template_name,
6646
                          instance_name, primary_node,
6647
                          secondary_nodes, disk_info,
6648
                          file_storage_dir, file_driver,
6649
                          base_index, feedback_fn):
6650
  """Generate the entire disk layout for a given template type.
6651

6652
  """
6653
  #TODO: compute space requirements
6654

    
6655
  vgname = lu.cfg.GetVGName()
6656
  disk_count = len(disk_info)
6657
  disks = []
6658
  if template_name == constants.DT_DISKLESS:
6659
    pass
6660
  elif template_name == constants.DT_PLAIN:
6661
    if len(secondary_nodes) != 0:
6662
      raise errors.ProgrammerError("Wrong template configuration")
6663

    
6664
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6665
                                      for i in range(disk_count)])
6666
    for idx, disk in enumerate(disk_info):
6667
      disk_index = idx + base_index
6668
      vg = disk.get("vg", vgname)
6669
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
6670
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
6671
                              logical_id=(vg, names[idx]),
6672
                              iv_name="disk/%d" % disk_index,
6673
                              mode=disk["mode"])
6674
      disks.append(disk_dev)
6675
  elif template_name == constants.DT_DRBD8:
6676
    if len(secondary_nodes) != 1:
6677
      raise errors.ProgrammerError("Wrong template configuration")
6678
    remote_node = secondary_nodes[0]
6679
    minors = lu.cfg.AllocateDRBDMinor(
6680
      [primary_node, remote_node] * len(disk_info), instance_name)
6681

    
6682
    names = []
6683
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6684
                                               for i in range(disk_count)]):
6685
      names.append(lv_prefix + "_data")
6686
      names.append(lv_prefix + "_meta")
6687
    for idx, disk in enumerate(disk_info):
6688
      disk_index = idx + base_index
6689
      vg = disk.get("vg", vgname)
6690
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
6691
                                      disk["size"], vg, names[idx*2:idx*2+2],
6692
                                      "disk/%d" % disk_index,
6693
                                      minors[idx*2], minors[idx*2+1])
6694
      disk_dev.mode = disk["mode"]
6695
      disks.append(disk_dev)
6696
  elif template_name == constants.DT_FILE:
6697
    if len(secondary_nodes) != 0:
6698
      raise errors.ProgrammerError("Wrong template configuration")
6699

    
6700
    opcodes.RequireFileStorage()
6701

    
6702
    for idx, disk in enumerate(disk_info):
6703
      disk_index = idx + base_index
6704
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6705
                              iv_name="disk/%d" % disk_index,
6706
                              logical_id=(file_driver,
6707
                                          "%s/disk%d" % (file_storage_dir,
6708
                                                         disk_index)),
6709
                              mode=disk["mode"])
6710
      disks.append(disk_dev)
6711
  elif template_name == constants.DT_SHARED_FILE:
6712
    if len(secondary_nodes) != 0:
6713
      raise errors.ProgrammerError("Wrong template configuration")
6714

    
6715
    opcodes.RequireSharedFileStorage()
6716

    
6717
    for idx, disk in enumerate(disk_info):
6718
      disk_index = idx + base_index
6719
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6720
                              iv_name="disk/%d" % disk_index,
6721
                              logical_id=(file_driver,
6722
                                          "%s/disk%d" % (file_storage_dir,
6723
                                                         disk_index)),
6724
                              mode=disk["mode"])
6725
      disks.append(disk_dev)
6726
  elif template_name == constants.DT_BLOCK:
6727
    if len(secondary_nodes) != 0:
6728
      raise errors.ProgrammerError("Wrong template configuration")
6729

    
6730
    for idx, disk in enumerate(disk_info):
6731
      disk_index = idx + base_index
6732
      disk_dev = objects.Disk(dev_type=constants.LD_BLOCKDEV, size=disk["size"],
6733
                              logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
6734
                                          disk["adopt"]),
6735
                              iv_name="disk/%d" % disk_index,
6736
                              mode=disk["mode"])
6737
      disks.append(disk_dev)
6738

    
6739
  else:
6740
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
6741
  return disks
6742

    
6743

    
6744
def _GetInstanceInfoText(instance):
6745
  """Compute that text that should be added to the disk's metadata.
6746

6747
  """
6748
  return "originstname+%s" % instance.name
6749

    
6750

    
6751
def _CalcEta(time_taken, written, total_size):
6752
  """Calculates the ETA based on size written and total size.
6753

6754
  @param time_taken: The time taken so far
6755
  @param written: amount written so far
6756
  @param total_size: The total size of data to be written
6757
  @return: The remaining time in seconds
6758

6759
  """
6760
  avg_time = time_taken / float(written)
6761
  return (total_size - written) * avg_time
6762

    
6763

    
6764
def _WipeDisks(lu, instance):
6765
  """Wipes instance disks.
6766

6767
  @type lu: L{LogicalUnit}
6768
  @param lu: the logical unit on whose behalf we execute
6769
  @type instance: L{objects.Instance}
6770
  @param instance: the instance whose disks we should create
6771
  @return: the success of the wipe
6772

6773
  """
6774
  node = instance.primary_node
6775

    
6776
  for device in instance.disks:
6777
    lu.cfg.SetDiskID(device, node)
6778

    
6779
  logging.info("Pause sync of instance %s disks", instance.name)
6780
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
6781

    
6782
  for idx, success in enumerate(result.payload):
6783
    if not success:
6784
      logging.warn("pause-sync of instance %s for disks %d failed",
6785
                   instance.name, idx)
6786

    
6787
  try:
6788
    for idx, device in enumerate(instance.disks):
6789
      lu.LogInfo("* Wiping disk %d", idx)
6790
      logging.info("Wiping disk %d for instance %s, node %s",
6791
                   idx, instance.name, node)
6792

    
6793
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
6794
      # MAX_WIPE_CHUNK at max
6795
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
6796
                            constants.MIN_WIPE_CHUNK_PERCENT)
6797

    
6798
      offset = 0
6799
      size = device.size
6800
      last_output = 0
6801
      start_time = time.time()
6802

    
6803
      while offset < size:
6804
        wipe_size = min(wipe_chunk_size, size - offset)
6805
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
6806
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
6807
                     (idx, offset, wipe_size))
6808
        now = time.time()
6809
        offset += wipe_size
6810
        if now - last_output >= 60:
6811
          eta = _CalcEta(now - start_time, offset, size)
6812
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
6813
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
6814
          last_output = now
6815
  finally:
6816
    logging.info("Resume sync of instance %s disks", instance.name)
6817

    
6818
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
6819

    
6820
    for idx, success in enumerate(result.payload):
6821
      if not success:
6822
        lu.LogWarning("Warning: Resume sync of disk %d failed. Please have a"
6823
                      " look at the status and troubleshoot the issue.", idx)
6824
        logging.warn("resume-sync of instance %s for disks %d failed",
6825
                     instance.name, idx)
6826

    
6827

    
6828
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
6829
  """Create all disks for an instance.
6830

6831
  This abstracts away some work from AddInstance.
6832

6833
  @type lu: L{LogicalUnit}
6834
  @param lu: the logical unit on whose behalf we execute
6835
  @type instance: L{objects.Instance}
6836
  @param instance: the instance whose disks we should create
6837
  @type to_skip: list
6838
  @param to_skip: list of indices to skip
6839
  @type target_node: string
6840
  @param target_node: if passed, overrides the target node for creation
6841
  @rtype: boolean
6842
  @return: the success of the creation
6843

6844
  """
6845
  info = _GetInstanceInfoText(instance)
6846
  if target_node is None:
6847
    pnode = instance.primary_node
6848
    all_nodes = instance.all_nodes
6849
  else:
6850
    pnode = target_node
6851
    all_nodes = [pnode]
6852

    
6853
  if instance.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE):
6854
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6855
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
6856

    
6857
    result.Raise("Failed to create directory '%s' on"
6858
                 " node %s" % (file_storage_dir, pnode))
6859

    
6860
  # Note: this needs to be kept in sync with adding of disks in
6861
  # LUInstanceSetParams
6862
  for idx, device in enumerate(instance.disks):
6863
    if to_skip and idx in to_skip:
6864
      continue
6865
    logging.info("Creating volume %s for instance %s",
6866
                 device.iv_name, instance.name)
6867
    #HARDCODE
6868
    for node in all_nodes:
6869
      f_create = node == pnode
6870
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
6871

    
6872

    
6873
def _RemoveDisks(lu, instance, target_node=None):
6874
  """Remove all disks for an instance.
6875

6876
  This abstracts away some work from `AddInstance()` and
6877
  `RemoveInstance()`. Note that in case some of the devices couldn't
6878
  be removed, the removal will continue with the other ones (compare
6879
  with `_CreateDisks()`).
6880

6881
  @type lu: L{LogicalUnit}
6882
  @param lu: the logical unit on whose behalf we execute
6883
  @type instance: L{objects.Instance}
6884
  @param instance: the instance whose disks we should remove
6885
  @type target_node: string
6886
  @param target_node: used to override the node on which to remove the disks
6887
  @rtype: boolean
6888
  @return: the success of the removal
6889

6890
  """
6891
  logging.info("Removing block devices for instance %s", instance.name)
6892

    
6893
  all_result = True
6894
  for device in instance.disks:
6895
    if target_node:
6896
      edata = [(target_node, device)]
6897
    else:
6898
      edata = device.ComputeNodeTree(instance.primary_node)
6899
    for node, disk in edata:
6900
      lu.cfg.SetDiskID(disk, node)
6901
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
6902
      if msg:
6903
        lu.LogWarning("Could not remove block device %s on node %s,"
6904
                      " continuing anyway: %s", device.iv_name, node, msg)
6905
        all_result = False
6906

    
6907
  if instance.disk_template == constants.DT_FILE:
6908
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6909
    if target_node:
6910
      tgt = target_node
6911
    else:
6912
      tgt = instance.primary_node
6913
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
6914
    if result.fail_msg:
6915
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
6916
                    file_storage_dir, instance.primary_node, result.fail_msg)
6917
      all_result = False
6918

    
6919
  return all_result
6920

    
6921

    
6922
def _ComputeDiskSizePerVG(disk_template, disks):
6923
  """Compute disk size requirements in the volume group
6924

6925
  """
6926
  def _compute(disks, payload):
6927
    """Universal algorithm
6928

6929
    """
6930
    vgs = {}
6931
    for disk in disks:
6932
      vgs[disk["vg"]] = vgs.get("vg", 0) + disk["size"] + payload
6933

    
6934
    return vgs
6935

    
6936
  # Required free disk space as a function of disk and swap space
6937
  req_size_dict = {
6938
    constants.DT_DISKLESS: {},
6939
    constants.DT_PLAIN: _compute(disks, 0),
6940
    # 128 MB are added for drbd metadata for each disk
6941
    constants.DT_DRBD8: _compute(disks, 128),
6942
    constants.DT_FILE: {},
6943
    constants.DT_SHARED_FILE: {},
6944
  }
6945

    
6946
  if disk_template not in req_size_dict:
6947
    raise errors.ProgrammerError("Disk template '%s' size requirement"
6948
                                 " is unknown" %  disk_template)
6949

    
6950
  return req_size_dict[disk_template]
6951

    
6952

    
6953
def _ComputeDiskSize(disk_template, disks):
6954
  """Compute disk size requirements in the volume group
6955

6956
  """
6957
  # Required free disk space as a function of disk and swap space
6958
  req_size_dict = {
6959
    constants.DT_DISKLESS: None,
6960
    constants.DT_PLAIN: sum(d["size"] for d in disks),
6961
    # 128 MB are added for drbd metadata for each disk
6962
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
6963
    constants.DT_FILE: None,
6964
    constants.DT_SHARED_FILE: 0,
6965
    constants.DT_BLOCK: 0,
6966
  }
6967

    
6968
  if disk_template not in req_size_dict:
6969
    raise errors.ProgrammerError("Disk template '%s' size requirement"
6970
                                 " is unknown" %  disk_template)
6971

    
6972
  return req_size_dict[disk_template]
6973

    
6974

    
6975
def _FilterVmNodes(lu, nodenames):
6976
  """Filters out non-vm_capable nodes from a list.
6977

6978
  @type lu: L{LogicalUnit}
6979
  @param lu: the logical unit for which we check
6980
  @type nodenames: list
6981
  @param nodenames: the list of nodes on which we should check
6982
  @rtype: list
6983
  @return: the list of vm-capable nodes
6984

6985
  """
6986
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
6987
  return [name for name in nodenames if name not in vm_nodes]
6988

    
6989

    
6990
def _CheckHVParams(lu, nodenames, hvname, hvparams):
6991
  """Hypervisor parameter validation.
6992

6993
  This function abstract the hypervisor parameter validation to be
6994
  used in both instance create and instance modify.
6995

6996
  @type lu: L{LogicalUnit}
6997
  @param lu: the logical unit for which we check
6998
  @type nodenames: list
6999
  @param nodenames: the list of nodes on which we should check
7000
  @type hvname: string
7001
  @param hvname: the name of the hypervisor we should use
7002
  @type hvparams: dict
7003
  @param hvparams: the parameters which we need to check
7004
  @raise errors.OpPrereqError: if the parameters are not valid
7005

7006
  """
7007
  nodenames = _FilterVmNodes(lu, nodenames)
7008
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
7009
                                                  hvname,
7010
                                                  hvparams)
7011
  for node in nodenames:
7012
    info = hvinfo[node]
7013
    if info.offline:
7014
      continue
7015
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
7016

    
7017

    
7018
def _CheckOSParams(lu, required, nodenames, osname, osparams):
7019
  """OS parameters validation.
7020

7021
  @type lu: L{LogicalUnit}
7022
  @param lu: the logical unit for which we check
7023
  @type required: boolean
7024
  @param required: whether the validation should fail if the OS is not
7025
      found
7026
  @type nodenames: list
7027
  @param nodenames: the list of nodes on which we should check
7028
  @type osname: string
7029
  @param osname: the name of the hypervisor we should use
7030
  @type osparams: dict
7031
  @param osparams: the parameters which we need to check
7032
  @raise errors.OpPrereqError: if the parameters are not valid
7033

7034
  """
7035
  nodenames = _FilterVmNodes(lu, nodenames)
7036
  result = lu.rpc.call_os_validate(required, nodenames, osname,
7037
                                   [constants.OS_VALIDATE_PARAMETERS],
7038
                                   osparams)
7039
  for node, nres in result.items():
7040
    # we don't check for offline cases since this should be run only
7041
    # against the master node and/or an instance's nodes
7042
    nres.Raise("OS Parameters validation failed on node %s" % node)
7043
    if not nres.payload:
7044
      lu.LogInfo("OS %s not found on node %s, validation skipped",
7045
                 osname, node)
7046

    
7047

    
7048
class LUInstanceCreate(LogicalUnit):
7049
  """Create an instance.
7050

7051
  """
7052
  HPATH = "instance-add"
7053
  HTYPE = constants.HTYPE_INSTANCE
7054
  REQ_BGL = False
7055

    
7056
  def CheckArguments(self):
7057
    """Check arguments.
7058

7059
    """
7060
    # do not require name_check to ease forward/backward compatibility
7061
    # for tools
7062
    if self.op.no_install and self.op.start:
7063
      self.LogInfo("No-installation mode selected, disabling startup")
7064
      self.op.start = False
7065
    # validate/normalize the instance name
7066
    self.op.instance_name = \
7067
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
7068

    
7069
    if self.op.ip_check and not self.op.name_check:
7070
      # TODO: make the ip check more flexible and not depend on the name check
7071
      raise errors.OpPrereqError("Cannot do ip check without a name check",
7072
                                 errors.ECODE_INVAL)
7073

    
7074
    # check nics' parameter names
7075
    for nic in self.op.nics:
7076
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
7077

    
7078
    # check disks. parameter names and consistent adopt/no-adopt strategy
7079
    has_adopt = has_no_adopt = False
7080
    for disk in self.op.disks:
7081
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
7082
      if "adopt" in disk:
7083
        has_adopt = True
7084
      else:
7085
        has_no_adopt = True
7086
    if has_adopt and has_no_adopt:
7087
      raise errors.OpPrereqError("Either all disks are adopted or none is",
7088
                                 errors.ECODE_INVAL)
7089
    if has_adopt:
7090
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
7091
        raise errors.OpPrereqError("Disk adoption is not supported for the"
7092
                                   " '%s' disk template" %
7093
                                   self.op.disk_template,
7094
                                   errors.ECODE_INVAL)
7095
      if self.op.iallocator is not None:
7096
        raise errors.OpPrereqError("Disk adoption not allowed with an"
7097
                                   " iallocator script", errors.ECODE_INVAL)
7098
      if self.op.mode == constants.INSTANCE_IMPORT:
7099
        raise errors.OpPrereqError("Disk adoption not allowed for"
7100
                                   " instance import", errors.ECODE_INVAL)
7101
    else:
7102
      if self.op.disk_template in constants.DTS_MUST_ADOPT:
7103
        raise errors.OpPrereqError("Disk template %s requires disk adoption,"
7104
                                   " but no 'adopt' parameter given" %
7105
                                   self.op.disk_template,
7106
                                   errors.ECODE_INVAL)
7107

    
7108
    self.adopt_disks = has_adopt
7109

    
7110
    # instance name verification
7111
    if self.op.name_check:
7112
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
7113
      self.op.instance_name = self.hostname1.name
7114
      # used in CheckPrereq for ip ping check
7115
      self.check_ip = self.hostname1.ip
7116
    else:
7117
      self.check_ip = None
7118

    
7119
    # file storage checks
7120
    if (self.op.file_driver and
7121
        not self.op.file_driver in constants.FILE_DRIVER):
7122
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
7123
                                 self.op.file_driver, errors.ECODE_INVAL)
7124

    
7125
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
7126
      raise errors.OpPrereqError("File storage directory path not absolute",
7127
                                 errors.ECODE_INVAL)
7128

    
7129
    ### Node/iallocator related checks
7130
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
7131

    
7132
    if self.op.pnode is not None:
7133
      if self.op.disk_template in constants.DTS_NET_MIRROR:
7134
        if self.op.snode is None:
7135
          raise errors.OpPrereqError("The networked disk templates need"
7136
                                     " a mirror node", errors.ECODE_INVAL)
7137
      elif self.op.snode:
7138
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
7139
                        " template")
7140
        self.op.snode = None
7141

    
7142
    self._cds = _GetClusterDomainSecret()
7143

    
7144
    if self.op.mode == constants.INSTANCE_IMPORT:
7145
      # On import force_variant must be True, because if we forced it at
7146
      # initial install, our only chance when importing it back is that it
7147
      # works again!
7148
      self.op.force_variant = True
7149

    
7150
      if self.op.no_install:
7151
        self.LogInfo("No-installation mode has no effect during import")
7152

    
7153
    elif self.op.mode == constants.INSTANCE_CREATE:
7154
      if self.op.os_type is None:
7155
        raise errors.OpPrereqError("No guest OS specified",
7156
                                   errors.ECODE_INVAL)
7157
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
7158
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
7159
                                   " installation" % self.op.os_type,
7160
                                   errors.ECODE_STATE)
7161
      if self.op.disk_template is None:
7162
        raise errors.OpPrereqError("No disk template specified",
7163
                                   errors.ECODE_INVAL)
7164

    
7165
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7166
      # Check handshake to ensure both clusters have the same domain secret
7167
      src_handshake = self.op.source_handshake
7168
      if not src_handshake:
7169
        raise errors.OpPrereqError("Missing source handshake",
7170
                                   errors.ECODE_INVAL)
7171

    
7172
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
7173
                                                           src_handshake)
7174
      if errmsg:
7175
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
7176
                                   errors.ECODE_INVAL)
7177

    
7178
      # Load and check source CA
7179
      self.source_x509_ca_pem = self.op.source_x509_ca
7180
      if not self.source_x509_ca_pem:
7181
        raise errors.OpPrereqError("Missing source X509 CA",
7182
                                   errors.ECODE_INVAL)
7183

    
7184
      try:
7185
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
7186
                                                    self._cds)
7187
      except OpenSSL.crypto.Error, err:
7188
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
7189
                                   (err, ), errors.ECODE_INVAL)
7190

    
7191
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
7192
      if errcode is not None:
7193
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
7194
                                   errors.ECODE_INVAL)
7195

    
7196
      self.source_x509_ca = cert
7197

    
7198
      src_instance_name = self.op.source_instance_name
7199
      if not src_instance_name:
7200
        raise errors.OpPrereqError("Missing source instance name",
7201
                                   errors.ECODE_INVAL)
7202

    
7203
      self.source_instance_name = \
7204
          netutils.GetHostname(name=src_instance_name).name
7205

    
7206
    else:
7207
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
7208
                                 self.op.mode, errors.ECODE_INVAL)
7209

    
7210
  def ExpandNames(self):
7211
    """ExpandNames for CreateInstance.
7212

7213
    Figure out the right locks for instance creation.
7214

7215
    """
7216
    self.needed_locks = {}
7217

    
7218
    instance_name = self.op.instance_name
7219
    # this is just a preventive check, but someone might still add this
7220
    # instance in the meantime, and creation will fail at lock-add time
7221
    if instance_name in self.cfg.GetInstanceList():
7222
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7223
                                 instance_name, errors.ECODE_EXISTS)
7224

    
7225
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
7226

    
7227
    if self.op.iallocator:
7228
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7229
    else:
7230
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
7231
      nodelist = [self.op.pnode]
7232
      if self.op.snode is not None:
7233
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
7234
        nodelist.append(self.op.snode)
7235
      self.needed_locks[locking.LEVEL_NODE] = nodelist
7236

    
7237
    # in case of import lock the source node too
7238
    if self.op.mode == constants.INSTANCE_IMPORT:
7239
      src_node = self.op.src_node
7240
      src_path = self.op.src_path
7241

    
7242
      if src_path is None:
7243
        self.op.src_path = src_path = self.op.instance_name
7244

    
7245
      if src_node is None:
7246
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7247
        self.op.src_node = None
7248
        if os.path.isabs(src_path):
7249
          raise errors.OpPrereqError("Importing an instance from an absolute"
7250
                                     " path requires a source node option.",
7251
                                     errors.ECODE_INVAL)
7252
      else:
7253
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
7254
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
7255
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
7256
        if not os.path.isabs(src_path):
7257
          self.op.src_path = src_path = \
7258
            utils.PathJoin(constants.EXPORT_DIR, src_path)
7259

    
7260
  def _RunAllocator(self):
7261
    """Run the allocator based on input opcode.
7262

7263
    """
7264
    nics = [n.ToDict() for n in self.nics]
7265
    ial = IAllocator(self.cfg, self.rpc,
7266
                     mode=constants.IALLOCATOR_MODE_ALLOC,
7267
                     name=self.op.instance_name,
7268
                     disk_template=self.op.disk_template,
7269
                     tags=[],
7270
                     os=self.op.os_type,
7271
                     vcpus=self.be_full[constants.BE_VCPUS],
7272
                     mem_size=self.be_full[constants.BE_MEMORY],
7273
                     disks=self.disks,
7274
                     nics=nics,
7275
                     hypervisor=self.op.hypervisor,
7276
                     )
7277

    
7278
    ial.Run(self.op.iallocator)
7279

    
7280
    if not ial.success:
7281
      raise errors.OpPrereqError("Can't compute nodes using"
7282
                                 " iallocator '%s': %s" %
7283
                                 (self.op.iallocator, ial.info),
7284
                                 errors.ECODE_NORES)
7285
    if len(ial.result) != ial.required_nodes:
7286
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
7287
                                 " of nodes (%s), required %s" %
7288
                                 (self.op.iallocator, len(ial.result),
7289
                                  ial.required_nodes), errors.ECODE_FAULT)
7290
    self.op.pnode = ial.result[0]
7291
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
7292
                 self.op.instance_name, self.op.iallocator,
7293
                 utils.CommaJoin(ial.result))
7294
    if ial.required_nodes == 2:
7295
      self.op.snode = ial.result[1]
7296

    
7297
  def BuildHooksEnv(self):
7298
    """Build hooks env.
7299

7300
    This runs on master, primary and secondary nodes of the instance.
7301

7302
    """
7303
    env = {
7304
      "ADD_MODE": self.op.mode,
7305
      }
7306
    if self.op.mode == constants.INSTANCE_IMPORT:
7307
      env["SRC_NODE"] = self.op.src_node
7308
      env["SRC_PATH"] = self.op.src_path
7309
      env["SRC_IMAGES"] = self.src_images
7310

    
7311
    env.update(_BuildInstanceHookEnv(
7312
      name=self.op.instance_name,
7313
      primary_node=self.op.pnode,
7314
      secondary_nodes=self.secondaries,
7315
      status=self.op.start,
7316
      os_type=self.op.os_type,
7317
      memory=self.be_full[constants.BE_MEMORY],
7318
      vcpus=self.be_full[constants.BE_VCPUS],
7319
      nics=_NICListToTuple(self, self.nics),
7320
      disk_template=self.op.disk_template,
7321
      disks=[(d["size"], d["mode"]) for d in self.disks],
7322
      bep=self.be_full,
7323
      hvp=self.hv_full,
7324
      hypervisor_name=self.op.hypervisor,
7325
    ))
7326

    
7327
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
7328
          self.secondaries)
7329
    return env, nl, nl
7330

    
7331
  def _ReadExportInfo(self):
7332
    """Reads the export information from disk.
7333

7334
    It will override the opcode source node and path with the actual
7335
    information, if these two were not specified before.
7336

7337
    @return: the export information
7338

7339
    """
7340
    assert self.op.mode == constants.INSTANCE_IMPORT
7341

    
7342
    src_node = self.op.src_node
7343
    src_path = self.op.src_path
7344

    
7345
    if src_node is None:
7346
      locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
7347
      exp_list = self.rpc.call_export_list(locked_nodes)
7348
      found = False
7349
      for node in exp_list:
7350
        if exp_list[node].fail_msg:
7351
          continue
7352
        if src_path in exp_list[node].payload:
7353
          found = True
7354
          self.op.src_node = src_node = node
7355
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
7356
                                                       src_path)
7357
          break
7358
      if not found:
7359
        raise errors.OpPrereqError("No export found for relative path %s" %
7360
                                    src_path, errors.ECODE_INVAL)
7361

    
7362
    _CheckNodeOnline(self, src_node)
7363
    result = self.rpc.call_export_info(src_node, src_path)
7364
    result.Raise("No export or invalid export found in dir %s" % src_path)
7365

    
7366
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
7367
    if not export_info.has_section(constants.INISECT_EXP):
7368
      raise errors.ProgrammerError("Corrupted export config",
7369
                                   errors.ECODE_ENVIRON)
7370

    
7371
    ei_version = export_info.get(constants.INISECT_EXP, "version")
7372
    if (int(ei_version) != constants.EXPORT_VERSION):
7373
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
7374
                                 (ei_version, constants.EXPORT_VERSION),
7375
                                 errors.ECODE_ENVIRON)
7376
    return export_info
7377

    
7378
  def _ReadExportParams(self, einfo):
7379
    """Use export parameters as defaults.
7380

7381
    In case the opcode doesn't specify (as in override) some instance
7382
    parameters, then try to use them from the export information, if
7383
    that declares them.
7384

7385
    """
7386
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
7387

    
7388
    if self.op.disk_template is None:
7389
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
7390
        self.op.disk_template = einfo.get(constants.INISECT_INS,
7391
                                          "disk_template")
7392
      else:
7393
        raise errors.OpPrereqError("No disk template specified and the export"
7394
                                   " is missing the disk_template information",
7395
                                   errors.ECODE_INVAL)
7396

    
7397
    if not self.op.disks:
7398
      if einfo.has_option(constants.INISECT_INS, "disk_count"):
7399
        disks = []
7400
        # TODO: import the disk iv_name too
7401
        for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
7402
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
7403
          disks.append({"size": disk_sz})
7404
        self.op.disks = disks
7405
      else:
7406
        raise errors.OpPrereqError("No disk info specified and the export"
7407
                                   " is missing the disk information",
7408
                                   errors.ECODE_INVAL)
7409

    
7410
    if (not self.op.nics and
7411
        einfo.has_option(constants.INISECT_INS, "nic_count")):
7412
      nics = []
7413
      for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
7414
        ndict = {}
7415
        for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
7416
          v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
7417
          ndict[name] = v
7418
        nics.append(ndict)
7419
      self.op.nics = nics
7420

    
7421
    if (self.op.hypervisor is None and
7422
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
7423
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
7424
    if einfo.has_section(constants.INISECT_HYP):
7425
      # use the export parameters but do not override the ones
7426
      # specified by the user
7427
      for name, value in einfo.items(constants.INISECT_HYP):
7428
        if name not in self.op.hvparams:
7429
          self.op.hvparams[name] = value
7430

    
7431
    if einfo.has_section(constants.INISECT_BEP):
7432
      # use the parameters, without overriding
7433
      for name, value in einfo.items(constants.INISECT_BEP):
7434
        if name not in self.op.beparams:
7435
          self.op.beparams[name] = value
7436
    else:
7437
      # try to read the parameters old style, from the main section
7438
      for name in constants.BES_PARAMETERS:
7439
        if (name not in self.op.beparams and
7440
            einfo.has_option(constants.INISECT_INS, name)):
7441
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
7442

    
7443
    if einfo.has_section(constants.INISECT_OSP):
7444
      # use the parameters, without overriding
7445
      for name, value in einfo.items(constants.INISECT_OSP):
7446
        if name not in self.op.osparams:
7447
          self.op.osparams[name] = value
7448

    
7449
  def _RevertToDefaults(self, cluster):
7450
    """Revert the instance parameters to the default values.
7451

7452
    """
7453
    # hvparams
7454
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
7455
    for name in self.op.hvparams.keys():
7456
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
7457
        del self.op.hvparams[name]
7458
    # beparams
7459
    be_defs = cluster.SimpleFillBE({})
7460
    for name in self.op.beparams.keys():
7461
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
7462
        del self.op.beparams[name]
7463
    # nic params
7464
    nic_defs = cluster.SimpleFillNIC({})
7465
    for nic in self.op.nics:
7466
      for name in constants.NICS_PARAMETERS:
7467
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
7468
          del nic[name]
7469
    # osparams
7470
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
7471
    for name in self.op.osparams.keys():
7472
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
7473
        del self.op.osparams[name]
7474

    
7475
  def CheckPrereq(self):
7476
    """Check prerequisites.
7477

7478
    """
7479
    if self.op.mode == constants.INSTANCE_IMPORT:
7480
      export_info = self._ReadExportInfo()
7481
      self._ReadExportParams(export_info)
7482

    
7483
    if (not self.cfg.GetVGName() and
7484
        self.op.disk_template not in constants.DTS_NOT_LVM):
7485
      raise errors.OpPrereqError("Cluster does not support lvm-based"
7486
                                 " instances", errors.ECODE_STATE)
7487

    
7488
    if self.op.hypervisor is None:
7489
      self.op.hypervisor = self.cfg.GetHypervisorType()
7490

    
7491
    cluster = self.cfg.GetClusterInfo()
7492
    enabled_hvs = cluster.enabled_hypervisors
7493
    if self.op.hypervisor not in enabled_hvs:
7494
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
7495
                                 " cluster (%s)" % (self.op.hypervisor,
7496
                                  ",".join(enabled_hvs)),
7497
                                 errors.ECODE_STATE)
7498

    
7499
    # check hypervisor parameter syntax (locally)
7500
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
7501
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
7502
                                      self.op.hvparams)
7503
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
7504
    hv_type.CheckParameterSyntax(filled_hvp)
7505
    self.hv_full = filled_hvp
7506
    # check that we don't specify global parameters on an instance
7507
    _CheckGlobalHvParams(self.op.hvparams)
7508

    
7509
    # fill and remember the beparams dict
7510
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
7511
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
7512

    
7513
    # build os parameters
7514
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
7515

    
7516
    # now that hvp/bep are in final format, let's reset to defaults,
7517
    # if told to do so
7518
    if self.op.identify_defaults:
7519
      self._RevertToDefaults(cluster)
7520

    
7521
    # NIC buildup
7522
    self.nics = []
7523
    for idx, nic in enumerate(self.op.nics):
7524
      nic_mode_req = nic.get("mode", None)
7525
      nic_mode = nic_mode_req
7526
      if nic_mode is None:
7527
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
7528

    
7529
      # in routed mode, for the first nic, the default ip is 'auto'
7530
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
7531
        default_ip_mode = constants.VALUE_AUTO
7532
      else:
7533
        default_ip_mode = constants.VALUE_NONE
7534

    
7535
      # ip validity checks
7536
      ip = nic.get("ip", default_ip_mode)
7537
      if ip is None or ip.lower() == constants.VALUE_NONE:
7538
        nic_ip = None
7539
      elif ip.lower() == constants.VALUE_AUTO:
7540
        if not self.op.name_check:
7541
          raise errors.OpPrereqError("IP address set to auto but name checks"
7542
                                     " have been skipped",
7543
                                     errors.ECODE_INVAL)
7544
        nic_ip = self.hostname1.ip
7545
      else:
7546
        if not netutils.IPAddress.IsValid(ip):
7547
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
7548
                                     errors.ECODE_INVAL)
7549
        nic_ip = ip
7550

    
7551
      # TODO: check the ip address for uniqueness
7552
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
7553
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
7554
                                   errors.ECODE_INVAL)
7555

    
7556
      # MAC address verification
7557
      mac = nic.get("mac", constants.VALUE_AUTO)
7558
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7559
        mac = utils.NormalizeAndValidateMac(mac)
7560

    
7561
        try:
7562
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
7563
        except errors.ReservationError:
7564
          raise errors.OpPrereqError("MAC address %s already in use"
7565
                                     " in cluster" % mac,
7566
                                     errors.ECODE_NOTUNIQUE)
7567

    
7568
      # bridge verification
7569
      bridge = nic.get("bridge", None)
7570
      link = nic.get("link", None)
7571
      if bridge and link:
7572
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
7573
                                   " at the same time", errors.ECODE_INVAL)
7574
      elif bridge and nic_mode == constants.NIC_MODE_ROUTED:
7575
        raise errors.OpPrereqError("Cannot pass 'bridge' on a routed nic",
7576
                                   errors.ECODE_INVAL)
7577
      elif bridge:
7578
        link = bridge
7579

    
7580
      nicparams = {}
7581
      if nic_mode_req:
7582
        nicparams[constants.NIC_MODE] = nic_mode_req
7583
      if link:
7584
        nicparams[constants.NIC_LINK] = link
7585

    
7586
      check_params = cluster.SimpleFillNIC(nicparams)
7587
      objects.NIC.CheckParameterSyntax(check_params)
7588
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
7589

    
7590
    # disk checks/pre-build
7591
    self.disks = []
7592
    for disk in self.op.disks:
7593
      mode = disk.get("mode", constants.DISK_RDWR)
7594
      if mode not in constants.DISK_ACCESS_SET:
7595
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
7596
                                   mode, errors.ECODE_INVAL)
7597
      size = disk.get("size", None)
7598
      if size is None:
7599
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
7600
      try:
7601
        size = int(size)
7602
      except (TypeError, ValueError):
7603
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
7604
                                   errors.ECODE_INVAL)
7605
      vg = disk.get("vg", self.cfg.GetVGName())
7606
      new_disk = {"size": size, "mode": mode, "vg": vg}
7607
      if "adopt" in disk:
7608
        new_disk["adopt"] = disk["adopt"]
7609
      self.disks.append(new_disk)
7610

    
7611
    if self.op.mode == constants.INSTANCE_IMPORT:
7612

    
7613
      # Check that the new instance doesn't have less disks than the export
7614
      instance_disks = len(self.disks)
7615
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
7616
      if instance_disks < export_disks:
7617
        raise errors.OpPrereqError("Not enough disks to import."
7618
                                   " (instance: %d, export: %d)" %
7619
                                   (instance_disks, export_disks),
7620
                                   errors.ECODE_INVAL)
7621

    
7622
      disk_images = []
7623
      for idx in range(export_disks):
7624
        option = 'disk%d_dump' % idx
7625
        if export_info.has_option(constants.INISECT_INS, option):
7626
          # FIXME: are the old os-es, disk sizes, etc. useful?
7627
          export_name = export_info.get(constants.INISECT_INS, option)
7628
          image = utils.PathJoin(self.op.src_path, export_name)
7629
          disk_images.append(image)
7630
        else:
7631
          disk_images.append(False)
7632

    
7633
      self.src_images = disk_images
7634

    
7635
      old_name = export_info.get(constants.INISECT_INS, 'name')
7636
      try:
7637
        exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
7638
      except (TypeError, ValueError), err:
7639
        raise errors.OpPrereqError("Invalid export file, nic_count is not"
7640
                                   " an integer: %s" % str(err),
7641
                                   errors.ECODE_STATE)
7642
      if self.op.instance_name == old_name:
7643
        for idx, nic in enumerate(self.nics):
7644
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
7645
            nic_mac_ini = 'nic%d_mac' % idx
7646
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
7647

    
7648
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
7649

    
7650
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
7651
    if self.op.ip_check:
7652
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
7653
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
7654
                                   (self.check_ip, self.op.instance_name),
7655
                                   errors.ECODE_NOTUNIQUE)
7656

    
7657
    #### mac address generation
7658
    # By generating here the mac address both the allocator and the hooks get
7659
    # the real final mac address rather than the 'auto' or 'generate' value.
7660
    # There is a race condition between the generation and the instance object
7661
    # creation, which means that we know the mac is valid now, but we're not
7662
    # sure it will be when we actually add the instance. If things go bad
7663
    # adding the instance will abort because of a duplicate mac, and the
7664
    # creation job will fail.
7665
    for nic in self.nics:
7666
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7667
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
7668

    
7669
    #### allocator run
7670

    
7671
    if self.op.iallocator is not None:
7672
      self._RunAllocator()
7673

    
7674
    #### node related checks
7675

    
7676
    # check primary node
7677
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
7678
    assert self.pnode is not None, \
7679
      "Cannot retrieve locked node %s" % self.op.pnode
7680
    if pnode.offline:
7681
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
7682
                                 pnode.name, errors.ECODE_STATE)
7683
    if pnode.drained:
7684
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
7685
                                 pnode.name, errors.ECODE_STATE)
7686
    if not pnode.vm_capable:
7687
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
7688
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
7689

    
7690
    self.secondaries = []
7691

    
7692
    # mirror node verification
7693
    if self.op.disk_template in constants.DTS_NET_MIRROR:
7694
      if self.op.snode == pnode.name:
7695
        raise errors.OpPrereqError("The secondary node cannot be the"
7696
                                   " primary node.", errors.ECODE_INVAL)
7697
      _CheckNodeOnline(self, self.op.snode)
7698
      _CheckNodeNotDrained(self, self.op.snode)
7699
      _CheckNodeVmCapable(self, self.op.snode)
7700
      self.secondaries.append(self.op.snode)
7701

    
7702
    nodenames = [pnode.name] + self.secondaries
7703

    
7704
    if not self.adopt_disks:
7705
      # Check lv size requirements, if not adopting
7706
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
7707
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
7708

    
7709
    elif self.op.disk_template == constants.DT_PLAIN: # Check the adoption data
7710
      all_lvs = set([i["vg"] + "/" + i["adopt"] for i in self.disks])
7711
      if len(all_lvs) != len(self.disks):
7712
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
7713
                                   errors.ECODE_INVAL)
7714
      for lv_name in all_lvs:
7715
        try:
7716
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
7717
          # to ReserveLV uses the same syntax
7718
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
7719
        except errors.ReservationError:
7720
          raise errors.OpPrereqError("LV named %s used by another instance" %
7721
                                     lv_name, errors.ECODE_NOTUNIQUE)
7722

    
7723
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
7724
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
7725

    
7726
      node_lvs = self.rpc.call_lv_list([pnode.name],
7727
                                       vg_names.payload.keys())[pnode.name]
7728
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
7729
      node_lvs = node_lvs.payload
7730

    
7731
      delta = all_lvs.difference(node_lvs.keys())
7732
      if delta:
7733
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
7734
                                   utils.CommaJoin(delta),
7735
                                   errors.ECODE_INVAL)
7736
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
7737
      if online_lvs:
7738
        raise errors.OpPrereqError("Online logical volumes found, cannot"
7739
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
7740
                                   errors.ECODE_STATE)
7741
      # update the size of disk based on what is found
7742
      for dsk in self.disks:
7743
        dsk["size"] = int(float(node_lvs[dsk["vg"] + "/" + dsk["adopt"]][0]))
7744

    
7745
    elif self.op.disk_template == constants.DT_BLOCK:
7746
      # Normalize and de-duplicate device paths
7747
      all_disks = set([os.path.abspath(i["adopt"]) for i in self.disks])
7748
      if len(all_disks) != len(self.disks):
7749
        raise errors.OpPrereqError("Duplicate disk names given for adoption",
7750
                                   errors.ECODE_INVAL)
7751
      baddisks = [d for d in all_disks
7752
                  if not d.startswith(constants.ADOPTABLE_BLOCKDEV_ROOT)]
7753
      if baddisks:
7754
        raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
7755
                                   " cannot be adopted" %
7756
                                   (", ".join(baddisks),
7757
                                    constants.ADOPTABLE_BLOCKDEV_ROOT),
7758
                                   errors.ECODE_INVAL)
7759

    
7760
      node_disks = self.rpc.call_bdev_sizes([pnode.name],
7761
                                            list(all_disks))[pnode.name]
7762
      node_disks.Raise("Cannot get block device information from node %s" %
7763
                       pnode.name)
7764
      node_disks = node_disks.payload
7765
      delta = all_disks.difference(node_disks.keys())
7766
      if delta:
7767
        raise errors.OpPrereqError("Missing block device(s): %s" %
7768
                                   utils.CommaJoin(delta),
7769
                                   errors.ECODE_INVAL)
7770
      for dsk in self.disks:
7771
        dsk["size"] = int(float(node_disks[dsk["adopt"]]))
7772

    
7773
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
7774

    
7775
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
7776
    # check OS parameters (remotely)
7777
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
7778

    
7779
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
7780

    
7781
    # memory check on primary node
7782
    if self.op.start:
7783
      _CheckNodeFreeMemory(self, self.pnode.name,
7784
                           "creating instance %s" % self.op.instance_name,
7785
                           self.be_full[constants.BE_MEMORY],
7786
                           self.op.hypervisor)
7787

    
7788
    self.dry_run_result = list(nodenames)
7789

    
7790
  def Exec(self, feedback_fn):
7791
    """Create and add the instance to the cluster.
7792

7793
    """
7794
    instance = self.op.instance_name
7795
    pnode_name = self.pnode.name
7796

    
7797
    ht_kind = self.op.hypervisor
7798
    if ht_kind in constants.HTS_REQ_PORT:
7799
      network_port = self.cfg.AllocatePort()
7800
    else:
7801
      network_port = None
7802

    
7803
    if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE:
7804
      # this is needed because os.path.join does not accept None arguments
7805
      if self.op.file_storage_dir is None:
7806
        string_file_storage_dir = ""
7807
      else:
7808
        string_file_storage_dir = self.op.file_storage_dir
7809

    
7810
      # build the full file storage dir path
7811
      if self.op.disk_template == constants.DT_SHARED_FILE:
7812
        get_fsd_fn = self.cfg.GetSharedFileStorageDir
7813
      else:
7814
        get_fsd_fn = self.cfg.GetFileStorageDir
7815

    
7816
      file_storage_dir = utils.PathJoin(get_fsd_fn(),
7817
                                        string_file_storage_dir, instance)
7818
    else:
7819
      file_storage_dir = ""
7820

    
7821
    disks = _GenerateDiskTemplate(self,
7822
                                  self.op.disk_template,
7823
                                  instance, pnode_name,
7824
                                  self.secondaries,
7825
                                  self.disks,
7826
                                  file_storage_dir,
7827
                                  self.op.file_driver,
7828
                                  0,
7829
                                  feedback_fn)
7830

    
7831
    iobj = objects.Instance(name=instance, os=self.op.os_type,
7832
                            primary_node=pnode_name,
7833
                            nics=self.nics, disks=disks,
7834
                            disk_template=self.op.disk_template,
7835
                            admin_up=False,
7836
                            network_port=network_port,
7837
                            beparams=self.op.beparams,
7838
                            hvparams=self.op.hvparams,
7839
                            hypervisor=self.op.hypervisor,
7840
                            osparams=self.op.osparams,
7841
                            )
7842

    
7843
    if self.adopt_disks:
7844
      if self.op.disk_template == constants.DT_PLAIN:
7845
        # rename LVs to the newly-generated names; we need to construct
7846
        # 'fake' LV disks with the old data, plus the new unique_id
7847
        tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
7848
        rename_to = []
7849
        for t_dsk, a_dsk in zip (tmp_disks, self.disks):
7850
          rename_to.append(t_dsk.logical_id)
7851
          t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk["adopt"])
7852
          self.cfg.SetDiskID(t_dsk, pnode_name)
7853
        result = self.rpc.call_blockdev_rename(pnode_name,
7854
                                               zip(tmp_disks, rename_to))
7855
        result.Raise("Failed to rename adoped LVs")
7856
    else:
7857
      feedback_fn("* creating instance disks...")
7858
      try:
7859
        _CreateDisks(self, iobj)
7860
      except errors.OpExecError:
7861
        self.LogWarning("Device creation failed, reverting...")
7862
        try:
7863
          _RemoveDisks(self, iobj)
7864
        finally:
7865
          self.cfg.ReleaseDRBDMinors(instance)
7866
          raise
7867

    
7868
      if self.cfg.GetClusterInfo().prealloc_wipe_disks:
7869
        feedback_fn("* wiping instance disks...")
7870
        try:
7871
          _WipeDisks(self, iobj)
7872
        except errors.OpExecError:
7873
          self.LogWarning("Device wiping failed, reverting...")
7874
          try:
7875
            _RemoveDisks(self, iobj)
7876
          finally:
7877
            self.cfg.ReleaseDRBDMinors(instance)
7878
            raise
7879

    
7880
    feedback_fn("adding instance %s to cluster config" % instance)
7881

    
7882
    self.cfg.AddInstance(iobj, self.proc.GetECId())
7883

    
7884
    # Declare that we don't want to remove the instance lock anymore, as we've
7885
    # added the instance to the config
7886
    del self.remove_locks[locking.LEVEL_INSTANCE]
7887
    # Unlock all the nodes
7888
    if self.op.mode == constants.INSTANCE_IMPORT:
7889
      nodes_keep = [self.op.src_node]
7890
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
7891
                       if node != self.op.src_node]
7892
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
7893
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
7894
    else:
7895
      self.context.glm.release(locking.LEVEL_NODE)
7896
      del self.acquired_locks[locking.LEVEL_NODE]
7897

    
7898
    if self.op.wait_for_sync:
7899
      disk_abort = not _WaitForSync(self, iobj)
7900
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
7901
      # make sure the disks are not degraded (still sync-ing is ok)
7902
      time.sleep(15)
7903
      feedback_fn("* checking mirrors status")
7904
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
7905
    else:
7906
      disk_abort = False
7907

    
7908
    if disk_abort:
7909
      _RemoveDisks(self, iobj)
7910
      self.cfg.RemoveInstance(iobj.name)
7911
      # Make sure the instance lock gets removed
7912
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
7913
      raise errors.OpExecError("There are some degraded disks for"
7914
                               " this instance")
7915

    
7916
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
7917
      if self.op.mode == constants.INSTANCE_CREATE:
7918
        if not self.op.no_install:
7919
          feedback_fn("* running the instance OS create scripts...")
7920
          # FIXME: pass debug option from opcode to backend
7921
          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
7922
                                                 self.op.debug_level)
7923
          result.Raise("Could not add os for instance %s"
7924
                       " on node %s" % (instance, pnode_name))
7925

    
7926
      elif self.op.mode == constants.INSTANCE_IMPORT:
7927
        feedback_fn("* running the instance OS import scripts...")
7928

    
7929
        transfers = []
7930

    
7931
        for idx, image in enumerate(self.src_images):
7932
          if not image:
7933
            continue
7934

    
7935
          # FIXME: pass debug option from opcode to backend
7936
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
7937
                                             constants.IEIO_FILE, (image, ),
7938
                                             constants.IEIO_SCRIPT,
7939
                                             (iobj.disks[idx], idx),
7940
                                             None)
7941
          transfers.append(dt)
7942

    
7943
        import_result = \
7944
          masterd.instance.TransferInstanceData(self, feedback_fn,
7945
                                                self.op.src_node, pnode_name,
7946
                                                self.pnode.secondary_ip,
7947
                                                iobj, transfers)
7948
        if not compat.all(import_result):
7949
          self.LogWarning("Some disks for instance %s on node %s were not"
7950
                          " imported successfully" % (instance, pnode_name))
7951

    
7952
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7953
        feedback_fn("* preparing remote import...")
7954
        # The source cluster will stop the instance before attempting to make a
7955
        # connection. In some cases stopping an instance can take a long time,
7956
        # hence the shutdown timeout is added to the connection timeout.
7957
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
7958
                           self.op.source_shutdown_timeout)
7959
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
7960

    
7961
        assert iobj.primary_node == self.pnode.name
7962
        disk_results = \
7963
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
7964
                                        self.source_x509_ca,
7965
                                        self._cds, timeouts)
7966
        if not compat.all(disk_results):
7967
          # TODO: Should the instance still be started, even if some disks
7968
          # failed to import (valid for local imports, too)?
7969
          self.LogWarning("Some disks for instance %s on node %s were not"
7970
                          " imported successfully" % (instance, pnode_name))
7971

    
7972
        # Run rename script on newly imported instance
7973
        assert iobj.name == instance
7974
        feedback_fn("Running rename script for %s" % instance)
7975
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
7976
                                                   self.source_instance_name,
7977
                                                   self.op.debug_level)
7978
        if result.fail_msg:
7979
          self.LogWarning("Failed to run rename script for %s on node"
7980
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
7981

    
7982
      else:
7983
        # also checked in the prereq part
7984
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
7985
                                     % self.op.mode)
7986

    
7987
    if self.op.start:
7988
      iobj.admin_up = True
7989
      self.cfg.Update(iobj, feedback_fn)
7990
      logging.info("Starting instance %s on node %s", instance, pnode_name)
7991
      feedback_fn("* starting instance...")
7992
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
7993
      result.Raise("Could not start instance")
7994

    
7995
    return list(iobj.all_nodes)
7996

    
7997

    
7998
class LUInstanceConsole(NoHooksLU):
7999
  """Connect to an instance's console.
8000

8001
  This is somewhat special in that it returns the command line that
8002
  you need to run on the master node in order to connect to the
8003
  console.
8004

8005
  """
8006
  REQ_BGL = False
8007

    
8008
  def ExpandNames(self):
8009
    self._ExpandAndLockInstance()
8010

    
8011
  def CheckPrereq(self):
8012
    """Check prerequisites.
8013

8014
    This checks that the instance is in the cluster.
8015

8016
    """
8017
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8018
    assert self.instance is not None, \
8019
      "Cannot retrieve locked instance %s" % self.op.instance_name
8020
    _CheckNodeOnline(self, self.instance.primary_node)
8021

    
8022
  def Exec(self, feedback_fn):
8023
    """Connect to the console of an instance
8024

8025
    """
8026
    instance = self.instance
8027
    node = instance.primary_node
8028

    
8029
    node_insts = self.rpc.call_instance_list([node],
8030
                                             [instance.hypervisor])[node]
8031
    node_insts.Raise("Can't get node information from %s" % node)
8032

    
8033
    if instance.name not in node_insts.payload:
8034
      if instance.admin_up:
8035
        state = "ERROR_down"
8036
      else:
8037
        state = "ADMIN_down"
8038
      raise errors.OpExecError("Instance %s is not running (state %s)" %
8039
                               (instance.name, state))
8040

    
8041
    logging.debug("Connecting to console of %s on %s", instance.name, node)
8042

    
8043
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
8044

    
8045

    
8046
def _GetInstanceConsole(cluster, instance):
8047
  """Returns console information for an instance.
8048

8049
  @type cluster: L{objects.Cluster}
8050
  @type instance: L{objects.Instance}
8051
  @rtype: dict
8052

8053
  """
8054
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
8055
  # beparams and hvparams are passed separately, to avoid editing the
8056
  # instance and then saving the defaults in the instance itself.
8057
  hvparams = cluster.FillHV(instance)
8058
  beparams = cluster.FillBE(instance)
8059
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
8060

    
8061
  assert console.instance == instance.name
8062
  assert console.Validate()
8063

    
8064
  return console.ToDict()
8065

    
8066

    
8067
class LUInstanceReplaceDisks(LogicalUnit):
8068
  """Replace the disks of an instance.
8069

8070
  """
8071
  HPATH = "mirrors-replace"
8072
  HTYPE = constants.HTYPE_INSTANCE
8073
  REQ_BGL = False
8074

    
8075
  def CheckArguments(self):
8076
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
8077
                                  self.op.iallocator)
8078

    
8079
  def ExpandNames(self):
8080
    self._ExpandAndLockInstance()
8081

    
8082
    if self.op.iallocator is not None:
8083
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8084

    
8085
    elif self.op.remote_node is not None:
8086
      remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8087
      self.op.remote_node = remote_node
8088

    
8089
      # Warning: do not remove the locking of the new secondary here
8090
      # unless DRBD8.AddChildren is changed to work in parallel;
8091
      # currently it doesn't since parallel invocations of
8092
      # FindUnusedMinor will conflict
8093
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
8094
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
8095

    
8096
    else:
8097
      self.needed_locks[locking.LEVEL_NODE] = []
8098
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8099

    
8100
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
8101
                                   self.op.iallocator, self.op.remote_node,
8102
                                   self.op.disks, False, self.op.early_release)
8103

    
8104
    self.tasklets = [self.replacer]
8105

    
8106
  def DeclareLocks(self, level):
8107
    # If we're not already locking all nodes in the set we have to declare the
8108
    # instance's primary/secondary nodes.
8109
    if (level == locking.LEVEL_NODE and
8110
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
8111
      self._LockInstancesNodes()
8112

    
8113
  def BuildHooksEnv(self):
8114
    """Build hooks env.
8115

8116
    This runs on the master, the primary and all the secondaries.
8117

8118
    """
8119
    instance = self.replacer.instance
8120
    env = {
8121
      "MODE": self.op.mode,
8122
      "NEW_SECONDARY": self.op.remote_node,
8123
      "OLD_SECONDARY": instance.secondary_nodes[0],
8124
      }
8125
    env.update(_BuildInstanceHookEnvByObject(self, instance))
8126
    nl = [
8127
      self.cfg.GetMasterNode(),
8128
      instance.primary_node,
8129
      ]
8130
    if self.op.remote_node is not None:
8131
      nl.append(self.op.remote_node)
8132
    return env, nl, nl
8133

    
8134

    
8135
class TLReplaceDisks(Tasklet):
8136
  """Replaces disks for an instance.
8137

8138
  Note: Locking is not within the scope of this class.
8139

8140
  """
8141
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
8142
               disks, delay_iallocator, early_release):
8143
    """Initializes this class.
8144

8145
    """
8146
    Tasklet.__init__(self, lu)
8147

    
8148
    # Parameters
8149
    self.instance_name = instance_name
8150
    self.mode = mode
8151
    self.iallocator_name = iallocator_name
8152
    self.remote_node = remote_node
8153
    self.disks = disks
8154
    self.delay_iallocator = delay_iallocator
8155
    self.early_release = early_release
8156

    
8157
    # Runtime data
8158
    self.instance = None
8159
    self.new_node = None
8160
    self.target_node = None
8161
    self.other_node = None
8162
    self.remote_node_info = None
8163
    self.node_secondary_ip = None
8164

    
8165
  @staticmethod
8166
  def CheckArguments(mode, remote_node, iallocator):
8167
    """Helper function for users of this class.
8168

8169
    """
8170
    # check for valid parameter combination
8171
    if mode == constants.REPLACE_DISK_CHG:
8172
      if remote_node is None and iallocator is None:
8173
        raise errors.OpPrereqError("When changing the secondary either an"
8174
                                   " iallocator script must be used or the"
8175
                                   " new node given", errors.ECODE_INVAL)
8176

    
8177
      if remote_node is not None and iallocator is not None:
8178
        raise errors.OpPrereqError("Give either the iallocator or the new"
8179
                                   " secondary, not both", errors.ECODE_INVAL)
8180

    
8181
    elif remote_node is not None or iallocator is not None:
8182
      # Not replacing the secondary
8183
      raise errors.OpPrereqError("The iallocator and new node options can"
8184
                                 " only be used when changing the"
8185
                                 " secondary node", errors.ECODE_INVAL)
8186

    
8187
  @staticmethod
8188
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
8189
    """Compute a new secondary node using an IAllocator.
8190

8191
    """
8192
    ial = IAllocator(lu.cfg, lu.rpc,
8193
                     mode=constants.IALLOCATOR_MODE_RELOC,
8194
                     name=instance_name,
8195
                     relocate_from=relocate_from)
8196

    
8197
    ial.Run(iallocator_name)
8198

    
8199
    if not ial.success:
8200
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
8201
                                 " %s" % (iallocator_name, ial.info),
8202
                                 errors.ECODE_NORES)
8203

    
8204
    if len(ial.result) != ial.required_nodes:
8205
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8206
                                 " of nodes (%s), required %s" %
8207
                                 (iallocator_name,
8208
                                  len(ial.result), ial.required_nodes),
8209
                                 errors.ECODE_FAULT)
8210

    
8211
    remote_node_name = ial.result[0]
8212

    
8213
    lu.LogInfo("Selected new secondary for instance '%s': %s",
8214
               instance_name, remote_node_name)
8215

    
8216
    return remote_node_name
8217

    
8218
  def _FindFaultyDisks(self, node_name):
8219
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
8220
                                    node_name, True)
8221

    
8222
  def CheckPrereq(self):
8223
    """Check prerequisites.
8224

8225
    This checks that the instance is in the cluster.
8226

8227
    """
8228
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
8229
    assert instance is not None, \
8230
      "Cannot retrieve locked instance %s" % self.instance_name
8231

    
8232
    if instance.disk_template != constants.DT_DRBD8:
8233
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
8234
                                 " instances", errors.ECODE_INVAL)
8235

    
8236
    if len(instance.secondary_nodes) != 1:
8237
      raise errors.OpPrereqError("The instance has a strange layout,"
8238
                                 " expected one secondary but found %d" %
8239
                                 len(instance.secondary_nodes),
8240
                                 errors.ECODE_FAULT)
8241

    
8242
    if not self.delay_iallocator:
8243
      self._CheckPrereq2()
8244

    
8245
  def _CheckPrereq2(self):
8246
    """Check prerequisites, second part.
8247

8248
    This function should always be part of CheckPrereq. It was separated and is
8249
    now called from Exec because during node evacuation iallocator was only
8250
    called with an unmodified cluster model, not taking planned changes into
8251
    account.
8252

8253
    """
8254
    instance = self.instance
8255
    secondary_node = instance.secondary_nodes[0]
8256

    
8257
    if self.iallocator_name is None:
8258
      remote_node = self.remote_node
8259
    else:
8260
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
8261
                                       instance.name, instance.secondary_nodes)
8262

    
8263
    if remote_node is not None:
8264
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
8265
      assert self.remote_node_info is not None, \
8266
        "Cannot retrieve locked node %s" % remote_node
8267
    else:
8268
      self.remote_node_info = None
8269

    
8270
    if remote_node == self.instance.primary_node:
8271
      raise errors.OpPrereqError("The specified node is the primary node of"
8272
                                 " the instance.", errors.ECODE_INVAL)
8273

    
8274
    if remote_node == secondary_node:
8275
      raise errors.OpPrereqError("The specified node is already the"
8276
                                 " secondary node of the instance.",
8277
                                 errors.ECODE_INVAL)
8278

    
8279
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
8280
                                    constants.REPLACE_DISK_CHG):
8281
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
8282
                                 errors.ECODE_INVAL)
8283

    
8284
    if self.mode == constants.REPLACE_DISK_AUTO:
8285
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
8286
      faulty_secondary = self._FindFaultyDisks(secondary_node)
8287

    
8288
      if faulty_primary and faulty_secondary:
8289
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
8290
                                   " one node and can not be repaired"
8291
                                   " automatically" % self.instance_name,
8292
                                   errors.ECODE_STATE)
8293

    
8294
      if faulty_primary:
8295
        self.disks = faulty_primary
8296
        self.target_node = instance.primary_node
8297
        self.other_node = secondary_node
8298
        check_nodes = [self.target_node, self.other_node]
8299
      elif faulty_secondary:
8300
        self.disks = faulty_secondary
8301
        self.target_node = secondary_node
8302
        self.other_node = instance.primary_node
8303
        check_nodes = [self.target_node, self.other_node]
8304
      else:
8305
        self.disks = []
8306
        check_nodes = []
8307

    
8308
    else:
8309
      # Non-automatic modes
8310
      if self.mode == constants.REPLACE_DISK_PRI:
8311
        self.target_node = instance.primary_node
8312
        self.other_node = secondary_node
8313
        check_nodes = [self.target_node, self.other_node]
8314

    
8315
      elif self.mode == constants.REPLACE_DISK_SEC:
8316
        self.target_node = secondary_node
8317
        self.other_node = instance.primary_node
8318
        check_nodes = [self.target_node, self.other_node]
8319

    
8320
      elif self.mode == constants.REPLACE_DISK_CHG:
8321
        self.new_node = remote_node
8322
        self.other_node = instance.primary_node
8323
        self.target_node = secondary_node
8324
        check_nodes = [self.new_node, self.other_node]
8325

    
8326
        _CheckNodeNotDrained(self.lu, remote_node)
8327
        _CheckNodeVmCapable(self.lu, remote_node)
8328

    
8329
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
8330
        assert old_node_info is not None
8331
        if old_node_info.offline and not self.early_release:
8332
          # doesn't make sense to delay the release
8333
          self.early_release = True
8334
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
8335
                          " early-release mode", secondary_node)
8336

    
8337
      else:
8338
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
8339
                                     self.mode)
8340

    
8341
      # If not specified all disks should be replaced
8342
      if not self.disks:
8343
        self.disks = range(len(self.instance.disks))
8344

    
8345
    for node in check_nodes:
8346
      _CheckNodeOnline(self.lu, node)
8347

    
8348
    # Check whether disks are valid
8349
    for disk_idx in self.disks:
8350
      instance.FindDisk(disk_idx)
8351

    
8352
    # Get secondary node IP addresses
8353
    node_2nd_ip = {}
8354

    
8355
    for node_name in [self.target_node, self.other_node, self.new_node]:
8356
      if node_name is not None:
8357
        node_2nd_ip[node_name] = self.cfg.GetNodeInfo(node_name).secondary_ip
8358

    
8359
    self.node_secondary_ip = node_2nd_ip
8360

    
8361
  def Exec(self, feedback_fn):
8362
    """Execute disk replacement.
8363

8364
    This dispatches the disk replacement to the appropriate handler.
8365

8366
    """
8367
    if self.delay_iallocator:
8368
      self._CheckPrereq2()
8369

    
8370
    if not self.disks:
8371
      feedback_fn("No disks need replacement")
8372
      return
8373

    
8374
    feedback_fn("Replacing disk(s) %s for %s" %
8375
                (utils.CommaJoin(self.disks), self.instance.name))
8376

    
8377
    activate_disks = (not self.instance.admin_up)
8378

    
8379
    # Activate the instance disks if we're replacing them on a down instance
8380
    if activate_disks:
8381
      _StartInstanceDisks(self.lu, self.instance, True)
8382

    
8383
    try:
8384
      # Should we replace the secondary node?
8385
      if self.new_node is not None:
8386
        fn = self._ExecDrbd8Secondary
8387
      else:
8388
        fn = self._ExecDrbd8DiskOnly
8389

    
8390
      return fn(feedback_fn)
8391

    
8392
    finally:
8393
      # Deactivate the instance disks if we're replacing them on a
8394
      # down instance
8395
      if activate_disks:
8396
        _SafeShutdownInstanceDisks(self.lu, self.instance)
8397

    
8398
  def _CheckVolumeGroup(self, nodes):
8399
    self.lu.LogInfo("Checking volume groups")
8400

    
8401
    vgname = self.cfg.GetVGName()
8402

    
8403
    # Make sure volume group exists on all involved nodes
8404
    results = self.rpc.call_vg_list(nodes)
8405
    if not results:
8406
      raise errors.OpExecError("Can't list volume groups on the nodes")
8407

    
8408
    for node in nodes:
8409
      res = results[node]
8410
      res.Raise("Error checking node %s" % node)
8411
      if vgname not in res.payload:
8412
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
8413
                                 (vgname, node))
8414

    
8415
  def _CheckDisksExistence(self, nodes):
8416
    # Check disk existence
8417
    for idx, dev in enumerate(self.instance.disks):
8418
      if idx not in self.disks:
8419
        continue
8420

    
8421
      for node in nodes:
8422
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
8423
        self.cfg.SetDiskID(dev, node)
8424

    
8425
        result = self.rpc.call_blockdev_find(node, dev)
8426

    
8427
        msg = result.fail_msg
8428
        if msg or not result.payload:
8429
          if not msg:
8430
            msg = "disk not found"
8431
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
8432
                                   (idx, node, msg))
8433

    
8434
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
8435
    for idx, dev in enumerate(self.instance.disks):
8436
      if idx not in self.disks:
8437
        continue
8438

    
8439
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
8440
                      (idx, node_name))
8441

    
8442
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
8443
                                   ldisk=ldisk):
8444
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
8445
                                 " replace disks for instance %s" %
8446
                                 (node_name, self.instance.name))
8447

    
8448
  def _CreateNewStorage(self, node_name):
8449
    vgname = self.cfg.GetVGName()
8450
    iv_names = {}
8451

    
8452
    for idx, dev in enumerate(self.instance.disks):
8453
      if idx not in self.disks:
8454
        continue
8455

    
8456
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
8457

    
8458
      self.cfg.SetDiskID(dev, node_name)
8459

    
8460
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
8461
      names = _GenerateUniqueNames(self.lu, lv_names)
8462

    
8463
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
8464
                             logical_id=(vgname, names[0]))
8465
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
8466
                             logical_id=(vgname, names[1]))
8467

    
8468
      new_lvs = [lv_data, lv_meta]
8469
      old_lvs = dev.children
8470
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
8471

    
8472
      # we pass force_create=True to force the LVM creation
8473
      for new_lv in new_lvs:
8474
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
8475
                        _GetInstanceInfoText(self.instance), False)
8476

    
8477
    return iv_names
8478

    
8479
  def _CheckDevices(self, node_name, iv_names):
8480
    for name, (dev, _, _) in iv_names.iteritems():
8481
      self.cfg.SetDiskID(dev, node_name)
8482

    
8483
      result = self.rpc.call_blockdev_find(node_name, dev)
8484

    
8485
      msg = result.fail_msg
8486
      if msg or not result.payload:
8487
        if not msg:
8488
          msg = "disk not found"
8489
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
8490
                                 (name, msg))
8491

    
8492
      if result.payload.is_degraded:
8493
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
8494

    
8495
  def _RemoveOldStorage(self, node_name, iv_names):
8496
    for name, (_, old_lvs, _) in iv_names.iteritems():
8497
      self.lu.LogInfo("Remove logical volumes for %s" % name)
8498

    
8499
      for lv in old_lvs:
8500
        self.cfg.SetDiskID(lv, node_name)
8501

    
8502
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
8503
        if msg:
8504
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
8505
                             hint="remove unused LVs manually")
8506

    
8507
  def _ReleaseNodeLock(self, node_name):
8508
    """Releases the lock for a given node."""
8509
    self.lu.context.glm.release(locking.LEVEL_NODE, node_name)
8510

    
8511
  def _ExecDrbd8DiskOnly(self, feedback_fn):
8512
    """Replace a disk on the primary or secondary for DRBD 8.
8513

8514
    The algorithm for replace is quite complicated:
8515

8516
      1. for each disk to be replaced:
8517

8518
        1. create new LVs on the target node with unique names
8519
        1. detach old LVs from the drbd device
8520
        1. rename old LVs to name_replaced.<time_t>
8521
        1. rename new LVs to old LVs
8522
        1. attach the new LVs (with the old names now) to the drbd device
8523

8524
      1. wait for sync across all devices
8525

8526
      1. for each modified disk:
8527

8528
        1. remove old LVs (which have the name name_replaces.<time_t>)
8529

8530
    Failures are not very well handled.
8531

8532
    """
8533
    steps_total = 6
8534

    
8535
    # Step: check device activation
8536
    self.lu.LogStep(1, steps_total, "Check device existence")
8537
    self._CheckDisksExistence([self.other_node, self.target_node])
8538
    self._CheckVolumeGroup([self.target_node, self.other_node])
8539

    
8540
    # Step: check other node consistency
8541
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8542
    self._CheckDisksConsistency(self.other_node,
8543
                                self.other_node == self.instance.primary_node,
8544
                                False)
8545

    
8546
    # Step: create new storage
8547
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8548
    iv_names = self._CreateNewStorage(self.target_node)
8549

    
8550
    # Step: for each lv, detach+rename*2+attach
8551
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8552
    for dev, old_lvs, new_lvs in iv_names.itervalues():
8553
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
8554

    
8555
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
8556
                                                     old_lvs)
8557
      result.Raise("Can't detach drbd from local storage on node"
8558
                   " %s for device %s" % (self.target_node, dev.iv_name))
8559
      #dev.children = []
8560
      #cfg.Update(instance)
8561

    
8562
      # ok, we created the new LVs, so now we know we have the needed
8563
      # storage; as such, we proceed on the target node to rename
8564
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
8565
      # using the assumption that logical_id == physical_id (which in
8566
      # turn is the unique_id on that node)
8567

    
8568
      # FIXME(iustin): use a better name for the replaced LVs
8569
      temp_suffix = int(time.time())
8570
      ren_fn = lambda d, suff: (d.physical_id[0],
8571
                                d.physical_id[1] + "_replaced-%s" % suff)
8572

    
8573
      # Build the rename list based on what LVs exist on the node
8574
      rename_old_to_new = []
8575
      for to_ren in old_lvs:
8576
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
8577
        if not result.fail_msg and result.payload:
8578
          # device exists
8579
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
8580

    
8581
      self.lu.LogInfo("Renaming the old LVs on the target node")
8582
      result = self.rpc.call_blockdev_rename(self.target_node,
8583
                                             rename_old_to_new)
8584
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
8585

    
8586
      # Now we rename the new LVs to the old LVs
8587
      self.lu.LogInfo("Renaming the new LVs on the target node")
8588
      rename_new_to_old = [(new, old.physical_id)
8589
                           for old, new in zip(old_lvs, new_lvs)]
8590
      result = self.rpc.call_blockdev_rename(self.target_node,
8591
                                             rename_new_to_old)
8592
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
8593

    
8594
      for old, new in zip(old_lvs, new_lvs):
8595
        new.logical_id = old.logical_id
8596
        self.cfg.SetDiskID(new, self.target_node)
8597

    
8598
      for disk in old_lvs:
8599
        disk.logical_id = ren_fn(disk, temp_suffix)
8600
        self.cfg.SetDiskID(disk, self.target_node)
8601

    
8602
      # Now that the new lvs have the old name, we can add them to the device
8603
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
8604
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
8605
                                                  new_lvs)
8606
      msg = result.fail_msg
8607
      if msg:
8608
        for new_lv in new_lvs:
8609
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
8610
                                               new_lv).fail_msg
8611
          if msg2:
8612
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
8613
                               hint=("cleanup manually the unused logical"
8614
                                     "volumes"))
8615
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
8616

    
8617
      dev.children = new_lvs
8618

    
8619
      self.cfg.Update(self.instance, feedback_fn)
8620

    
8621
    cstep = 5
8622
    if self.early_release:
8623
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8624
      cstep += 1
8625
      self._RemoveOldStorage(self.target_node, iv_names)
8626
      # WARNING: we release both node locks here, do not do other RPCs
8627
      # than WaitForSync to the primary node
8628
      self._ReleaseNodeLock([self.target_node, self.other_node])
8629

    
8630
    # Wait for sync
8631
    # This can fail as the old devices are degraded and _WaitForSync
8632
    # does a combined result over all disks, so we don't check its return value
8633
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8634
    cstep += 1
8635
    _WaitForSync(self.lu, self.instance)
8636

    
8637
    # Check all devices manually
8638
    self._CheckDevices(self.instance.primary_node, iv_names)
8639

    
8640
    # Step: remove old storage
8641
    if not self.early_release:
8642
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8643
      cstep += 1
8644
      self._RemoveOldStorage(self.target_node, iv_names)
8645

    
8646
  def _ExecDrbd8Secondary(self, feedback_fn):
8647
    """Replace the secondary node for DRBD 8.
8648

8649
    The algorithm for replace is quite complicated:
8650
      - for all disks of the instance:
8651
        - create new LVs on the new node with same names
8652
        - shutdown the drbd device on the old secondary
8653
        - disconnect the drbd network on the primary
8654
        - create the drbd device on the new secondary
8655
        - network attach the drbd on the primary, using an artifice:
8656
          the drbd code for Attach() will connect to the network if it
8657
          finds a device which is connected to the good local disks but
8658
          not network enabled
8659
      - wait for sync across all devices
8660
      - remove all disks from the old secondary
8661

8662
    Failures are not very well handled.
8663

8664
    """
8665
    steps_total = 6
8666

    
8667
    # Step: check device activation
8668
    self.lu.LogStep(1, steps_total, "Check device existence")
8669
    self._CheckDisksExistence([self.instance.primary_node])
8670
    self._CheckVolumeGroup([self.instance.primary_node])
8671

    
8672
    # Step: check other node consistency
8673
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8674
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
8675

    
8676
    # Step: create new storage
8677
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8678
    for idx, dev in enumerate(self.instance.disks):
8679
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
8680
                      (self.new_node, idx))
8681
      # we pass force_create=True to force LVM creation
8682
      for new_lv in dev.children:
8683
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
8684
                        _GetInstanceInfoText(self.instance), False)
8685

    
8686
    # Step 4: dbrd minors and drbd setups changes
8687
    # after this, we must manually remove the drbd minors on both the
8688
    # error and the success paths
8689
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8690
    minors = self.cfg.AllocateDRBDMinor([self.new_node
8691
                                         for dev in self.instance.disks],
8692
                                        self.instance.name)
8693
    logging.debug("Allocated minors %r", minors)
8694

    
8695
    iv_names = {}
8696
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
8697
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
8698
                      (self.new_node, idx))
8699
      # create new devices on new_node; note that we create two IDs:
8700
      # one without port, so the drbd will be activated without
8701
      # networking information on the new node at this stage, and one
8702
      # with network, for the latter activation in step 4
8703
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
8704
      if self.instance.primary_node == o_node1:
8705
        p_minor = o_minor1
8706
      else:
8707
        assert self.instance.primary_node == o_node2, "Three-node instance?"
8708
        p_minor = o_minor2
8709

    
8710
      new_alone_id = (self.instance.primary_node, self.new_node, None,
8711
                      p_minor, new_minor, o_secret)
8712
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
8713
                    p_minor, new_minor, o_secret)
8714

    
8715
      iv_names[idx] = (dev, dev.children, new_net_id)
8716
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
8717
                    new_net_id)
8718
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
8719
                              logical_id=new_alone_id,
8720
                              children=dev.children,
8721
                              size=dev.size)
8722
      try:
8723
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
8724
                              _GetInstanceInfoText(self.instance), False)
8725
      except errors.GenericError:
8726
        self.cfg.ReleaseDRBDMinors(self.instance.name)
8727
        raise
8728

    
8729
    # We have new devices, shutdown the drbd on the old secondary
8730
    for idx, dev in enumerate(self.instance.disks):
8731
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
8732
      self.cfg.SetDiskID(dev, self.target_node)
8733
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
8734
      if msg:
8735
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
8736
                           "node: %s" % (idx, msg),
8737
                           hint=("Please cleanup this device manually as"
8738
                                 " soon as possible"))
8739

    
8740
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
8741
    result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
8742
                                               self.node_secondary_ip,
8743
                                               self.instance.disks)\
8744
                                              [self.instance.primary_node]
8745

    
8746
    msg = result.fail_msg
8747
    if msg:
8748
      # detaches didn't succeed (unlikely)
8749
      self.cfg.ReleaseDRBDMinors(self.instance.name)
8750
      raise errors.OpExecError("Can't detach the disks from the network on"
8751
                               " old node: %s" % (msg,))
8752

    
8753
    # if we managed to detach at least one, we update all the disks of
8754
    # the instance to point to the new secondary
8755
    self.lu.LogInfo("Updating instance configuration")
8756
    for dev, _, new_logical_id in iv_names.itervalues():
8757
      dev.logical_id = new_logical_id
8758
      self.cfg.SetDiskID(dev, self.instance.primary_node)
8759

    
8760
    self.cfg.Update(self.instance, feedback_fn)
8761

    
8762
    # and now perform the drbd attach
8763
    self.lu.LogInfo("Attaching primary drbds to new secondary"
8764
                    " (standalone => connected)")
8765
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
8766
                                            self.new_node],
8767
                                           self.node_secondary_ip,
8768
                                           self.instance.disks,
8769
                                           self.instance.name,
8770
                                           False)
8771
    for to_node, to_result in result.items():
8772
      msg = to_result.fail_msg
8773
      if msg:
8774
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
8775
                           to_node, msg,
8776
                           hint=("please do a gnt-instance info to see the"
8777
                                 " status of disks"))
8778
    cstep = 5
8779
    if self.early_release:
8780
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8781
      cstep += 1
8782
      self._RemoveOldStorage(self.target_node, iv_names)
8783
      # WARNING: we release all node locks here, do not do other RPCs
8784
      # than WaitForSync to the primary node
8785
      self._ReleaseNodeLock([self.instance.primary_node,
8786
                             self.target_node,
8787
                             self.new_node])
8788

    
8789
    # Wait for sync
8790
    # This can fail as the old devices are degraded and _WaitForSync
8791
    # does a combined result over all disks, so we don't check its return value
8792
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8793
    cstep += 1
8794
    _WaitForSync(self.lu, self.instance)
8795

    
8796
    # Check all devices manually
8797
    self._CheckDevices(self.instance.primary_node, iv_names)
8798

    
8799
    # Step: remove old storage
8800
    if not self.early_release:
8801
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8802
      self._RemoveOldStorage(self.target_node, iv_names)
8803

    
8804

    
8805
class LURepairNodeStorage(NoHooksLU):
8806
  """Repairs the volume group on a node.
8807

8808
  """
8809
  REQ_BGL = False
8810

    
8811
  def CheckArguments(self):
8812
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
8813

    
8814
    storage_type = self.op.storage_type
8815

    
8816
    if (constants.SO_FIX_CONSISTENCY not in
8817
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
8818
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
8819
                                 " repaired" % storage_type,
8820
                                 errors.ECODE_INVAL)
8821

    
8822
  def ExpandNames(self):
8823
    self.needed_locks = {
8824
      locking.LEVEL_NODE: [self.op.node_name],
8825
      }
8826

    
8827
  def _CheckFaultyDisks(self, instance, node_name):
8828
    """Ensure faulty disks abort the opcode or at least warn."""
8829
    try:
8830
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
8831
                                  node_name, True):
8832
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
8833
                                   " node '%s'" % (instance.name, node_name),
8834
                                   errors.ECODE_STATE)
8835
    except errors.OpPrereqError, err:
8836
      if self.op.ignore_consistency:
8837
        self.proc.LogWarning(str(err.args[0]))
8838
      else:
8839
        raise
8840

    
8841
  def CheckPrereq(self):
8842
    """Check prerequisites.
8843

8844
    """
8845
    # Check whether any instance on this node has faulty disks
8846
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
8847
      if not inst.admin_up:
8848
        continue
8849
      check_nodes = set(inst.all_nodes)
8850
      check_nodes.discard(self.op.node_name)
8851
      for inst_node_name in check_nodes:
8852
        self._CheckFaultyDisks(inst, inst_node_name)
8853

    
8854
  def Exec(self, feedback_fn):
8855
    feedback_fn("Repairing storage unit '%s' on %s ..." %
8856
                (self.op.name, self.op.node_name))
8857

    
8858
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
8859
    result = self.rpc.call_storage_execute(self.op.node_name,
8860
                                           self.op.storage_type, st_args,
8861
                                           self.op.name,
8862
                                           constants.SO_FIX_CONSISTENCY)
8863
    result.Raise("Failed to repair storage unit '%s' on %s" %
8864
                 (self.op.name, self.op.node_name))
8865

    
8866

    
8867
class LUNodeEvacStrategy(NoHooksLU):
8868
  """Computes the node evacuation strategy.
8869

8870
  """
8871
  REQ_BGL = False
8872

    
8873
  def CheckArguments(self):
8874
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
8875

    
8876
  def ExpandNames(self):
8877
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
8878
    self.needed_locks = locks = {}
8879
    if self.op.remote_node is None:
8880
      locks[locking.LEVEL_NODE] = locking.ALL_SET
8881
    else:
8882
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8883
      locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
8884

    
8885
  def Exec(self, feedback_fn):
8886
    if self.op.remote_node is not None:
8887
      instances = []
8888
      for node in self.op.nodes:
8889
        instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
8890
      result = []
8891
      for i in instances:
8892
        if i.primary_node == self.op.remote_node:
8893
          raise errors.OpPrereqError("Node %s is the primary node of"
8894
                                     " instance %s, cannot use it as"
8895
                                     " secondary" %
8896
                                     (self.op.remote_node, i.name),
8897
                                     errors.ECODE_INVAL)
8898
        result.append([i.name, self.op.remote_node])
8899
    else:
8900
      ial = IAllocator(self.cfg, self.rpc,
8901
                       mode=constants.IALLOCATOR_MODE_MEVAC,
8902
                       evac_nodes=self.op.nodes)
8903
      ial.Run(self.op.iallocator, validate=True)
8904
      if not ial.success:
8905
        raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
8906
                                 errors.ECODE_NORES)
8907
      result = ial.result
8908
    return result
8909

    
8910

    
8911
class LUInstanceGrowDisk(LogicalUnit):
8912
  """Grow a disk of an instance.
8913

8914
  """
8915
  HPATH = "disk-grow"
8916
  HTYPE = constants.HTYPE_INSTANCE
8917
  REQ_BGL = False
8918

    
8919
  def ExpandNames(self):
8920
    self._ExpandAndLockInstance()
8921
    self.needed_locks[locking.LEVEL_NODE] = []
8922
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8923

    
8924
  def DeclareLocks(self, level):
8925
    if level == locking.LEVEL_NODE:
8926
      self._LockInstancesNodes()
8927

    
8928
  def BuildHooksEnv(self):
8929
    """Build hooks env.
8930

8931
    This runs on the master, the primary and all the secondaries.
8932

8933
    """
8934
    env = {
8935
      "DISK": self.op.disk,
8936
      "AMOUNT": self.op.amount,
8937
      }
8938
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
8939
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
8940
    return env, nl, nl
8941

    
8942
  def CheckPrereq(self):
8943
    """Check prerequisites.
8944

8945
    This checks that the instance is in the cluster.
8946

8947
    """
8948
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8949
    assert instance is not None, \
8950
      "Cannot retrieve locked instance %s" % self.op.instance_name
8951
    nodenames = list(instance.all_nodes)
8952
    for node in nodenames:
8953
      _CheckNodeOnline(self, node)
8954

    
8955
    self.instance = instance
8956

    
8957
    if instance.disk_template not in constants.DTS_GROWABLE:
8958
      raise errors.OpPrereqError("Instance's disk layout does not support"
8959
                                 " growing.", errors.ECODE_INVAL)
8960

    
8961
    self.disk = instance.FindDisk(self.op.disk)
8962

    
8963
    if instance.disk_template not in (constants.DT_FILE,
8964
                                      constants.DT_SHARED_FILE):
8965
      # TODO: check the free disk space for file, when that feature will be
8966
      # supported
8967
      _CheckNodesFreeDiskPerVG(self, nodenames,
8968
                               self.disk.ComputeGrowth(self.op.amount))
8969

    
8970
  def Exec(self, feedback_fn):
8971
    """Execute disk grow.
8972

8973
    """
8974
    instance = self.instance
8975
    disk = self.disk
8976

    
8977
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
8978
    if not disks_ok:
8979
      raise errors.OpExecError("Cannot activate block device to grow")
8980

    
8981
    for node in instance.all_nodes:
8982
      self.cfg.SetDiskID(disk, node)
8983
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
8984
      result.Raise("Grow request failed to node %s" % node)
8985

    
8986
      # TODO: Rewrite code to work properly
8987
      # DRBD goes into sync mode for a short amount of time after executing the
8988
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
8989
      # calling "resize" in sync mode fails. Sleeping for a short amount of
8990
      # time is a work-around.
8991
      time.sleep(5)
8992

    
8993
    disk.RecordGrow(self.op.amount)
8994
    self.cfg.Update(instance, feedback_fn)
8995
    if self.op.wait_for_sync:
8996
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
8997
      if disk_abort:
8998
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
8999
                             " status.\nPlease check the instance.")
9000
      if not instance.admin_up:
9001
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
9002
    elif not instance.admin_up:
9003
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
9004
                           " not supposed to be running because no wait for"
9005
                           " sync mode was requested.")
9006

    
9007

    
9008
class LUInstanceQueryData(NoHooksLU):
9009
  """Query runtime instance data.
9010

9011
  """
9012
  REQ_BGL = False
9013

    
9014
  def ExpandNames(self):
9015
    self.needed_locks = {}
9016
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
9017

    
9018
    if self.op.instances:
9019
      self.wanted_names = []
9020
      for name in self.op.instances:
9021
        full_name = _ExpandInstanceName(self.cfg, name)
9022
        self.wanted_names.append(full_name)
9023
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
9024
    else:
9025
      self.wanted_names = None
9026
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
9027

    
9028
    self.needed_locks[locking.LEVEL_NODE] = []
9029
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9030

    
9031
  def DeclareLocks(self, level):
9032
    if level == locking.LEVEL_NODE:
9033
      self._LockInstancesNodes()
9034

    
9035
  def CheckPrereq(self):
9036
    """Check prerequisites.
9037

9038
    This only checks the optional instance list against the existing names.
9039

9040
    """
9041
    if self.wanted_names is None:
9042
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
9043

    
9044
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
9045
                             in self.wanted_names]
9046

    
9047
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
9048
    """Returns the status of a block device
9049

9050
    """
9051
    if self.op.static or not node:
9052
      return None
9053

    
9054
    self.cfg.SetDiskID(dev, node)
9055

    
9056
    result = self.rpc.call_blockdev_find(node, dev)
9057
    if result.offline:
9058
      return None
9059

    
9060
    result.Raise("Can't compute disk status for %s" % instance_name)
9061

    
9062
    status = result.payload
9063
    if status is None:
9064
      return None
9065

    
9066
    return (status.dev_path, status.major, status.minor,
9067
            status.sync_percent, status.estimated_time,
9068
            status.is_degraded, status.ldisk_status)
9069

    
9070
  def _ComputeDiskStatus(self, instance, snode, dev):
9071
    """Compute block device status.
9072

9073
    """
9074
    if dev.dev_type in constants.LDS_DRBD:
9075
      # we change the snode then (otherwise we use the one passed in)
9076
      if dev.logical_id[0] == instance.primary_node:
9077
        snode = dev.logical_id[1]
9078
      else:
9079
        snode = dev.logical_id[0]
9080

    
9081
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
9082
                                              instance.name, dev)
9083
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
9084

    
9085
    if dev.children:
9086
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
9087
                      for child in dev.children]
9088
    else:
9089
      dev_children = []
9090

    
9091
    data = {
9092
      "iv_name": dev.iv_name,
9093
      "dev_type": dev.dev_type,
9094
      "logical_id": dev.logical_id,
9095
      "physical_id": dev.physical_id,
9096
      "pstatus": dev_pstatus,
9097
      "sstatus": dev_sstatus,
9098
      "children": dev_children,
9099
      "mode": dev.mode,
9100
      "size": dev.size,
9101
      }
9102

    
9103
    return data
9104

    
9105
  def Exec(self, feedback_fn):
9106
    """Gather and return data"""
9107
    result = {}
9108

    
9109
    cluster = self.cfg.GetClusterInfo()
9110

    
9111
    for instance in self.wanted_instances:
9112
      if not self.op.static:
9113
        remote_info = self.rpc.call_instance_info(instance.primary_node,
9114
                                                  instance.name,
9115
                                                  instance.hypervisor)
9116
        remote_info.Raise("Error checking node %s" % instance.primary_node)
9117
        remote_info = remote_info.payload
9118
        if remote_info and "state" in remote_info:
9119
          remote_state = "up"
9120
        else:
9121
          remote_state = "down"
9122
      else:
9123
        remote_state = None
9124
      if instance.admin_up:
9125
        config_state = "up"
9126
      else:
9127
        config_state = "down"
9128

    
9129
      disks = [self._ComputeDiskStatus(instance, None, device)
9130
               for device in instance.disks]
9131

    
9132
      idict = {
9133
        "name": instance.name,
9134
        "config_state": config_state,
9135
        "run_state": remote_state,
9136
        "pnode": instance.primary_node,
9137
        "snodes": instance.secondary_nodes,
9138
        "os": instance.os,
9139
        # this happens to be the same format used for hooks
9140
        "nics": _NICListToTuple(self, instance.nics),
9141
        "disk_template": instance.disk_template,
9142
        "disks": disks,
9143
        "hypervisor": instance.hypervisor,
9144
        "network_port": instance.network_port,
9145
        "hv_instance": instance.hvparams,
9146
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
9147
        "be_instance": instance.beparams,
9148
        "be_actual": cluster.FillBE(instance),
9149
        "os_instance": instance.osparams,
9150
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
9151
        "serial_no": instance.serial_no,
9152
        "mtime": instance.mtime,
9153
        "ctime": instance.ctime,
9154
        "uuid": instance.uuid,
9155
        }
9156

    
9157
      result[instance.name] = idict
9158

    
9159
    return result
9160

    
9161

    
9162
class LUInstanceSetParams(LogicalUnit):
9163
  """Modifies an instances's parameters.
9164

9165
  """
9166
  HPATH = "instance-modify"
9167
  HTYPE = constants.HTYPE_INSTANCE
9168
  REQ_BGL = False
9169

    
9170
  def CheckArguments(self):
9171
    if not (self.op.nics or self.op.disks or self.op.disk_template or
9172
            self.op.hvparams or self.op.beparams or self.op.os_name):
9173
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
9174

    
9175
    if self.op.hvparams:
9176
      _CheckGlobalHvParams(self.op.hvparams)
9177

    
9178
    # Disk validation
9179
    disk_addremove = 0
9180
    for disk_op, disk_dict in self.op.disks:
9181
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
9182
      if disk_op == constants.DDM_REMOVE:
9183
        disk_addremove += 1
9184
        continue
9185
      elif disk_op == constants.DDM_ADD:
9186
        disk_addremove += 1
9187
      else:
9188
        if not isinstance(disk_op, int):
9189
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
9190
        if not isinstance(disk_dict, dict):
9191
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
9192
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9193

    
9194
      if disk_op == constants.DDM_ADD:
9195
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
9196
        if mode not in constants.DISK_ACCESS_SET:
9197
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
9198
                                     errors.ECODE_INVAL)
9199
        size = disk_dict.get('size', None)
9200
        if size is None:
9201
          raise errors.OpPrereqError("Required disk parameter size missing",
9202
                                     errors.ECODE_INVAL)
9203
        try:
9204
          size = int(size)
9205
        except (TypeError, ValueError), err:
9206
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
9207
                                     str(err), errors.ECODE_INVAL)
9208
        disk_dict['size'] = size
9209
      else:
9210
        # modification of disk
9211
        if 'size' in disk_dict:
9212
          raise errors.OpPrereqError("Disk size change not possible, use"
9213
                                     " grow-disk", errors.ECODE_INVAL)
9214

    
9215
    if disk_addremove > 1:
9216
      raise errors.OpPrereqError("Only one disk add or remove operation"
9217
                                 " supported at a time", errors.ECODE_INVAL)
9218

    
9219
    if self.op.disks and self.op.disk_template is not None:
9220
      raise errors.OpPrereqError("Disk template conversion and other disk"
9221
                                 " changes not supported at the same time",
9222
                                 errors.ECODE_INVAL)
9223

    
9224
    if (self.op.disk_template and
9225
        self.op.disk_template in constants.DTS_NET_MIRROR and
9226
        self.op.remote_node is None):
9227
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
9228
                                 " one requires specifying a secondary node",
9229
                                 errors.ECODE_INVAL)
9230

    
9231
    # NIC validation
9232
    nic_addremove = 0
9233
    for nic_op, nic_dict in self.op.nics:
9234
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
9235
      if nic_op == constants.DDM_REMOVE:
9236
        nic_addremove += 1
9237
        continue
9238
      elif nic_op == constants.DDM_ADD:
9239
        nic_addremove += 1
9240
      else:
9241
        if not isinstance(nic_op, int):
9242
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
9243
        if not isinstance(nic_dict, dict):
9244
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
9245
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9246

    
9247
      # nic_dict should be a dict
9248
      nic_ip = nic_dict.get('ip', None)
9249
      if nic_ip is not None:
9250
        if nic_ip.lower() == constants.VALUE_NONE:
9251
          nic_dict['ip'] = None
9252
        else:
9253
          if not netutils.IPAddress.IsValid(nic_ip):
9254
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
9255
                                       errors.ECODE_INVAL)
9256

    
9257
      nic_bridge = nic_dict.get('bridge', None)
9258
      nic_link = nic_dict.get('link', None)
9259
      if nic_bridge and nic_link:
9260
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
9261
                                   " at the same time", errors.ECODE_INVAL)
9262
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
9263
        nic_dict['bridge'] = None
9264
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
9265
        nic_dict['link'] = None
9266

    
9267
      if nic_op == constants.DDM_ADD:
9268
        nic_mac = nic_dict.get('mac', None)
9269
        if nic_mac is None:
9270
          nic_dict['mac'] = constants.VALUE_AUTO
9271

    
9272
      if 'mac' in nic_dict:
9273
        nic_mac = nic_dict['mac']
9274
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9275
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
9276

    
9277
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
9278
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
9279
                                     " modifying an existing nic",
9280
                                     errors.ECODE_INVAL)
9281

    
9282
    if nic_addremove > 1:
9283
      raise errors.OpPrereqError("Only one NIC add or remove operation"
9284
                                 " supported at a time", errors.ECODE_INVAL)
9285

    
9286
  def ExpandNames(self):
9287
    self._ExpandAndLockInstance()
9288
    self.needed_locks[locking.LEVEL_NODE] = []
9289
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9290

    
9291
  def DeclareLocks(self, level):
9292
    if level == locking.LEVEL_NODE:
9293
      self._LockInstancesNodes()
9294
      if self.op.disk_template and self.op.remote_node:
9295
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9296
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
9297

    
9298
  def BuildHooksEnv(self):
9299
    """Build hooks env.
9300

9301
    This runs on the master, primary and secondaries.
9302

9303
    """
9304
    args = dict()
9305
    if constants.BE_MEMORY in self.be_new:
9306
      args['memory'] = self.be_new[constants.BE_MEMORY]
9307
    if constants.BE_VCPUS in self.be_new:
9308
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
9309
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
9310
    # information at all.
9311
    if self.op.nics:
9312
      args['nics'] = []
9313
      nic_override = dict(self.op.nics)
9314
      for idx, nic in enumerate(self.instance.nics):
9315
        if idx in nic_override:
9316
          this_nic_override = nic_override[idx]
9317
        else:
9318
          this_nic_override = {}
9319
        if 'ip' in this_nic_override:
9320
          ip = this_nic_override['ip']
9321
        else:
9322
          ip = nic.ip
9323
        if 'mac' in this_nic_override:
9324
          mac = this_nic_override['mac']
9325
        else:
9326
          mac = nic.mac
9327
        if idx in self.nic_pnew:
9328
          nicparams = self.nic_pnew[idx]
9329
        else:
9330
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
9331
        mode = nicparams[constants.NIC_MODE]
9332
        link = nicparams[constants.NIC_LINK]
9333
        args['nics'].append((ip, mac, mode, link))
9334
      if constants.DDM_ADD in nic_override:
9335
        ip = nic_override[constants.DDM_ADD].get('ip', None)
9336
        mac = nic_override[constants.DDM_ADD]['mac']
9337
        nicparams = self.nic_pnew[constants.DDM_ADD]
9338
        mode = nicparams[constants.NIC_MODE]
9339
        link = nicparams[constants.NIC_LINK]
9340
        args['nics'].append((ip, mac, mode, link))
9341
      elif constants.DDM_REMOVE in nic_override:
9342
        del args['nics'][-1]
9343

    
9344
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
9345
    if self.op.disk_template:
9346
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
9347
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
9348
    return env, nl, nl
9349

    
9350
  def CheckPrereq(self):
9351
    """Check prerequisites.
9352

9353
    This only checks the instance list against the existing names.
9354

9355
    """
9356
    # checking the new params on the primary/secondary nodes
9357

    
9358
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9359
    cluster = self.cluster = self.cfg.GetClusterInfo()
9360
    assert self.instance is not None, \
9361
      "Cannot retrieve locked instance %s" % self.op.instance_name
9362
    pnode = instance.primary_node
9363
    nodelist = list(instance.all_nodes)
9364

    
9365
    # OS change
9366
    if self.op.os_name and not self.op.force:
9367
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
9368
                      self.op.force_variant)
9369
      instance_os = self.op.os_name
9370
    else:
9371
      instance_os = instance.os
9372

    
9373
    if self.op.disk_template:
9374
      if instance.disk_template == self.op.disk_template:
9375
        raise errors.OpPrereqError("Instance already has disk template %s" %
9376
                                   instance.disk_template, errors.ECODE_INVAL)
9377

    
9378
      if (instance.disk_template,
9379
          self.op.disk_template) not in self._DISK_CONVERSIONS:
9380
        raise errors.OpPrereqError("Unsupported disk template conversion from"
9381
                                   " %s to %s" % (instance.disk_template,
9382
                                                  self.op.disk_template),
9383
                                   errors.ECODE_INVAL)
9384
      _CheckInstanceDown(self, instance, "cannot change disk template")
9385
      if self.op.disk_template in constants.DTS_NET_MIRROR:
9386
        if self.op.remote_node == pnode:
9387
          raise errors.OpPrereqError("Given new secondary node %s is the same"
9388
                                     " as the primary node of the instance" %
9389
                                     self.op.remote_node, errors.ECODE_STATE)
9390
        _CheckNodeOnline(self, self.op.remote_node)
9391
        _CheckNodeNotDrained(self, self.op.remote_node)
9392
        # FIXME: here we assume that the old instance type is DT_PLAIN
9393
        assert instance.disk_template == constants.DT_PLAIN
9394
        disks = [{"size": d.size, "vg": d.logical_id[0]}
9395
                 for d in instance.disks]
9396
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
9397
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
9398

    
9399
    # hvparams processing
9400
    if self.op.hvparams:
9401
      hv_type = instance.hypervisor
9402
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
9403
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
9404
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
9405

    
9406
      # local check
9407
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
9408
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
9409
      self.hv_new = hv_new # the new actual values
9410
      self.hv_inst = i_hvdict # the new dict (without defaults)
9411
    else:
9412
      self.hv_new = self.hv_inst = {}
9413

    
9414
    # beparams processing
9415
    if self.op.beparams:
9416
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
9417
                                   use_none=True)
9418
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
9419
      be_new = cluster.SimpleFillBE(i_bedict)
9420
      self.be_new = be_new # the new actual values
9421
      self.be_inst = i_bedict # the new dict (without defaults)
9422
    else:
9423
      self.be_new = self.be_inst = {}
9424

    
9425
    # osparams processing
9426
    if self.op.osparams:
9427
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
9428
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
9429
      self.os_inst = i_osdict # the new dict (without defaults)
9430
    else:
9431
      self.os_inst = {}
9432

    
9433
    self.warn = []
9434

    
9435
    if constants.BE_MEMORY in self.op.beparams and not self.op.force:
9436
      mem_check_list = [pnode]
9437
      if be_new[constants.BE_AUTO_BALANCE]:
9438
        # either we changed auto_balance to yes or it was from before
9439
        mem_check_list.extend(instance.secondary_nodes)
9440
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
9441
                                                  instance.hypervisor)
9442
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
9443
                                         instance.hypervisor)
9444
      pninfo = nodeinfo[pnode]
9445
      msg = pninfo.fail_msg
9446
      if msg:
9447
        # Assume the primary node is unreachable and go ahead
9448
        self.warn.append("Can't get info from primary node %s: %s" %
9449
                         (pnode,  msg))
9450
      elif not isinstance(pninfo.payload.get('memory_free', None), int):
9451
        self.warn.append("Node data from primary node %s doesn't contain"
9452
                         " free memory information" % pnode)
9453
      elif instance_info.fail_msg:
9454
        self.warn.append("Can't get instance runtime information: %s" %
9455
                        instance_info.fail_msg)
9456
      else:
9457
        if instance_info.payload:
9458
          current_mem = int(instance_info.payload['memory'])
9459
        else:
9460
          # Assume instance not running
9461
          # (there is a slight race condition here, but it's not very probable,
9462
          # and we have no other way to check)
9463
          current_mem = 0
9464
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
9465
                    pninfo.payload['memory_free'])
9466
        if miss_mem > 0:
9467
          raise errors.OpPrereqError("This change will prevent the instance"
9468
                                     " from starting, due to %d MB of memory"
9469
                                     " missing on its primary node" % miss_mem,
9470
                                     errors.ECODE_NORES)
9471

    
9472
      if be_new[constants.BE_AUTO_BALANCE]:
9473
        for node, nres in nodeinfo.items():
9474
          if node not in instance.secondary_nodes:
9475
            continue
9476
          msg = nres.fail_msg
9477
          if msg:
9478
            self.warn.append("Can't get info from secondary node %s: %s" %
9479
                             (node, msg))
9480
          elif not isinstance(nres.payload.get('memory_free', None), int):
9481
            self.warn.append("Secondary node %s didn't return free"
9482
                             " memory information" % node)
9483
          elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
9484
            self.warn.append("Not enough memory to failover instance to"
9485
                             " secondary node %s" % node)
9486

    
9487
    # NIC processing
9488
    self.nic_pnew = {}
9489
    self.nic_pinst = {}
9490
    for nic_op, nic_dict in self.op.nics:
9491
      if nic_op == constants.DDM_REMOVE:
9492
        if not instance.nics:
9493
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
9494
                                     errors.ECODE_INVAL)
9495
        continue
9496
      if nic_op != constants.DDM_ADD:
9497
        # an existing nic
9498
        if not instance.nics:
9499
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
9500
                                     " no NICs" % nic_op,
9501
                                     errors.ECODE_INVAL)
9502
        if nic_op < 0 or nic_op >= len(instance.nics):
9503
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
9504
                                     " are 0 to %d" %
9505
                                     (nic_op, len(instance.nics) - 1),
9506
                                     errors.ECODE_INVAL)
9507
        old_nic_params = instance.nics[nic_op].nicparams
9508
        old_nic_ip = instance.nics[nic_op].ip
9509
      else:
9510
        old_nic_params = {}
9511
        old_nic_ip = None
9512

    
9513
      update_params_dict = dict([(key, nic_dict[key])
9514
                                 for key in constants.NICS_PARAMETERS
9515
                                 if key in nic_dict])
9516

    
9517
      if 'bridge' in nic_dict:
9518
        update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
9519

    
9520
      new_nic_params = _GetUpdatedParams(old_nic_params,
9521
                                         update_params_dict)
9522
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
9523
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
9524
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
9525
      self.nic_pinst[nic_op] = new_nic_params
9526
      self.nic_pnew[nic_op] = new_filled_nic_params
9527
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
9528

    
9529
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
9530
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
9531
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
9532
        if msg:
9533
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
9534
          if self.op.force:
9535
            self.warn.append(msg)
9536
          else:
9537
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
9538
      if new_nic_mode == constants.NIC_MODE_ROUTED:
9539
        if 'ip' in nic_dict:
9540
          nic_ip = nic_dict['ip']
9541
        else:
9542
          nic_ip = old_nic_ip
9543
        if nic_ip is None:
9544
          raise errors.OpPrereqError('Cannot set the nic ip to None'
9545
                                     ' on a routed nic', errors.ECODE_INVAL)
9546
      if 'mac' in nic_dict:
9547
        nic_mac = nic_dict['mac']
9548
        if nic_mac is None:
9549
          raise errors.OpPrereqError('Cannot set the nic mac to None',
9550
                                     errors.ECODE_INVAL)
9551
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9552
          # otherwise generate the mac
9553
          nic_dict['mac'] = self.cfg.GenerateMAC(self.proc.GetECId())
9554
        else:
9555
          # or validate/reserve the current one
9556
          try:
9557
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
9558
          except errors.ReservationError:
9559
            raise errors.OpPrereqError("MAC address %s already in use"
9560
                                       " in cluster" % nic_mac,
9561
                                       errors.ECODE_NOTUNIQUE)
9562

    
9563
    # DISK processing
9564
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
9565
      raise errors.OpPrereqError("Disk operations not supported for"
9566
                                 " diskless instances",
9567
                                 errors.ECODE_INVAL)
9568
    for disk_op, _ in self.op.disks:
9569
      if disk_op == constants.DDM_REMOVE:
9570
        if len(instance.disks) == 1:
9571
          raise errors.OpPrereqError("Cannot remove the last disk of"
9572
                                     " an instance", errors.ECODE_INVAL)
9573
        _CheckInstanceDown(self, instance, "cannot remove disks")
9574

    
9575
      if (disk_op == constants.DDM_ADD and
9576
          len(instance.disks) >= constants.MAX_DISKS):
9577
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
9578
                                   " add more" % constants.MAX_DISKS,
9579
                                   errors.ECODE_STATE)
9580
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
9581
        # an existing disk
9582
        if disk_op < 0 or disk_op >= len(instance.disks):
9583
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
9584
                                     " are 0 to %d" %
9585
                                     (disk_op, len(instance.disks)),
9586
                                     errors.ECODE_INVAL)
9587

    
9588
    return
9589

    
9590
  def _ConvertPlainToDrbd(self, feedback_fn):
9591
    """Converts an instance from plain to drbd.
9592

9593
    """
9594
    feedback_fn("Converting template to drbd")
9595
    instance = self.instance
9596
    pnode = instance.primary_node
9597
    snode = self.op.remote_node
9598

    
9599
    # create a fake disk info for _GenerateDiskTemplate
9600
    disk_info = [{"size": d.size, "mode": d.mode} for d in instance.disks]
9601
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
9602
                                      instance.name, pnode, [snode],
9603
                                      disk_info, None, None, 0, feedback_fn)
9604
    info = _GetInstanceInfoText(instance)
9605
    feedback_fn("Creating aditional volumes...")
9606
    # first, create the missing data and meta devices
9607
    for disk in new_disks:
9608
      # unfortunately this is... not too nice
9609
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
9610
                            info, True)
9611
      for child in disk.children:
9612
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
9613
    # at this stage, all new LVs have been created, we can rename the
9614
    # old ones
9615
    feedback_fn("Renaming original volumes...")
9616
    rename_list = [(o, n.children[0].logical_id)
9617
                   for (o, n) in zip(instance.disks, new_disks)]
9618
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
9619
    result.Raise("Failed to rename original LVs")
9620

    
9621
    feedback_fn("Initializing DRBD devices...")
9622
    # all child devices are in place, we can now create the DRBD devices
9623
    for disk in new_disks:
9624
      for node in [pnode, snode]:
9625
        f_create = node == pnode
9626
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
9627

    
9628
    # at this point, the instance has been modified
9629
    instance.disk_template = constants.DT_DRBD8
9630
    instance.disks = new_disks
9631
    self.cfg.Update(instance, feedback_fn)
9632

    
9633
    # disks are created, waiting for sync
9634
    disk_abort = not _WaitForSync(self, instance)
9635
    if disk_abort:
9636
      raise errors.OpExecError("There are some degraded disks for"
9637
                               " this instance, please cleanup manually")
9638

    
9639
  def _ConvertDrbdToPlain(self, feedback_fn):
9640
    """Converts an instance from drbd to plain.
9641

9642
    """
9643
    instance = self.instance
9644
    assert len(instance.secondary_nodes) == 1
9645
    pnode = instance.primary_node
9646
    snode = instance.secondary_nodes[0]
9647
    feedback_fn("Converting template to plain")
9648

    
9649
    old_disks = instance.disks
9650
    new_disks = [d.children[0] for d in old_disks]
9651

    
9652
    # copy over size and mode
9653
    for parent, child in zip(old_disks, new_disks):
9654
      child.size = parent.size
9655
      child.mode = parent.mode
9656

    
9657
    # update instance structure
9658
    instance.disks = new_disks
9659
    instance.disk_template = constants.DT_PLAIN
9660
    self.cfg.Update(instance, feedback_fn)
9661

    
9662
    feedback_fn("Removing volumes on the secondary node...")
9663
    for disk in old_disks:
9664
      self.cfg.SetDiskID(disk, snode)
9665
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
9666
      if msg:
9667
        self.LogWarning("Could not remove block device %s on node %s,"
9668
                        " continuing anyway: %s", disk.iv_name, snode, msg)
9669

    
9670
    feedback_fn("Removing unneeded volumes on the primary node...")
9671
    for idx, disk in enumerate(old_disks):
9672
      meta = disk.children[1]
9673
      self.cfg.SetDiskID(meta, pnode)
9674
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
9675
      if msg:
9676
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
9677
                        " continuing anyway: %s", idx, pnode, msg)
9678

    
9679
  def Exec(self, feedback_fn):
9680
    """Modifies an instance.
9681

9682
    All parameters take effect only at the next restart of the instance.
9683

9684
    """
9685
    # Process here the warnings from CheckPrereq, as we don't have a
9686
    # feedback_fn there.
9687
    for warn in self.warn:
9688
      feedback_fn("WARNING: %s" % warn)
9689

    
9690
    result = []
9691
    instance = self.instance
9692
    # disk changes
9693
    for disk_op, disk_dict in self.op.disks:
9694
      if disk_op == constants.DDM_REMOVE:
9695
        # remove the last disk
9696
        device = instance.disks.pop()
9697
        device_idx = len(instance.disks)
9698
        for node, disk in device.ComputeNodeTree(instance.primary_node):
9699
          self.cfg.SetDiskID(disk, node)
9700
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
9701
          if msg:
9702
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
9703
                            " continuing anyway", device_idx, node, msg)
9704
        result.append(("disk/%d" % device_idx, "remove"))
9705
      elif disk_op == constants.DDM_ADD:
9706
        # add a new disk
9707
        if instance.disk_template in (constants.DT_FILE,
9708
                                        constants.DT_SHARED_FILE):
9709
          file_driver, file_path = instance.disks[0].logical_id
9710
          file_path = os.path.dirname(file_path)
9711
        else:
9712
          file_driver = file_path = None
9713
        disk_idx_base = len(instance.disks)
9714
        new_disk = _GenerateDiskTemplate(self,
9715
                                         instance.disk_template,
9716
                                         instance.name, instance.primary_node,
9717
                                         instance.secondary_nodes,
9718
                                         [disk_dict],
9719
                                         file_path,
9720
                                         file_driver,
9721
                                         disk_idx_base, feedback_fn)[0]
9722
        instance.disks.append(new_disk)
9723
        info = _GetInstanceInfoText(instance)
9724

    
9725
        logging.info("Creating volume %s for instance %s",
9726
                     new_disk.iv_name, instance.name)
9727
        # Note: this needs to be kept in sync with _CreateDisks
9728
        #HARDCODE
9729
        for node in instance.all_nodes:
9730
          f_create = node == instance.primary_node
9731
          try:
9732
            _CreateBlockDev(self, node, instance, new_disk,
9733
                            f_create, info, f_create)
9734
          except errors.OpExecError, err:
9735
            self.LogWarning("Failed to create volume %s (%s) on"
9736
                            " node %s: %s",
9737
                            new_disk.iv_name, new_disk, node, err)
9738
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
9739
                       (new_disk.size, new_disk.mode)))
9740
      else:
9741
        # change a given disk
9742
        instance.disks[disk_op].mode = disk_dict['mode']
9743
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
9744

    
9745
    if self.op.disk_template:
9746
      r_shut = _ShutdownInstanceDisks(self, instance)
9747
      if not r_shut:
9748
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
9749
                                 " proceed with disk template conversion")
9750
      mode = (instance.disk_template, self.op.disk_template)
9751
      try:
9752
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
9753
      except:
9754
        self.cfg.ReleaseDRBDMinors(instance.name)
9755
        raise
9756
      result.append(("disk_template", self.op.disk_template))
9757

    
9758
    # NIC changes
9759
    for nic_op, nic_dict in self.op.nics:
9760
      if nic_op == constants.DDM_REMOVE:
9761
        # remove the last nic
9762
        del instance.nics[-1]
9763
        result.append(("nic.%d" % len(instance.nics), "remove"))
9764
      elif nic_op == constants.DDM_ADD:
9765
        # mac and bridge should be set, by now
9766
        mac = nic_dict['mac']
9767
        ip = nic_dict.get('ip', None)
9768
        nicparams = self.nic_pinst[constants.DDM_ADD]
9769
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
9770
        instance.nics.append(new_nic)
9771
        result.append(("nic.%d" % (len(instance.nics) - 1),
9772
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
9773
                       (new_nic.mac, new_nic.ip,
9774
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
9775
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
9776
                       )))
9777
      else:
9778
        for key in 'mac', 'ip':
9779
          if key in nic_dict:
9780
            setattr(instance.nics[nic_op], key, nic_dict[key])
9781
        if nic_op in self.nic_pinst:
9782
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
9783
        for key, val in nic_dict.iteritems():
9784
          result.append(("nic.%s/%d" % (key, nic_op), val))
9785

    
9786
    # hvparams changes
9787
    if self.op.hvparams:
9788
      instance.hvparams = self.hv_inst
9789
      for key, val in self.op.hvparams.iteritems():
9790
        result.append(("hv/%s" % key, val))
9791

    
9792
    # beparams changes
9793
    if self.op.beparams:
9794
      instance.beparams = self.be_inst
9795
      for key, val in self.op.beparams.iteritems():
9796
        result.append(("be/%s" % key, val))
9797

    
9798
    # OS change
9799
    if self.op.os_name:
9800
      instance.os = self.op.os_name
9801

    
9802
    # osparams changes
9803
    if self.op.osparams:
9804
      instance.osparams = self.os_inst
9805
      for key, val in self.op.osparams.iteritems():
9806
        result.append(("os/%s" % key, val))
9807

    
9808
    self.cfg.Update(instance, feedback_fn)
9809

    
9810
    return result
9811

    
9812
  _DISK_CONVERSIONS = {
9813
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
9814
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
9815
    }
9816

    
9817

    
9818
class LUBackupQuery(NoHooksLU):
9819
  """Query the exports list
9820

9821
  """
9822
  REQ_BGL = False
9823

    
9824
  def ExpandNames(self):
9825
    self.needed_locks = {}
9826
    self.share_locks[locking.LEVEL_NODE] = 1
9827
    if not self.op.nodes:
9828
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9829
    else:
9830
      self.needed_locks[locking.LEVEL_NODE] = \
9831
        _GetWantedNodes(self, self.op.nodes)
9832

    
9833
  def Exec(self, feedback_fn):
9834
    """Compute the list of all the exported system images.
9835

9836
    @rtype: dict
9837
    @return: a dictionary with the structure node->(export-list)
9838
        where export-list is a list of the instances exported on
9839
        that node.
9840

9841
    """
9842
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
9843
    rpcresult = self.rpc.call_export_list(self.nodes)
9844
    result = {}
9845
    for node in rpcresult:
9846
      if rpcresult[node].fail_msg:
9847
        result[node] = False
9848
      else:
9849
        result[node] = rpcresult[node].payload
9850

    
9851
    return result
9852

    
9853

    
9854
class LUBackupPrepare(NoHooksLU):
9855
  """Prepares an instance for an export and returns useful information.
9856

9857
  """
9858
  REQ_BGL = False
9859

    
9860
  def ExpandNames(self):
9861
    self._ExpandAndLockInstance()
9862

    
9863
  def CheckPrereq(self):
9864
    """Check prerequisites.
9865

9866
    """
9867
    instance_name = self.op.instance_name
9868

    
9869
    self.instance = self.cfg.GetInstanceInfo(instance_name)
9870
    assert self.instance is not None, \
9871
          "Cannot retrieve locked instance %s" % self.op.instance_name
9872
    _CheckNodeOnline(self, self.instance.primary_node)
9873

    
9874
    self._cds = _GetClusterDomainSecret()
9875

    
9876
  def Exec(self, feedback_fn):
9877
    """Prepares an instance for an export.
9878

9879
    """
9880
    instance = self.instance
9881

    
9882
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
9883
      salt = utils.GenerateSecret(8)
9884

    
9885
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
9886
      result = self.rpc.call_x509_cert_create(instance.primary_node,
9887
                                              constants.RIE_CERT_VALIDITY)
9888
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
9889

    
9890
      (name, cert_pem) = result.payload
9891

    
9892
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
9893
                                             cert_pem)
9894

    
9895
      return {
9896
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
9897
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
9898
                          salt),
9899
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
9900
        }
9901

    
9902
    return None
9903

    
9904

    
9905
class LUBackupExport(LogicalUnit):
9906
  """Export an instance to an image in the cluster.
9907

9908
  """
9909
  HPATH = "instance-export"
9910
  HTYPE = constants.HTYPE_INSTANCE
9911
  REQ_BGL = False
9912

    
9913
  def CheckArguments(self):
9914
    """Check the arguments.
9915

9916
    """
9917
    self.x509_key_name = self.op.x509_key_name
9918
    self.dest_x509_ca_pem = self.op.destination_x509_ca
9919

    
9920
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
9921
      if not self.x509_key_name:
9922
        raise errors.OpPrereqError("Missing X509 key name for encryption",
9923
                                   errors.ECODE_INVAL)
9924

    
9925
      if not self.dest_x509_ca_pem:
9926
        raise errors.OpPrereqError("Missing destination X509 CA",
9927
                                   errors.ECODE_INVAL)
9928

    
9929
  def ExpandNames(self):
9930
    self._ExpandAndLockInstance()
9931

    
9932
    # Lock all nodes for local exports
9933
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9934
      # FIXME: lock only instance primary and destination node
9935
      #
9936
      # Sad but true, for now we have do lock all nodes, as we don't know where
9937
      # the previous export might be, and in this LU we search for it and
9938
      # remove it from its current node. In the future we could fix this by:
9939
      #  - making a tasklet to search (share-lock all), then create the
9940
      #    new one, then one to remove, after
9941
      #  - removing the removal operation altogether
9942
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9943

    
9944
  def DeclareLocks(self, level):
9945
    """Last minute lock declaration."""
9946
    # All nodes are locked anyway, so nothing to do here.
9947

    
9948
  def BuildHooksEnv(self):
9949
    """Build hooks env.
9950

9951
    This will run on the master, primary node and target node.
9952

9953
    """
9954
    env = {
9955
      "EXPORT_MODE": self.op.mode,
9956
      "EXPORT_NODE": self.op.target_node,
9957
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
9958
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
9959
      # TODO: Generic function for boolean env variables
9960
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
9961
      }
9962

    
9963
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
9964

    
9965
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
9966

    
9967
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9968
      nl.append(self.op.target_node)
9969

    
9970
    return env, nl, nl
9971

    
9972
  def CheckPrereq(self):
9973
    """Check prerequisites.
9974

9975
    This checks that the instance and node names are valid.
9976

9977
    """
9978
    instance_name = self.op.instance_name
9979

    
9980
    self.instance = self.cfg.GetInstanceInfo(instance_name)
9981
    assert self.instance is not None, \
9982
          "Cannot retrieve locked instance %s" % self.op.instance_name
9983
    _CheckNodeOnline(self, self.instance.primary_node)
9984

    
9985
    if (self.op.remove_instance and self.instance.admin_up and
9986
        not self.op.shutdown):
9987
      raise errors.OpPrereqError("Can not remove instance without shutting it"
9988
                                 " down before")
9989

    
9990
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9991
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
9992
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
9993
      assert self.dst_node is not None
9994

    
9995
      _CheckNodeOnline(self, self.dst_node.name)
9996
      _CheckNodeNotDrained(self, self.dst_node.name)
9997

    
9998
      self._cds = None
9999
      self.dest_disk_info = None
10000
      self.dest_x509_ca = None
10001

    
10002
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10003
      self.dst_node = None
10004

    
10005
      if len(self.op.target_node) != len(self.instance.disks):
10006
        raise errors.OpPrereqError(("Received destination information for %s"
10007
                                    " disks, but instance %s has %s disks") %
10008
                                   (len(self.op.target_node), instance_name,
10009
                                    len(self.instance.disks)),
10010
                                   errors.ECODE_INVAL)
10011

    
10012
      cds = _GetClusterDomainSecret()
10013

    
10014
      # Check X509 key name
10015
      try:
10016
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
10017
      except (TypeError, ValueError), err:
10018
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
10019

    
10020
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
10021
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
10022
                                   errors.ECODE_INVAL)
10023

    
10024
      # Load and verify CA
10025
      try:
10026
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
10027
      except OpenSSL.crypto.Error, err:
10028
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
10029
                                   (err, ), errors.ECODE_INVAL)
10030

    
10031
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
10032
      if errcode is not None:
10033
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
10034
                                   (msg, ), errors.ECODE_INVAL)
10035

    
10036
      self.dest_x509_ca = cert
10037

    
10038
      # Verify target information
10039
      disk_info = []
10040
      for idx, disk_data in enumerate(self.op.target_node):
10041
        try:
10042
          (host, port, magic) = \
10043
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
10044
        except errors.GenericError, err:
10045
          raise errors.OpPrereqError("Target info for disk %s: %s" %
10046
                                     (idx, err), errors.ECODE_INVAL)
10047

    
10048
        disk_info.append((host, port, magic))
10049

    
10050
      assert len(disk_info) == len(self.op.target_node)
10051
      self.dest_disk_info = disk_info
10052

    
10053
    else:
10054
      raise errors.ProgrammerError("Unhandled export mode %r" %
10055
                                   self.op.mode)
10056

    
10057
    # instance disk type verification
10058
    # TODO: Implement export support for file-based disks
10059
    for disk in self.instance.disks:
10060
      if disk.dev_type == constants.LD_FILE:
10061
        raise errors.OpPrereqError("Export not supported for instances with"
10062
                                   " file-based disks", errors.ECODE_INVAL)
10063

    
10064
  def _CleanupExports(self, feedback_fn):
10065
    """Removes exports of current instance from all other nodes.
10066

10067
    If an instance in a cluster with nodes A..D was exported to node C, its
10068
    exports will be removed from the nodes A, B and D.
10069

10070
    """
10071
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
10072

    
10073
    nodelist = self.cfg.GetNodeList()
10074
    nodelist.remove(self.dst_node.name)
10075

    
10076
    # on one-node clusters nodelist will be empty after the removal
10077
    # if we proceed the backup would be removed because OpBackupQuery
10078
    # substitutes an empty list with the full cluster node list.
10079
    iname = self.instance.name
10080
    if nodelist:
10081
      feedback_fn("Removing old exports for instance %s" % iname)
10082
      exportlist = self.rpc.call_export_list(nodelist)
10083
      for node in exportlist:
10084
        if exportlist[node].fail_msg:
10085
          continue
10086
        if iname in exportlist[node].payload:
10087
          msg = self.rpc.call_export_remove(node, iname).fail_msg
10088
          if msg:
10089
            self.LogWarning("Could not remove older export for instance %s"
10090
                            " on node %s: %s", iname, node, msg)
10091

    
10092
  def Exec(self, feedback_fn):
10093
    """Export an instance to an image in the cluster.
10094

10095
    """
10096
    assert self.op.mode in constants.EXPORT_MODES
10097

    
10098
    instance = self.instance
10099
    src_node = instance.primary_node
10100

    
10101
    if self.op.shutdown:
10102
      # shutdown the instance, but not the disks
10103
      feedback_fn("Shutting down instance %s" % instance.name)
10104
      result = self.rpc.call_instance_shutdown(src_node, instance,
10105
                                               self.op.shutdown_timeout)
10106
      # TODO: Maybe ignore failures if ignore_remove_failures is set
10107
      result.Raise("Could not shutdown instance %s on"
10108
                   " node %s" % (instance.name, src_node))
10109

    
10110
    # set the disks ID correctly since call_instance_start needs the
10111
    # correct drbd minor to create the symlinks
10112
    for disk in instance.disks:
10113
      self.cfg.SetDiskID(disk, src_node)
10114

    
10115
    activate_disks = (not instance.admin_up)
10116

    
10117
    if activate_disks:
10118
      # Activate the instance disks if we'exporting a stopped instance
10119
      feedback_fn("Activating disks for %s" % instance.name)
10120
      _StartInstanceDisks(self, instance, None)
10121

    
10122
    try:
10123
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
10124
                                                     instance)
10125

    
10126
      helper.CreateSnapshots()
10127
      try:
10128
        if (self.op.shutdown and instance.admin_up and
10129
            not self.op.remove_instance):
10130
          assert not activate_disks
10131
          feedback_fn("Starting instance %s" % instance.name)
10132
          result = self.rpc.call_instance_start(src_node, instance, None, None)
10133
          msg = result.fail_msg
10134
          if msg:
10135
            feedback_fn("Failed to start instance: %s" % msg)
10136
            _ShutdownInstanceDisks(self, instance)
10137
            raise errors.OpExecError("Could not start instance: %s" % msg)
10138

    
10139
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
10140
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
10141
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10142
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
10143
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
10144

    
10145
          (key_name, _, _) = self.x509_key_name
10146

    
10147
          dest_ca_pem = \
10148
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
10149
                                            self.dest_x509_ca)
10150

    
10151
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
10152
                                                     key_name, dest_ca_pem,
10153
                                                     timeouts)
10154
      finally:
10155
        helper.Cleanup()
10156

    
10157
      # Check for backwards compatibility
10158
      assert len(dresults) == len(instance.disks)
10159
      assert compat.all(isinstance(i, bool) for i in dresults), \
10160
             "Not all results are boolean: %r" % dresults
10161

    
10162
    finally:
10163
      if activate_disks:
10164
        feedback_fn("Deactivating disks for %s" % instance.name)
10165
        _ShutdownInstanceDisks(self, instance)
10166

    
10167
    if not (compat.all(dresults) and fin_resu):
10168
      failures = []
10169
      if not fin_resu:
10170
        failures.append("export finalization")
10171
      if not compat.all(dresults):
10172
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
10173
                               if not dsk)
10174
        failures.append("disk export: disk(s) %s" % fdsk)
10175

    
10176
      raise errors.OpExecError("Export failed, errors in %s" %
10177
                               utils.CommaJoin(failures))
10178

    
10179
    # At this point, the export was successful, we can cleanup/finish
10180

    
10181
    # Remove instance if requested
10182
    if self.op.remove_instance:
10183
      feedback_fn("Removing instance %s" % instance.name)
10184
      _RemoveInstance(self, feedback_fn, instance,
10185
                      self.op.ignore_remove_failures)
10186

    
10187
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10188
      self._CleanupExports(feedback_fn)
10189

    
10190
    return fin_resu, dresults
10191

    
10192

    
10193
class LUBackupRemove(NoHooksLU):
10194
  """Remove exports related to the named instance.
10195

10196
  """
10197
  REQ_BGL = False
10198

    
10199
  def ExpandNames(self):
10200
    self.needed_locks = {}
10201
    # We need all nodes to be locked in order for RemoveExport to work, but we
10202
    # don't need to lock the instance itself, as nothing will happen to it (and
10203
    # we can remove exports also for a removed instance)
10204
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10205

    
10206
  def Exec(self, feedback_fn):
10207
    """Remove any export.
10208

10209
    """
10210
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
10211
    # If the instance was not found we'll try with the name that was passed in.
10212
    # This will only work if it was an FQDN, though.
10213
    fqdn_warn = False
10214
    if not instance_name:
10215
      fqdn_warn = True
10216
      instance_name = self.op.instance_name
10217

    
10218
    locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
10219
    exportlist = self.rpc.call_export_list(locked_nodes)
10220
    found = False
10221
    for node in exportlist:
10222
      msg = exportlist[node].fail_msg
10223
      if msg:
10224
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
10225
        continue
10226
      if instance_name in exportlist[node].payload:
10227
        found = True
10228
        result = self.rpc.call_export_remove(node, instance_name)
10229
        msg = result.fail_msg
10230
        if msg:
10231
          logging.error("Could not remove export for instance %s"
10232
                        " on node %s: %s", instance_name, node, msg)
10233

    
10234
    if fqdn_warn and not found:
10235
      feedback_fn("Export not found. If trying to remove an export belonging"
10236
                  " to a deleted instance please use its Fully Qualified"
10237
                  " Domain Name.")
10238

    
10239

    
10240
class LUGroupAdd(LogicalUnit):
10241
  """Logical unit for creating node groups.
10242

10243
  """
10244
  HPATH = "group-add"
10245
  HTYPE = constants.HTYPE_GROUP
10246
  REQ_BGL = False
10247

    
10248
  def ExpandNames(self):
10249
    # We need the new group's UUID here so that we can create and acquire the
10250
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
10251
    # that it should not check whether the UUID exists in the configuration.
10252
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
10253
    self.needed_locks = {}
10254
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10255

    
10256
  def CheckPrereq(self):
10257
    """Check prerequisites.
10258

10259
    This checks that the given group name is not an existing node group
10260
    already.
10261

10262
    """
10263
    try:
10264
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10265
    except errors.OpPrereqError:
10266
      pass
10267
    else:
10268
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
10269
                                 " node group (UUID: %s)" %
10270
                                 (self.op.group_name, existing_uuid),
10271
                                 errors.ECODE_EXISTS)
10272

    
10273
    if self.op.ndparams:
10274
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10275

    
10276
  def BuildHooksEnv(self):
10277
    """Build hooks env.
10278

10279
    """
10280
    env = {
10281
      "GROUP_NAME": self.op.group_name,
10282
      }
10283
    mn = self.cfg.GetMasterNode()
10284
    return env, [mn], [mn]
10285

    
10286
  def Exec(self, feedback_fn):
10287
    """Add the node group to the cluster.
10288

10289
    """
10290
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
10291
                                  uuid=self.group_uuid,
10292
                                  alloc_policy=self.op.alloc_policy,
10293
                                  ndparams=self.op.ndparams)
10294

    
10295
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
10296
    del self.remove_locks[locking.LEVEL_NODEGROUP]
10297

    
10298

    
10299
class LUGroupAssignNodes(NoHooksLU):
10300
  """Logical unit for assigning nodes to groups.
10301

10302
  """
10303
  REQ_BGL = False
10304

    
10305
  def ExpandNames(self):
10306
    # These raise errors.OpPrereqError on their own:
10307
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10308
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
10309

    
10310
    # We want to lock all the affected nodes and groups. We have readily
10311
    # available the list of nodes, and the *destination* group. To gather the
10312
    # list of "source" groups, we need to fetch node information.
10313
    self.node_data = self.cfg.GetAllNodesInfo()
10314
    affected_groups = set(self.node_data[node].group for node in self.op.nodes)
10315
    affected_groups.add(self.group_uuid)
10316

    
10317
    self.needed_locks = {
10318
      locking.LEVEL_NODEGROUP: list(affected_groups),
10319
      locking.LEVEL_NODE: self.op.nodes,
10320
      }
10321

    
10322
  def CheckPrereq(self):
10323
    """Check prerequisites.
10324

10325
    """
10326
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10327
    instance_data = self.cfg.GetAllInstancesInfo()
10328

    
10329
    if self.group is None:
10330
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10331
                               (self.op.group_name, self.group_uuid))
10332

    
10333
    (new_splits, previous_splits) = \
10334
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
10335
                                             for node in self.op.nodes],
10336
                                            self.node_data, instance_data)
10337

    
10338
    if new_splits:
10339
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
10340

    
10341
      if not self.op.force:
10342
        raise errors.OpExecError("The following instances get split by this"
10343
                                 " change and --force was not given: %s" %
10344
                                 fmt_new_splits)
10345
      else:
10346
        self.LogWarning("This operation will split the following instances: %s",
10347
                        fmt_new_splits)
10348

    
10349
        if previous_splits:
10350
          self.LogWarning("In addition, these already-split instances continue"
10351
                          " to be spit across groups: %s",
10352
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
10353

    
10354
  def Exec(self, feedback_fn):
10355
    """Assign nodes to a new group.
10356

10357
    """
10358
    for node in self.op.nodes:
10359
      self.node_data[node].group = self.group_uuid
10360

    
10361
    self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
10362

    
10363
  @staticmethod
10364
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
10365
    """Check for split instances after a node assignment.
10366

10367
    This method considers a series of node assignments as an atomic operation,
10368
    and returns information about split instances after applying the set of
10369
    changes.
10370

10371
    In particular, it returns information about newly split instances, and
10372
    instances that were already split, and remain so after the change.
10373

10374
    Only instances whose disk template is listed in constants.DTS_NET_MIRROR are
10375
    considered.
10376

10377
    @type changes: list of (node_name, new_group_uuid) pairs.
10378
    @param changes: list of node assignments to consider.
10379
    @param node_data: a dict with data for all nodes
10380
    @param instance_data: a dict with all instances to consider
10381
    @rtype: a two-tuple
10382
    @return: a list of instances that were previously okay and result split as a
10383
      consequence of this change, and a list of instances that were previously
10384
      split and this change does not fix.
10385

10386
    """
10387
    changed_nodes = dict((node, group) for node, group in changes
10388
                         if node_data[node].group != group)
10389

    
10390
    all_split_instances = set()
10391
    previously_split_instances = set()
10392

    
10393
    def InstanceNodes(instance):
10394
      return [instance.primary_node] + list(instance.secondary_nodes)
10395

    
10396
    for inst in instance_data.values():
10397
      if inst.disk_template not in constants.DTS_NET_MIRROR:
10398
        continue
10399

    
10400
      instance_nodes = InstanceNodes(inst)
10401

    
10402
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
10403
        previously_split_instances.add(inst.name)
10404

    
10405
      if len(set(changed_nodes.get(node, node_data[node].group)
10406
                 for node in instance_nodes)) > 1:
10407
        all_split_instances.add(inst.name)
10408

    
10409
    return (list(all_split_instances - previously_split_instances),
10410
            list(previously_split_instances & all_split_instances))
10411

    
10412

    
10413
class _GroupQuery(_QueryBase):
10414

    
10415
  FIELDS = query.GROUP_FIELDS
10416

    
10417
  def ExpandNames(self, lu):
10418
    lu.needed_locks = {}
10419

    
10420
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
10421
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
10422

    
10423
    if not self.names:
10424
      self.wanted = [name_to_uuid[name]
10425
                     for name in utils.NiceSort(name_to_uuid.keys())]
10426
    else:
10427
      # Accept names to be either names or UUIDs.
10428
      missing = []
10429
      self.wanted = []
10430
      all_uuid = frozenset(self._all_groups.keys())
10431

    
10432
      for name in self.names:
10433
        if name in all_uuid:
10434
          self.wanted.append(name)
10435
        elif name in name_to_uuid:
10436
          self.wanted.append(name_to_uuid[name])
10437
        else:
10438
          missing.append(name)
10439

    
10440
      if missing:
10441
        raise errors.OpPrereqError("Some groups do not exist: %s" % missing,
10442
                                   errors.ECODE_NOENT)
10443

    
10444
  def DeclareLocks(self, lu, level):
10445
    pass
10446

    
10447
  def _GetQueryData(self, lu):
10448
    """Computes the list of node groups and their attributes.
10449

10450
    """
10451
    do_nodes = query.GQ_NODE in self.requested_data
10452
    do_instances = query.GQ_INST in self.requested_data
10453

    
10454
    group_to_nodes = None
10455
    group_to_instances = None
10456

    
10457
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
10458
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
10459
    # latter GetAllInstancesInfo() is not enough, for we have to go through
10460
    # instance->node. Hence, we will need to process nodes even if we only need
10461
    # instance information.
10462
    if do_nodes or do_instances:
10463
      all_nodes = lu.cfg.GetAllNodesInfo()
10464
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
10465
      node_to_group = {}
10466

    
10467
      for node in all_nodes.values():
10468
        if node.group in group_to_nodes:
10469
          group_to_nodes[node.group].append(node.name)
10470
          node_to_group[node.name] = node.group
10471

    
10472
      if do_instances:
10473
        all_instances = lu.cfg.GetAllInstancesInfo()
10474
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
10475

    
10476
        for instance in all_instances.values():
10477
          node = instance.primary_node
10478
          if node in node_to_group:
10479
            group_to_instances[node_to_group[node]].append(instance.name)
10480

    
10481
        if not do_nodes:
10482
          # Do not pass on node information if it was not requested.
10483
          group_to_nodes = None
10484

    
10485
    return query.GroupQueryData([self._all_groups[uuid]
10486
                                 for uuid in self.wanted],
10487
                                group_to_nodes, group_to_instances)
10488

    
10489

    
10490
class LUGroupQuery(NoHooksLU):
10491
  """Logical unit for querying node groups.
10492

10493
  """
10494
  REQ_BGL = False
10495

    
10496
  def CheckArguments(self):
10497
    self.gq = _GroupQuery(self.op.names, self.op.output_fields, False)
10498

    
10499
  def ExpandNames(self):
10500
    self.gq.ExpandNames(self)
10501

    
10502
  def Exec(self, feedback_fn):
10503
    return self.gq.OldStyleQuery(self)
10504

    
10505

    
10506
class LUGroupSetParams(LogicalUnit):
10507
  """Modifies the parameters of a node group.
10508

10509
  """
10510
  HPATH = "group-modify"
10511
  HTYPE = constants.HTYPE_GROUP
10512
  REQ_BGL = False
10513

    
10514
  def CheckArguments(self):
10515
    all_changes = [
10516
      self.op.ndparams,
10517
      self.op.alloc_policy,
10518
      ]
10519

    
10520
    if all_changes.count(None) == len(all_changes):
10521
      raise errors.OpPrereqError("Please pass at least one modification",
10522
                                 errors.ECODE_INVAL)
10523

    
10524
  def ExpandNames(self):
10525
    # This raises errors.OpPrereqError on its own:
10526
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10527

    
10528
    self.needed_locks = {
10529
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10530
      }
10531

    
10532
  def CheckPrereq(self):
10533
    """Check prerequisites.
10534

10535
    """
10536
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10537

    
10538
    if self.group is None:
10539
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10540
                               (self.op.group_name, self.group_uuid))
10541

    
10542
    if self.op.ndparams:
10543
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
10544
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10545
      self.new_ndparams = new_ndparams
10546

    
10547
  def BuildHooksEnv(self):
10548
    """Build hooks env.
10549

10550
    """
10551
    env = {
10552
      "GROUP_NAME": self.op.group_name,
10553
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
10554
      }
10555
    mn = self.cfg.GetMasterNode()
10556
    return env, [mn], [mn]
10557

    
10558
  def Exec(self, feedback_fn):
10559
    """Modifies the node group.
10560

10561
    """
10562
    result = []
10563

    
10564
    if self.op.ndparams:
10565
      self.group.ndparams = self.new_ndparams
10566
      result.append(("ndparams", str(self.group.ndparams)))
10567

    
10568
    if self.op.alloc_policy:
10569
      self.group.alloc_policy = self.op.alloc_policy
10570

    
10571
    self.cfg.Update(self.group, feedback_fn)
10572
    return result
10573

    
10574

    
10575

    
10576
class LUGroupRemove(LogicalUnit):
10577
  HPATH = "group-remove"
10578
  HTYPE = constants.HTYPE_GROUP
10579
  REQ_BGL = False
10580

    
10581
  def ExpandNames(self):
10582
    # This will raises errors.OpPrereqError on its own:
10583
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10584
    self.needed_locks = {
10585
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10586
      }
10587

    
10588
  def CheckPrereq(self):
10589
    """Check prerequisites.
10590

10591
    This checks that the given group name exists as a node group, that is
10592
    empty (i.e., contains no nodes), and that is not the last group of the
10593
    cluster.
10594

10595
    """
10596
    # Verify that the group is empty.
10597
    group_nodes = [node.name
10598
                   for node in self.cfg.GetAllNodesInfo().values()
10599
                   if node.group == self.group_uuid]
10600

    
10601
    if group_nodes:
10602
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
10603
                                 " nodes: %s" %
10604
                                 (self.op.group_name,
10605
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
10606
                                 errors.ECODE_STATE)
10607

    
10608
    # Verify the cluster would not be left group-less.
10609
    if len(self.cfg.GetNodeGroupList()) == 1:
10610
      raise errors.OpPrereqError("Group '%s' is the only group,"
10611
                                 " cannot be removed" %
10612
                                 self.op.group_name,
10613
                                 errors.ECODE_STATE)
10614

    
10615
  def BuildHooksEnv(self):
10616
    """Build hooks env.
10617

10618
    """
10619
    env = {
10620
      "GROUP_NAME": self.op.group_name,
10621
      }
10622
    mn = self.cfg.GetMasterNode()
10623
    return env, [mn], [mn]
10624

    
10625
  def Exec(self, feedback_fn):
10626
    """Remove the node group.
10627

10628
    """
10629
    try:
10630
      self.cfg.RemoveNodeGroup(self.group_uuid)
10631
    except errors.ConfigurationError:
10632
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
10633
                               (self.op.group_name, self.group_uuid))
10634

    
10635
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10636

    
10637

    
10638
class LUGroupRename(LogicalUnit):
10639
  HPATH = "group-rename"
10640
  HTYPE = constants.HTYPE_GROUP
10641
  REQ_BGL = False
10642

    
10643
  def ExpandNames(self):
10644
    # This raises errors.OpPrereqError on its own:
10645
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.old_name)
10646

    
10647
    self.needed_locks = {
10648
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10649
      }
10650

    
10651
  def CheckPrereq(self):
10652
    """Check prerequisites.
10653

10654
    This checks that the given old_name exists as a node group, and that
10655
    new_name doesn't.
10656

10657
    """
10658
    try:
10659
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
10660
    except errors.OpPrereqError:
10661
      pass
10662
    else:
10663
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
10664
                                 " node group (UUID: %s)" %
10665
                                 (self.op.new_name, new_name_uuid),
10666
                                 errors.ECODE_EXISTS)
10667

    
10668
  def BuildHooksEnv(self):
10669
    """Build hooks env.
10670

10671
    """
10672
    env = {
10673
      "OLD_NAME": self.op.old_name,
10674
      "NEW_NAME": self.op.new_name,
10675
      }
10676

    
10677
    mn = self.cfg.GetMasterNode()
10678
    all_nodes = self.cfg.GetAllNodesInfo()
10679
    run_nodes = [mn]
10680
    all_nodes.pop(mn, None)
10681

    
10682
    for node in all_nodes.values():
10683
      if node.group == self.group_uuid:
10684
        run_nodes.append(node.name)
10685

    
10686
    return env, run_nodes, run_nodes
10687

    
10688
  def Exec(self, feedback_fn):
10689
    """Rename the node group.
10690

10691
    """
10692
    group = self.cfg.GetNodeGroup(self.group_uuid)
10693

    
10694
    if group is None:
10695
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10696
                               (self.op.old_name, self.group_uuid))
10697

    
10698
    group.name = self.op.new_name
10699
    self.cfg.Update(group, feedback_fn)
10700

    
10701
    return self.op.new_name
10702

    
10703

    
10704
class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
10705
  """Generic tags LU.
10706

10707
  This is an abstract class which is the parent of all the other tags LUs.
10708

10709
  """
10710

    
10711
  def ExpandNames(self):
10712
    self.needed_locks = {}
10713
    if self.op.kind == constants.TAG_NODE:
10714
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
10715
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
10716
    elif self.op.kind == constants.TAG_INSTANCE:
10717
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
10718
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
10719

    
10720
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
10721
    # not possible to acquire the BGL based on opcode parameters)
10722

    
10723
  def CheckPrereq(self):
10724
    """Check prerequisites.
10725

10726
    """
10727
    if self.op.kind == constants.TAG_CLUSTER:
10728
      self.target = self.cfg.GetClusterInfo()
10729
    elif self.op.kind == constants.TAG_NODE:
10730
      self.target = self.cfg.GetNodeInfo(self.op.name)
10731
    elif self.op.kind == constants.TAG_INSTANCE:
10732
      self.target = self.cfg.GetInstanceInfo(self.op.name)
10733
    else:
10734
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
10735
                                 str(self.op.kind), errors.ECODE_INVAL)
10736

    
10737

    
10738
class LUTagsGet(TagsLU):
10739
  """Returns the tags of a given object.
10740

10741
  """
10742
  REQ_BGL = False
10743

    
10744
  def ExpandNames(self):
10745
    TagsLU.ExpandNames(self)
10746

    
10747
    # Share locks as this is only a read operation
10748
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
10749

    
10750
  def Exec(self, feedback_fn):
10751
    """Returns the tag list.
10752

10753
    """
10754
    return list(self.target.GetTags())
10755

    
10756

    
10757
class LUTagsSearch(NoHooksLU):
10758
  """Searches the tags for a given pattern.
10759

10760
  """
10761
  REQ_BGL = False
10762

    
10763
  def ExpandNames(self):
10764
    self.needed_locks = {}
10765

    
10766
  def CheckPrereq(self):
10767
    """Check prerequisites.
10768

10769
    This checks the pattern passed for validity by compiling it.
10770

10771
    """
10772
    try:
10773
      self.re = re.compile(self.op.pattern)
10774
    except re.error, err:
10775
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
10776
                                 (self.op.pattern, err), errors.ECODE_INVAL)
10777

    
10778
  def Exec(self, feedback_fn):
10779
    """Returns the tag list.
10780

10781
    """
10782
    cfg = self.cfg
10783
    tgts = [("/cluster", cfg.GetClusterInfo())]
10784
    ilist = cfg.GetAllInstancesInfo().values()
10785
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
10786
    nlist = cfg.GetAllNodesInfo().values()
10787
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
10788
    results = []
10789
    for path, target in tgts:
10790
      for tag in target.GetTags():
10791
        if self.re.search(tag):
10792
          results.append((path, tag))
10793
    return results
10794

    
10795

    
10796
class LUTagsSet(TagsLU):
10797
  """Sets a tag on a given object.
10798

10799
  """
10800
  REQ_BGL = False
10801

    
10802
  def CheckPrereq(self):
10803
    """Check prerequisites.
10804

10805
    This checks the type and length of the tag name and value.
10806

10807
    """
10808
    TagsLU.CheckPrereq(self)
10809
    for tag in self.op.tags:
10810
      objects.TaggableObject.ValidateTag(tag)
10811

    
10812
  def Exec(self, feedback_fn):
10813
    """Sets the tag.
10814

10815
    """
10816
    try:
10817
      for tag in self.op.tags:
10818
        self.target.AddTag(tag)
10819
    except errors.TagError, err:
10820
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
10821
    self.cfg.Update(self.target, feedback_fn)
10822

    
10823

    
10824
class LUTagsDel(TagsLU):
10825
  """Delete a list of tags from a given object.
10826

10827
  """
10828
  REQ_BGL = False
10829

    
10830
  def CheckPrereq(self):
10831
    """Check prerequisites.
10832

10833
    This checks that we have the given tag.
10834

10835
    """
10836
    TagsLU.CheckPrereq(self)
10837
    for tag in self.op.tags:
10838
      objects.TaggableObject.ValidateTag(tag)
10839
    del_tags = frozenset(self.op.tags)
10840
    cur_tags = self.target.GetTags()
10841

    
10842
    diff_tags = del_tags - cur_tags
10843
    if diff_tags:
10844
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
10845
      raise errors.OpPrereqError("Tag(s) %s not found" %
10846
                                 (utils.CommaJoin(diff_names), ),
10847
                                 errors.ECODE_NOENT)
10848

    
10849
  def Exec(self, feedback_fn):
10850
    """Remove the tag from the object.
10851

10852
    """
10853
    for tag in self.op.tags:
10854
      self.target.RemoveTag(tag)
10855
    self.cfg.Update(self.target, feedback_fn)
10856

    
10857

    
10858
class LUTestDelay(NoHooksLU):
10859
  """Sleep for a specified amount of time.
10860

10861
  This LU sleeps on the master and/or nodes for a specified amount of
10862
  time.
10863

10864
  """
10865
  REQ_BGL = False
10866

    
10867
  def ExpandNames(self):
10868
    """Expand names and set required locks.
10869

10870
    This expands the node list, if any.
10871

10872
    """
10873
    self.needed_locks = {}
10874
    if self.op.on_nodes:
10875
      # _GetWantedNodes can be used here, but is not always appropriate to use
10876
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
10877
      # more information.
10878
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
10879
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
10880

    
10881
  def _TestDelay(self):
10882
    """Do the actual sleep.
10883

10884
    """
10885
    if self.op.on_master:
10886
      if not utils.TestDelay(self.op.duration):
10887
        raise errors.OpExecError("Error during master delay test")
10888
    if self.op.on_nodes:
10889
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
10890
      for node, node_result in result.items():
10891
        node_result.Raise("Failure during rpc call to node %s" % node)
10892

    
10893
  def Exec(self, feedback_fn):
10894
    """Execute the test delay opcode, with the wanted repetitions.
10895

10896
    """
10897
    if self.op.repeat == 0:
10898
      self._TestDelay()
10899
    else:
10900
      top_value = self.op.repeat - 1
10901
      for i in range(self.op.repeat):
10902
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
10903
        self._TestDelay()
10904

    
10905

    
10906
class LUTestJqueue(NoHooksLU):
10907
  """Utility LU to test some aspects of the job queue.
10908

10909
  """
10910
  REQ_BGL = False
10911

    
10912
  # Must be lower than default timeout for WaitForJobChange to see whether it
10913
  # notices changed jobs
10914
  _CLIENT_CONNECT_TIMEOUT = 20.0
10915
  _CLIENT_CONFIRM_TIMEOUT = 60.0
10916

    
10917
  @classmethod
10918
  def _NotifyUsingSocket(cls, cb, errcls):
10919
    """Opens a Unix socket and waits for another program to connect.
10920

10921
    @type cb: callable
10922
    @param cb: Callback to send socket name to client
10923
    @type errcls: class
10924
    @param errcls: Exception class to use for errors
10925

10926
    """
10927
    # Using a temporary directory as there's no easy way to create temporary
10928
    # sockets without writing a custom loop around tempfile.mktemp and
10929
    # socket.bind
10930
    tmpdir = tempfile.mkdtemp()
10931
    try:
10932
      tmpsock = utils.PathJoin(tmpdir, "sock")
10933

    
10934
      logging.debug("Creating temporary socket at %s", tmpsock)
10935
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
10936
      try:
10937
        sock.bind(tmpsock)
10938
        sock.listen(1)
10939

    
10940
        # Send details to client
10941
        cb(tmpsock)
10942

    
10943
        # Wait for client to connect before continuing
10944
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
10945
        try:
10946
          (conn, _) = sock.accept()
10947
        except socket.error, err:
10948
          raise errcls("Client didn't connect in time (%s)" % err)
10949
      finally:
10950
        sock.close()
10951
    finally:
10952
      # Remove as soon as client is connected
10953
      shutil.rmtree(tmpdir)
10954

    
10955
    # Wait for client to close
10956
    try:
10957
      try:
10958
        # pylint: disable-msg=E1101
10959
        # Instance of '_socketobject' has no ... member
10960
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
10961
        conn.recv(1)
10962
      except socket.error, err:
10963
        raise errcls("Client failed to confirm notification (%s)" % err)
10964
    finally:
10965
      conn.close()
10966

    
10967
  def _SendNotification(self, test, arg, sockname):
10968
    """Sends a notification to the client.
10969

10970
    @type test: string
10971
    @param test: Test name
10972
    @param arg: Test argument (depends on test)
10973
    @type sockname: string
10974
    @param sockname: Socket path
10975

10976
    """
10977
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
10978

    
10979
  def _Notify(self, prereq, test, arg):
10980
    """Notifies the client of a test.
10981

10982
    @type prereq: bool
10983
    @param prereq: Whether this is a prereq-phase test
10984
    @type test: string
10985
    @param test: Test name
10986
    @param arg: Test argument (depends on test)
10987

10988
    """
10989
    if prereq:
10990
      errcls = errors.OpPrereqError
10991
    else:
10992
      errcls = errors.OpExecError
10993

    
10994
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
10995
                                                  test, arg),
10996
                                   errcls)
10997

    
10998
  def CheckArguments(self):
10999
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
11000
    self.expandnames_calls = 0
11001

    
11002
  def ExpandNames(self):
11003
    checkargs_calls = getattr(self, "checkargs_calls", 0)
11004
    if checkargs_calls < 1:
11005
      raise errors.ProgrammerError("CheckArguments was not called")
11006

    
11007
    self.expandnames_calls += 1
11008

    
11009
    if self.op.notify_waitlock:
11010
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
11011

    
11012
    self.LogInfo("Expanding names")
11013

    
11014
    # Get lock on master node (just to get a lock, not for a particular reason)
11015
    self.needed_locks = {
11016
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
11017
      }
11018

    
11019
  def Exec(self, feedback_fn):
11020
    if self.expandnames_calls < 1:
11021
      raise errors.ProgrammerError("ExpandNames was not called")
11022

    
11023
    if self.op.notify_exec:
11024
      self._Notify(False, constants.JQT_EXEC, None)
11025

    
11026
    self.LogInfo("Executing")
11027

    
11028
    if self.op.log_messages:
11029
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
11030
      for idx, msg in enumerate(self.op.log_messages):
11031
        self.LogInfo("Sending log message %s", idx + 1)
11032
        feedback_fn(constants.JQT_MSGPREFIX + msg)
11033
        # Report how many test messages have been sent
11034
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
11035

    
11036
    if self.op.fail:
11037
      raise errors.OpExecError("Opcode failure was requested")
11038

    
11039
    return True
11040

    
11041

    
11042
class IAllocator(object):
11043
  """IAllocator framework.
11044

11045
  An IAllocator instance has three sets of attributes:
11046
    - cfg that is needed to query the cluster
11047
    - input data (all members of the _KEYS class attribute are required)
11048
    - four buffer attributes (in|out_data|text), that represent the
11049
      input (to the external script) in text and data structure format,
11050
      and the output from it, again in two formats
11051
    - the result variables from the script (success, info, nodes) for
11052
      easy usage
11053

11054
  """
11055
  # pylint: disable-msg=R0902
11056
  # lots of instance attributes
11057
  _ALLO_KEYS = [
11058
    "name", "mem_size", "disks", "disk_template",
11059
    "os", "tags", "nics", "vcpus", "hypervisor",
11060
    ]
11061
  _RELO_KEYS = [
11062
    "name", "relocate_from",
11063
    ]
11064
  _EVAC_KEYS = [
11065
    "evac_nodes",
11066
    ]
11067

    
11068
  def __init__(self, cfg, rpc, mode, **kwargs):
11069
    self.cfg = cfg
11070
    self.rpc = rpc
11071
    # init buffer variables
11072
    self.in_text = self.out_text = self.in_data = self.out_data = None
11073
    # init all input fields so that pylint is happy
11074
    self.mode = mode
11075
    self.mem_size = self.disks = self.disk_template = None
11076
    self.os = self.tags = self.nics = self.vcpus = None
11077
    self.hypervisor = None
11078
    self.relocate_from = None
11079
    self.name = None
11080
    self.evac_nodes = None
11081
    # computed fields
11082
    self.required_nodes = None
11083
    # init result fields
11084
    self.success = self.info = self.result = None
11085
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11086
      keyset = self._ALLO_KEYS
11087
      fn = self._AddNewInstance
11088
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11089
      keyset = self._RELO_KEYS
11090
      fn = self._AddRelocateInstance
11091
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11092
      keyset = self._EVAC_KEYS
11093
      fn = self._AddEvacuateNodes
11094
    else:
11095
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
11096
                                   " IAllocator" % self.mode)
11097
    for key in kwargs:
11098
      if key not in keyset:
11099
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
11100
                                     " IAllocator" % key)
11101
      setattr(self, key, kwargs[key])
11102

    
11103
    for key in keyset:
11104
      if key not in kwargs:
11105
        raise errors.ProgrammerError("Missing input parameter '%s' to"
11106
                                     " IAllocator" % key)
11107
    self._BuildInputData(fn)
11108

    
11109
  def _ComputeClusterData(self):
11110
    """Compute the generic allocator input data.
11111

11112
    This is the data that is independent of the actual operation.
11113

11114
    """
11115
    cfg = self.cfg
11116
    cluster_info = cfg.GetClusterInfo()
11117
    # cluster data
11118
    data = {
11119
      "version": constants.IALLOCATOR_VERSION,
11120
      "cluster_name": cfg.GetClusterName(),
11121
      "cluster_tags": list(cluster_info.GetTags()),
11122
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
11123
      # we don't have job IDs
11124
      }
11125
    ninfo = cfg.GetAllNodesInfo()
11126
    iinfo = cfg.GetAllInstancesInfo().values()
11127
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
11128

    
11129
    # node data
11130
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
11131

    
11132
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11133
      hypervisor_name = self.hypervisor
11134
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11135
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
11136
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11137
      hypervisor_name = cluster_info.enabled_hypervisors[0]
11138

    
11139
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
11140
                                        hypervisor_name)
11141
    node_iinfo = \
11142
      self.rpc.call_all_instances_info(node_list,
11143
                                       cluster_info.enabled_hypervisors)
11144

    
11145
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
11146

    
11147
    config_ndata = self._ComputeBasicNodeData(ninfo)
11148
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
11149
                                                 i_list, config_ndata)
11150
    assert len(data["nodes"]) == len(ninfo), \
11151
        "Incomplete node data computed"
11152

    
11153
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
11154

    
11155
    self.in_data = data
11156

    
11157
  @staticmethod
11158
  def _ComputeNodeGroupData(cfg):
11159
    """Compute node groups data.
11160

11161
    """
11162
    ng = {}
11163
    for guuid, gdata in cfg.GetAllNodeGroupsInfo().items():
11164
      ng[guuid] = {
11165
        "name": gdata.name,
11166
        "alloc_policy": gdata.alloc_policy,
11167
        }
11168
    return ng
11169

    
11170
  @staticmethod
11171
  def _ComputeBasicNodeData(node_cfg):
11172
    """Compute global node data.
11173

11174
    @rtype: dict
11175
    @returns: a dict of name: (node dict, node config)
11176

11177
    """
11178
    node_results = {}
11179
    for ninfo in node_cfg.values():
11180
      # fill in static (config-based) values
11181
      pnr = {
11182
        "tags": list(ninfo.GetTags()),
11183
        "primary_ip": ninfo.primary_ip,
11184
        "secondary_ip": ninfo.secondary_ip,
11185
        "offline": ninfo.offline,
11186
        "drained": ninfo.drained,
11187
        "master_candidate": ninfo.master_candidate,
11188
        "group": ninfo.group,
11189
        "master_capable": ninfo.master_capable,
11190
        "vm_capable": ninfo.vm_capable,
11191
        }
11192

    
11193
      node_results[ninfo.name] = pnr
11194

    
11195
    return node_results
11196

    
11197
  @staticmethod
11198
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
11199
                              node_results):
11200
    """Compute global node data.
11201

11202
    @param node_results: the basic node structures as filled from the config
11203

11204
    """
11205
    # make a copy of the current dict
11206
    node_results = dict(node_results)
11207
    for nname, nresult in node_data.items():
11208
      assert nname in node_results, "Missing basic data for node %s" % nname
11209
      ninfo = node_cfg[nname]
11210

    
11211
      if not (ninfo.offline or ninfo.drained):
11212
        nresult.Raise("Can't get data for node %s" % nname)
11213
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
11214
                                nname)
11215
        remote_info = nresult.payload
11216

    
11217
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
11218
                     'vg_size', 'vg_free', 'cpu_total']:
11219
          if attr not in remote_info:
11220
            raise errors.OpExecError("Node '%s' didn't return attribute"
11221
                                     " '%s'" % (nname, attr))
11222
          if not isinstance(remote_info[attr], int):
11223
            raise errors.OpExecError("Node '%s' returned invalid value"
11224
                                     " for '%s': %s" %
11225
                                     (nname, attr, remote_info[attr]))
11226
        # compute memory used by primary instances
11227
        i_p_mem = i_p_up_mem = 0
11228
        for iinfo, beinfo in i_list:
11229
          if iinfo.primary_node == nname:
11230
            i_p_mem += beinfo[constants.BE_MEMORY]
11231
            if iinfo.name not in node_iinfo[nname].payload:
11232
              i_used_mem = 0
11233
            else:
11234
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
11235
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
11236
            remote_info['memory_free'] -= max(0, i_mem_diff)
11237

    
11238
            if iinfo.admin_up:
11239
              i_p_up_mem += beinfo[constants.BE_MEMORY]
11240

    
11241
        # compute memory used by instances
11242
        pnr_dyn = {
11243
          "total_memory": remote_info['memory_total'],
11244
          "reserved_memory": remote_info['memory_dom0'],
11245
          "free_memory": remote_info['memory_free'],
11246
          "total_disk": remote_info['vg_size'],
11247
          "free_disk": remote_info['vg_free'],
11248
          "total_cpus": remote_info['cpu_total'],
11249
          "i_pri_memory": i_p_mem,
11250
          "i_pri_up_memory": i_p_up_mem,
11251
          }
11252
        pnr_dyn.update(node_results[nname])
11253
        node_results[nname] = pnr_dyn
11254

    
11255
    return node_results
11256

    
11257
  @staticmethod
11258
  def _ComputeInstanceData(cluster_info, i_list):
11259
    """Compute global instance data.
11260

11261
    """
11262
    instance_data = {}
11263
    for iinfo, beinfo in i_list:
11264
      nic_data = []
11265
      for nic in iinfo.nics:
11266
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
11267
        nic_dict = {"mac": nic.mac,
11268
                    "ip": nic.ip,
11269
                    "mode": filled_params[constants.NIC_MODE],
11270
                    "link": filled_params[constants.NIC_LINK],
11271
                   }
11272
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
11273
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
11274
        nic_data.append(nic_dict)
11275
      pir = {
11276
        "tags": list(iinfo.GetTags()),
11277
        "admin_up": iinfo.admin_up,
11278
        "vcpus": beinfo[constants.BE_VCPUS],
11279
        "memory": beinfo[constants.BE_MEMORY],
11280
        "os": iinfo.os,
11281
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
11282
        "nics": nic_data,
11283
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
11284
        "disk_template": iinfo.disk_template,
11285
        "hypervisor": iinfo.hypervisor,
11286
        }
11287
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
11288
                                                 pir["disks"])
11289
      instance_data[iinfo.name] = pir
11290

    
11291
    return instance_data
11292

    
11293
  def _AddNewInstance(self):
11294
    """Add new instance data to allocator structure.
11295

11296
    This in combination with _AllocatorGetClusterData will create the
11297
    correct structure needed as input for the allocator.
11298

11299
    The checks for the completeness of the opcode must have already been
11300
    done.
11301

11302
    """
11303
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
11304

    
11305
    if self.disk_template in constants.DTS_NET_MIRROR:
11306
      self.required_nodes = 2
11307
    else:
11308
      self.required_nodes = 1
11309
    request = {
11310
      "name": self.name,
11311
      "disk_template": self.disk_template,
11312
      "tags": self.tags,
11313
      "os": self.os,
11314
      "vcpus": self.vcpus,
11315
      "memory": self.mem_size,
11316
      "disks": self.disks,
11317
      "disk_space_total": disk_space,
11318
      "nics": self.nics,
11319
      "required_nodes": self.required_nodes,
11320
      }
11321
    return request
11322

    
11323
  def _AddRelocateInstance(self):
11324
    """Add relocate instance data to allocator structure.
11325

11326
    This in combination with _IAllocatorGetClusterData will create the
11327
    correct structure needed as input for the allocator.
11328

11329
    The checks for the completeness of the opcode must have already been
11330
    done.
11331

11332
    """
11333
    instance = self.cfg.GetInstanceInfo(self.name)
11334
    if instance is None:
11335
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
11336
                                   " IAllocator" % self.name)
11337

    
11338
    if instance.disk_template not in constants.DTS_MIRRORED:
11339
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
11340
                                 errors.ECODE_INVAL)
11341

    
11342
    if instance.disk_template in constants.DTS_NET_MIRROR and \
11343
        len(instance.secondary_nodes) != 1:
11344
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
11345
                                 errors.ECODE_STATE)
11346

    
11347
    self.required_nodes = 1
11348
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
11349
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
11350

    
11351
    request = {
11352
      "name": self.name,
11353
      "disk_space_total": disk_space,
11354
      "required_nodes": self.required_nodes,
11355
      "relocate_from": self.relocate_from,
11356
      }
11357
    return request
11358

    
11359
  def _AddEvacuateNodes(self):
11360
    """Add evacuate nodes data to allocator structure.
11361

11362
    """
11363
    request = {
11364
      "evac_nodes": self.evac_nodes
11365
      }
11366
    return request
11367

    
11368
  def _BuildInputData(self, fn):
11369
    """Build input data structures.
11370

11371
    """
11372
    self._ComputeClusterData()
11373

    
11374
    request = fn()
11375
    request["type"] = self.mode
11376
    self.in_data["request"] = request
11377

    
11378
    self.in_text = serializer.Dump(self.in_data)
11379

    
11380
  def Run(self, name, validate=True, call_fn=None):
11381
    """Run an instance allocator and return the results.
11382

11383
    """
11384
    if call_fn is None:
11385
      call_fn = self.rpc.call_iallocator_runner
11386

    
11387
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
11388
    result.Raise("Failure while running the iallocator script")
11389

    
11390
    self.out_text = result.payload
11391
    if validate:
11392
      self._ValidateResult()
11393

    
11394
  def _ValidateResult(self):
11395
    """Process the allocator results.
11396

11397
    This will process and if successful save the result in
11398
    self.out_data and the other parameters.
11399

11400
    """
11401
    try:
11402
      rdict = serializer.Load(self.out_text)
11403
    except Exception, err:
11404
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
11405

    
11406
    if not isinstance(rdict, dict):
11407
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
11408

    
11409
    # TODO: remove backwards compatiblity in later versions
11410
    if "nodes" in rdict and "result" not in rdict:
11411
      rdict["result"] = rdict["nodes"]
11412
      del rdict["nodes"]
11413

    
11414
    for key in "success", "info", "result":
11415
      if key not in rdict:
11416
        raise errors.OpExecError("Can't parse iallocator results:"
11417
                                 " missing key '%s'" % key)
11418
      setattr(self, key, rdict[key])
11419

    
11420
    if not isinstance(rdict["result"], list):
11421
      raise errors.OpExecError("Can't parse iallocator results: 'result' key"
11422
                               " is not a list")
11423
    self.out_data = rdict
11424

    
11425

    
11426
class LUTestAllocator(NoHooksLU):
11427
  """Run allocator tests.
11428

11429
  This LU runs the allocator tests
11430

11431
  """
11432
  def CheckPrereq(self):
11433
    """Check prerequisites.
11434

11435
    This checks the opcode parameters depending on the director and mode test.
11436

11437
    """
11438
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11439
      for attr in ["mem_size", "disks", "disk_template",
11440
                   "os", "tags", "nics", "vcpus"]:
11441
        if not hasattr(self.op, attr):
11442
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
11443
                                     attr, errors.ECODE_INVAL)
11444
      iname = self.cfg.ExpandInstanceName(self.op.name)
11445
      if iname is not None:
11446
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
11447
                                   iname, errors.ECODE_EXISTS)
11448
      if not isinstance(self.op.nics, list):
11449
        raise errors.OpPrereqError("Invalid parameter 'nics'",
11450
                                   errors.ECODE_INVAL)
11451
      if not isinstance(self.op.disks, list):
11452
        raise errors.OpPrereqError("Invalid parameter 'disks'",
11453
                                   errors.ECODE_INVAL)
11454
      for row in self.op.disks:
11455
        if (not isinstance(row, dict) or
11456
            "size" not in row or
11457
            not isinstance(row["size"], int) or
11458
            "mode" not in row or
11459
            row["mode"] not in ['r', 'w']):
11460
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
11461
                                     " parameter", errors.ECODE_INVAL)
11462
      if self.op.hypervisor is None:
11463
        self.op.hypervisor = self.cfg.GetHypervisorType()
11464
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11465
      fname = _ExpandInstanceName(self.cfg, self.op.name)
11466
      self.op.name = fname
11467
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
11468
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11469
      if not hasattr(self.op, "evac_nodes"):
11470
        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
11471
                                   " opcode input", errors.ECODE_INVAL)
11472
    else:
11473
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
11474
                                 self.op.mode, errors.ECODE_INVAL)
11475

    
11476
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
11477
      if self.op.allocator is None:
11478
        raise errors.OpPrereqError("Missing allocator name",
11479
                                   errors.ECODE_INVAL)
11480
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
11481
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
11482
                                 self.op.direction, errors.ECODE_INVAL)
11483

    
11484
  def Exec(self, feedback_fn):
11485
    """Run the allocator test.
11486

11487
    """
11488
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11489
      ial = IAllocator(self.cfg, self.rpc,
11490
                       mode=self.op.mode,
11491
                       name=self.op.name,
11492
                       mem_size=self.op.mem_size,
11493
                       disks=self.op.disks,
11494
                       disk_template=self.op.disk_template,
11495
                       os=self.op.os,
11496
                       tags=self.op.tags,
11497
                       nics=self.op.nics,
11498
                       vcpus=self.op.vcpus,
11499
                       hypervisor=self.op.hypervisor,
11500
                       )
11501
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11502
      ial = IAllocator(self.cfg, self.rpc,
11503
                       mode=self.op.mode,
11504
                       name=self.op.name,
11505
                       relocate_from=list(self.relocate_from),
11506
                       )
11507
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11508
      ial = IAllocator(self.cfg, self.rpc,
11509
                       mode=self.op.mode,
11510
                       evac_nodes=self.op.evac_nodes)
11511
    else:
11512
      raise errors.ProgrammerError("Uncatched mode %s in"
11513
                                   " LUTestAllocator.Exec", self.op.mode)
11514

    
11515
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
11516
      result = ial.in_text
11517
    else:
11518
      ial.Run(self.op.allocator, validate=False)
11519
      result = ial.out_text
11520
    return result
11521

    
11522

    
11523
#: Query type implementations
11524
_QUERY_IMPL = {
11525
  constants.QR_INSTANCE: _InstanceQuery,
11526
  constants.QR_NODE: _NodeQuery,
11527
  constants.QR_GROUP: _GroupQuery,
11528
  }
11529

    
11530

    
11531
def _GetQueryImplementation(name):
11532
  """Returns the implemtnation for a query type.
11533

11534
  @param name: Query type, must be one of L{constants.QR_OP_QUERY}
11535

11536
  """
11537
  try:
11538
    return _QUERY_IMPL[name]
11539
  except KeyError:
11540
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
11541
                               errors.ECODE_INVAL)