Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ b92a2f3e

History | View | Annotate | Download (405.7 kB)

1
#
2
#
3

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

    
21

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

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

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

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

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

    
44
from ganeti import ssh
45
from ganeti import utils
46
from ganeti import errors
47
from ganeti import hypervisor
48
from ganeti import locking
49
from ganeti import constants
50
from ganeti import objects
51
from ganeti import serializer
52
from ganeti import ssconf
53
from ganeti import uidpool
54
from ganeti import compat
55
from ganeti import masterd
56
from ganeti import netutils
57
from ganeti import query
58
from ganeti import qlang
59
from ganeti import opcodes
60

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

    
63

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

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

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

    
76

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

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

90
  Note that all commands require root permissions.
91

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

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

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

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

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

    
133
    # Tasklets
134
    self.tasklets = None
135

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

    
139
    self.CheckArguments()
140

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

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

    
149
  ssh = property(fget=__GetSSH)
150

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

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

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

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

166
    """
167
    pass
168

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

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

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

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

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

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

194
    Examples::
195

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

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

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

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

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

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

233
    """
234

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

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

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

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

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

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

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

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

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

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

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

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

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

292
    """
293
    raise NotImplementedError
294

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
381
    del self.recalculate_locks[locking.LEVEL_NODE]
382

    
383

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

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

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

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

397
    This just raises an error.
398

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

    
402

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

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

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

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

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

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

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

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

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

435
    """
436
    pass
437

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

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

445
    """
446
    raise NotImplementedError
447

    
448

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

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

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

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

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

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

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

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

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

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

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

    
492
    # Return expanded names
493
    return self.wanted
494

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

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

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

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

507
    See L{LogicalUnit.ExpandNames}.
508

509
    """
510
    raise NotImplementedError()
511

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

515
    See L{LogicalUnit.DeclareLocks}.
516

517
    """
518
    raise NotImplementedError()
519

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

523
    @return: Query data object
524

525
    """
526
    raise NotImplementedError()
527

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

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

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

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

    
540

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

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

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

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

    
558

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

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

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

    
578

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

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

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

    
611

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

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

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

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

    
630

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

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

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

    
645

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

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

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

    
660

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

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

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

    
673

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

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

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

    
686

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

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

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

    
704

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

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

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

    
731

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

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

    
739

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

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

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

    
755

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

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

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

    
772

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

    
777

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

    
782

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

788
  This builds the hook environment from individual variables.
789

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

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

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

    
852
  env["INSTANCE_NIC_COUNT"] = nic_count
853

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

    
862
  env["INSTANCE_DISK_COUNT"] = disk_count
863

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

    
868
  return env
869

    
870

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

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

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

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

    
894

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

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

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

    
932

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

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

    
948

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

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

    
959

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

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

    
973

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

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

    
982

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

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

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

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

    
1002

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

    
1006

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

1010
  """
1011

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

    
1014

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

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

    
1022

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

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

    
1030

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

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

    
1040
  return []
1041

    
1042

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

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

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

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

    
1057
  return faulty
1058

    
1059

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

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

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

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

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

    
1091

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

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

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

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

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

1110
    """
1111
    return True
1112

    
1113

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

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

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

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

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

1131
    This checks whether the cluster is empty.
1132

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

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

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

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

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

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

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

    
1166
    return master
1167

    
1168

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

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

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

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

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

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

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

    
1201

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
1412
    return True
1413

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
1797
    nimg.os_fail = test
1798

    
1799
    if test:
1800
      return
1801

    
1802
    os_dict = {}
1803

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

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

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

    
1816
    nimg.oslist = os_dict
1817

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2013
      node_disks[nname] = disks
2014

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

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

    
2022
      node_disks_devonly[nname] = devonly
2023

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

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

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

    
2032
    instdisk = {}
2033

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

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

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

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

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

    
2073
    return instdisk
2074

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

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

    
2089

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

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

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

    
2104
    return env, [], all_nodes
2105

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

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

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

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

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

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

    
2152
    local_checksums = utils.FingerprintFiles(file_names)
2153

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

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

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

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

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

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

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

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

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

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

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

    
2243
      inst_config.MapLVsByNode(node_vol_should)
2244

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

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

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

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

    
2267
    all_drbd_map = self.cfg.ComputeDRBDMap()
2268

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

    
2272
    feedback_fn("* Verifying node status")
2273

    
2274
    refos_img = None
2275

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

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

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

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

    
2304
      nresult = all_nvinfo[node].payload
2305

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

    
2312
      self._VerifyOob(node_i, nresult)
2313

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2429
    return not self.bad
2430

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

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

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

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

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

    
2474
      return lu_result
2475

    
2476

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

2480
  """
2481
  REQ_BGL = False
2482

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

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

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

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

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

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

    
2515
    if not nv_dict:
2516
      return result
2517

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

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

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

    
2542
    return result
2543

    
2544

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

2548
  """
2549
  REQ_BGL = False
2550

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2661

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

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

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

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

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

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

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

    
2702
    self.op.name = new_name
2703

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

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

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

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

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

    
2737
    return clustername
2738

    
2739

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
3077

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

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

    
3091

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

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

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

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

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

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

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

    
3140

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

3144
  This is a very simple LU.
3145

3146
  """
3147
  REQ_BGL = False
3148

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

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

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

    
3162

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

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

    
3170
  disks = _ExpandCheckDisks(instance, disks)
3171

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

    
3175
  node = instance.primary_node
3176

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

    
3180
  # TODO: Convert to utils.Retry
3181

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

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

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

    
3228
    if done or oneshot:
3229
      break
3230

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

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

    
3237

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

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

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

    
3248
  result = True
3249

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

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

    
3269
  return result
3270

    
3271

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

3275
  """
3276
  REG_BGL = False
3277

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
3378
    return ret
3379

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

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

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

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

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

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

    
3416

    
3417

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
3551
    return output
3552

    
3553

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

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

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

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

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

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

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

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

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

    
3595
    instance_list = self.cfg.GetInstanceList()
3596

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

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

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

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

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

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

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

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

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

    
3649

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

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

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

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

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

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

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

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

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

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

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

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

    
3697
      inst_data = lu.cfg.GetAllInstancesInfo()
3698

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

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

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

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

    
3725

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

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

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

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

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

    
3743

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

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

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

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

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

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

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

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

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

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

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

    
3817
        output.append(node_output)
3818

    
3819
    return output
3820

    
3821

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

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

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

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

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

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

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

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

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

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

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

    
3869
    result = []
3870

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

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

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

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

    
3886
        out = []
3887

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

    
3898
          out.append(val)
3899

    
3900
        result.append(out)
3901

    
3902
    return result
3903

    
3904

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

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

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

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

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

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

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

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

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

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

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

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

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

    
3999

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

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

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

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

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

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

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

    
4022

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

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

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

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

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

    
4039

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

4043
  """
4044
  REQ_BGL = False
4045

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

    
4049
    storage_type = self.op.storage_type
4050

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

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

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

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

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

    
4081

    
4082
class LUNodeAdd(LogicalUnit):
4083
  """Logical unit for adding node to the cluster.
4084

4085
  """
4086
  HPATH = "node-add"
4087
  HTYPE = constants.HTYPE_NODE
4088
  _NFLAGS = ["master_capable", "vm_capable"]
4089

    
4090
  def CheckArguments(self):
4091
    self.primary_ip_family = self.cfg.GetPrimaryIPFamily()
4092
    # validate/normalize the node name
4093
    self.hostname = netutils.GetHostname(name=self.op.node_name,
4094
                                         family=self.primary_ip_family)
4095
    self.op.node_name = self.hostname.name
4096

    
4097
    if self.op.readd and self.op.node_name == self.cfg.GetMasterNode():
4098
      raise errors.OpPrereqError("Cannot readd the master node",
4099
                                 errors.ECODE_STATE)
4100

    
4101
    if self.op.readd and self.op.group:
4102
      raise errors.OpPrereqError("Cannot pass a node group when a node is"
4103
                                 " being readded", errors.ECODE_INVAL)
4104

    
4105
  def BuildHooksEnv(self):
4106
    """Build hooks env.
4107

4108
    This will run on all nodes before, and on all nodes + the new node after.
4109

4110
    """
4111
    env = {
4112
      "OP_TARGET": self.op.node_name,
4113
      "NODE_NAME": self.op.node_name,
4114
      "NODE_PIP": self.op.primary_ip,
4115
      "NODE_SIP": self.op.secondary_ip,
4116
      "MASTER_CAPABLE": str(self.op.master_capable),
4117
      "VM_CAPABLE": str(self.op.vm_capable),
4118
      }
4119
    nodes_0 = self.cfg.GetNodeList()
4120
    nodes_1 = nodes_0 + [self.op.node_name, ]
4121
    return env, nodes_0, nodes_1
4122

    
4123
  def CheckPrereq(self):
4124
    """Check prerequisites.
4125

4126
    This checks:
4127
     - the new node is not already in the config
4128
     - it is resolvable
4129
     - its parameters (single/dual homed) matches the cluster
4130

4131
    Any errors are signaled by raising errors.OpPrereqError.
4132

4133
    """
4134
    cfg = self.cfg
4135
    hostname = self.hostname
4136
    node = hostname.name
4137
    primary_ip = self.op.primary_ip = hostname.ip
4138
    if self.op.secondary_ip is None:
4139
      if self.primary_ip_family == netutils.IP6Address.family:
4140
        raise errors.OpPrereqError("When using a IPv6 primary address, a valid"
4141
                                   " IPv4 address must be given as secondary",
4142
                                   errors.ECODE_INVAL)
4143
      self.op.secondary_ip = primary_ip
4144

    
4145
    secondary_ip = self.op.secondary_ip
4146
    if not netutils.IP4Address.IsValid(secondary_ip):
4147
      raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
4148
                                 " address" % secondary_ip, errors.ECODE_INVAL)
4149

    
4150
    node_list = cfg.GetNodeList()
4151
    if not self.op.readd and node in node_list:
4152
      raise errors.OpPrereqError("Node %s is already in the configuration" %
4153
                                 node, errors.ECODE_EXISTS)
4154
    elif self.op.readd and node not in node_list:
4155
      raise errors.OpPrereqError("Node %s is not in the configuration" % node,
4156
                                 errors.ECODE_NOENT)
4157

    
4158
    self.changed_primary_ip = False
4159

    
4160
    for existing_node_name in node_list:
4161
      existing_node = cfg.GetNodeInfo(existing_node_name)
4162

    
4163
      if self.op.readd and node == existing_node_name:
4164
        if existing_node.secondary_ip != secondary_ip:
4165
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
4166
                                     " address configuration as before",
4167
                                     errors.ECODE_INVAL)
4168
        if existing_node.primary_ip != primary_ip:
4169
          self.changed_primary_ip = True
4170

    
4171
        continue
4172

    
4173
      if (existing_node.primary_ip == primary_ip or
4174
          existing_node.secondary_ip == primary_ip or
4175
          existing_node.primary_ip == secondary_ip or
4176
          existing_node.secondary_ip == secondary_ip):
4177
        raise errors.OpPrereqError("New node ip address(es) conflict with"
4178
                                   " existing node %s" % existing_node.name,
4179
                                   errors.ECODE_NOTUNIQUE)
4180

    
4181
    # After this 'if' block, None is no longer a valid value for the
4182
    # _capable op attributes
4183
    if self.op.readd:
4184
      old_node = self.cfg.GetNodeInfo(node)
4185
      assert old_node is not None, "Can't retrieve locked node %s" % node
4186
      for attr in self._NFLAGS:
4187
        if getattr(self.op, attr) is None:
4188
          setattr(self.op, attr, getattr(old_node, attr))
4189
    else:
4190
      for attr in self._NFLAGS:
4191
        if getattr(self.op, attr) is None:
4192
          setattr(self.op, attr, True)
4193

    
4194
    if self.op.readd and not self.op.vm_capable:
4195
      pri, sec = cfg.GetNodeInstances(node)
4196
      if pri or sec:
4197
        raise errors.OpPrereqError("Node %s being re-added with vm_capable"
4198
                                   " flag set to false, but it already holds"
4199
                                   " instances" % node,
4200
                                   errors.ECODE_STATE)
4201

    
4202
    # check that the type of the node (single versus dual homed) is the
4203
    # same as for the master
4204
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
4205
    master_singlehomed = myself.secondary_ip == myself.primary_ip
4206
    newbie_singlehomed = secondary_ip == primary_ip
4207
    if master_singlehomed != newbie_singlehomed:
4208
      if master_singlehomed:
4209
        raise errors.OpPrereqError("The master has no secondary ip but the"
4210
                                   " new node has one",
4211
                                   errors.ECODE_INVAL)
4212
      else:
4213
        raise errors.OpPrereqError("The master has a secondary ip but the"
4214
                                   " new node doesn't have one",
4215
                                   errors.ECODE_INVAL)
4216

    
4217
    # checks reachability
4218
    if not netutils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
4219
      raise errors.OpPrereqError("Node not reachable by ping",
4220
                                 errors.ECODE_ENVIRON)
4221

    
4222
    if not newbie_singlehomed:
4223
      # check reachability from my secondary ip to newbie's secondary ip
4224
      if not netutils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
4225
                           source=myself.secondary_ip):
4226
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
4227
                                   " based ping to node daemon port",
4228
                                   errors.ECODE_ENVIRON)
4229

    
4230
    if self.op.readd:
4231
      exceptions = [node]
4232
    else:
4233
      exceptions = []
4234

    
4235
    if self.op.master_capable:
4236
      self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
4237
    else:
4238
      self.master_candidate = False
4239

    
4240
    if self.op.readd:
4241
      self.new_node = old_node
4242
    else:
4243
      node_group = cfg.LookupNodeGroup(self.op.group)
4244
      self.new_node = objects.Node(name=node,
4245
                                   primary_ip=primary_ip,
4246
                                   secondary_ip=secondary_ip,
4247
                                   master_candidate=self.master_candidate,
4248
                                   offline=False, drained=False,
4249
                                   group=node_group)
4250

    
4251
    if self.op.ndparams:
4252
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
4253

    
4254
  def Exec(self, feedback_fn):
4255
    """Adds the new node to the cluster.
4256

4257
    """
4258
    new_node = self.new_node
4259
    node = new_node.name
4260

    
4261
    # We adding a new node so we assume it's powered
4262
    new_node.powered = True
4263

    
4264
    # for re-adds, reset the offline/drained/master-candidate flags;
4265
    # we need to reset here, otherwise offline would prevent RPC calls
4266
    # later in the procedure; this also means that if the re-add
4267
    # fails, we are left with a non-offlined, broken node
4268
    if self.op.readd:
4269
      new_node.drained = new_node.offline = False # pylint: disable-msg=W0201
4270
      self.LogInfo("Readding a node, the offline/drained flags were reset")
4271
      # if we demote the node, we do cleanup later in the procedure
4272
      new_node.master_candidate = self.master_candidate
4273
      if self.changed_primary_ip:
4274
        new_node.primary_ip = self.op.primary_ip
4275

    
4276
    # copy the master/vm_capable flags
4277
    for attr in self._NFLAGS:
4278
      setattr(new_node, attr, getattr(self.op, attr))
4279

    
4280
    # notify the user about any possible mc promotion
4281
    if new_node.master_candidate:
4282
      self.LogInfo("Node will be a master candidate")
4283

    
4284
    if self.op.ndparams:
4285
      new_node.ndparams = self.op.ndparams
4286
    else:
4287
      new_node.ndparams = {}
4288

    
4289
    # check connectivity
4290
    result = self.rpc.call_version([node])[node]
4291
    result.Raise("Can't get version information from node %s" % node)
4292
    if constants.PROTOCOL_VERSION == result.payload:
4293
      logging.info("Communication to node %s fine, sw version %s match",
4294
                   node, result.payload)
4295
    else:
4296
      raise errors.OpExecError("Version mismatch master version %s,"
4297
                               " node version %s" %
4298
                               (constants.PROTOCOL_VERSION, result.payload))
4299

    
4300
    # Add node to our /etc/hosts, and add key to known_hosts
4301
    if self.cfg.GetClusterInfo().modify_etc_hosts:
4302
      master_node = self.cfg.GetMasterNode()
4303
      result = self.rpc.call_etc_hosts_modify(master_node,
4304
                                              constants.ETC_HOSTS_ADD,
4305
                                              self.hostname.name,
4306
                                              self.hostname.ip)
4307
      result.Raise("Can't update hosts file with new host data")
4308

    
4309
    if new_node.secondary_ip != new_node.primary_ip:
4310
      _CheckNodeHasSecondaryIP(self, new_node.name, new_node.secondary_ip,
4311
                               False)
4312

    
4313
    node_verify_list = [self.cfg.GetMasterNode()]
4314
    node_verify_param = {
4315
      constants.NV_NODELIST: [node],
4316
      # TODO: do a node-net-test as well?
4317
    }
4318

    
4319
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
4320
                                       self.cfg.GetClusterName())
4321
    for verifier in node_verify_list:
4322
      result[verifier].Raise("Cannot communicate with node %s" % verifier)
4323
      nl_payload = result[verifier].payload[constants.NV_NODELIST]
4324
      if nl_payload:
4325
        for failed in nl_payload:
4326
          feedback_fn("ssh/hostname verification failed"
4327
                      " (checking from %s): %s" %
4328
                      (verifier, nl_payload[failed]))
4329
        raise errors.OpExecError("ssh/hostname verification failed")
4330

    
4331
    if self.op.readd:
4332
      _RedistributeAncillaryFiles(self)
4333
      self.context.ReaddNode(new_node)
4334
      # make sure we redistribute the config
4335
      self.cfg.Update(new_node, feedback_fn)
4336
      # and make sure the new node will not have old files around
4337
      if not new_node.master_candidate:
4338
        result = self.rpc.call_node_demote_from_mc(new_node.name)
4339
        msg = result.fail_msg
4340
        if msg:
4341
          self.LogWarning("Node failed to demote itself from master"
4342
                          " candidate status: %s" % msg)
4343
    else:
4344
      _RedistributeAncillaryFiles(self, additional_nodes=[node],
4345
                                  additional_vm=self.op.vm_capable)
4346
      self.context.AddNode(new_node, self.proc.GetECId())
4347

    
4348

    
4349
class LUNodeSetParams(LogicalUnit):
4350
  """Modifies the parameters of a node.
4351

4352
  @cvar _F2R: a dictionary from tuples of flags (mc, drained, offline)
4353
      to the node role (as _ROLE_*)
4354
  @cvar _R2F: a dictionary from node role to tuples of flags
4355
  @cvar _FLAGS: a list of attribute names corresponding to the flags
4356

4357
  """
4358
  HPATH = "node-modify"
4359
  HTYPE = constants.HTYPE_NODE
4360
  REQ_BGL = False
4361
  (_ROLE_CANDIDATE, _ROLE_DRAINED, _ROLE_OFFLINE, _ROLE_REGULAR) = range(4)
4362
  _F2R = {
4363
    (True, False, False): _ROLE_CANDIDATE,
4364
    (False, True, False): _ROLE_DRAINED,
4365
    (False, False, True): _ROLE_OFFLINE,
4366
    (False, False, False): _ROLE_REGULAR,
4367
    }
4368
  _R2F = dict((v, k) for k, v in _F2R.items())
4369
  _FLAGS = ["master_candidate", "drained", "offline"]
4370

    
4371
  def CheckArguments(self):
4372
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4373
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
4374
                self.op.master_capable, self.op.vm_capable,
4375
                self.op.secondary_ip, self.op.ndparams]
4376
    if all_mods.count(None) == len(all_mods):
4377
      raise errors.OpPrereqError("Please pass at least one modification",
4378
                                 errors.ECODE_INVAL)
4379
    if all_mods.count(True) > 1:
4380
      raise errors.OpPrereqError("Can't set the node into more than one"
4381
                                 " state at the same time",
4382
                                 errors.ECODE_INVAL)
4383

    
4384
    # Boolean value that tells us whether we might be demoting from MC
4385
    self.might_demote = (self.op.master_candidate == False or
4386
                         self.op.offline == True or
4387
                         self.op.drained == True or
4388
                         self.op.master_capable == False)
4389

    
4390
    if self.op.secondary_ip:
4391
      if not netutils.IP4Address.IsValid(self.op.secondary_ip):
4392
        raise errors.OpPrereqError("Secondary IP (%s) needs to be a valid IPv4"
4393
                                   " address" % self.op.secondary_ip,
4394
                                   errors.ECODE_INVAL)
4395

    
4396
    self.lock_all = self.op.auto_promote and self.might_demote
4397
    self.lock_instances = self.op.secondary_ip is not None
4398

    
4399
  def ExpandNames(self):
4400
    if self.lock_all:
4401
      self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
4402
    else:
4403
      self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
4404

    
4405
    if self.lock_instances:
4406
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
4407

    
4408
  def DeclareLocks(self, level):
4409
    # If we have locked all instances, before waiting to lock nodes, release
4410
    # all the ones living on nodes unrelated to the current operation.
4411
    if level == locking.LEVEL_NODE and self.lock_instances:
4412
      instances_release = []
4413
      instances_keep = []
4414
      self.affected_instances = []
4415
      if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4416
        for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
4417
          instance = self.context.cfg.GetInstanceInfo(instance_name)
4418
          i_mirrored = instance.disk_template in constants.DTS_NET_MIRROR
4419
          if i_mirrored and self.op.node_name in instance.all_nodes:
4420
            instances_keep.append(instance_name)
4421
            self.affected_instances.append(instance)
4422
          else:
4423
            instances_release.append(instance_name)
4424
        if instances_release:
4425
          self.context.glm.release(locking.LEVEL_INSTANCE, instances_release)
4426
          self.acquired_locks[locking.LEVEL_INSTANCE] = instances_keep
4427

    
4428
  def BuildHooksEnv(self):
4429
    """Build hooks env.
4430

4431
    This runs on the master node.
4432

4433
    """
4434
    env = {
4435
      "OP_TARGET": self.op.node_name,
4436
      "MASTER_CANDIDATE": str(self.op.master_candidate),
4437
      "OFFLINE": str(self.op.offline),
4438
      "DRAINED": str(self.op.drained),
4439
      "MASTER_CAPABLE": str(self.op.master_capable),
4440
      "VM_CAPABLE": str(self.op.vm_capable),
4441
      }
4442
    nl = [self.cfg.GetMasterNode(),
4443
          self.op.node_name]
4444
    return env, nl, nl
4445

    
4446
  def CheckPrereq(self):
4447
    """Check prerequisites.
4448

4449
    This only checks the instance list against the existing names.
4450

4451
    """
4452
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
4453

    
4454
    if (self.op.master_candidate is not None or
4455
        self.op.drained is not None or
4456
        self.op.offline is not None):
4457
      # we can't change the master's node flags
4458
      if self.op.node_name == self.cfg.GetMasterNode():
4459
        raise errors.OpPrereqError("The master role can be changed"
4460
                                   " only via master-failover",
4461
                                   errors.ECODE_INVAL)
4462

    
4463
    if self.op.master_candidate and not node.master_capable:
4464
      raise errors.OpPrereqError("Node %s is not master capable, cannot make"
4465
                                 " it a master candidate" % node.name,
4466
                                 errors.ECODE_STATE)
4467

    
4468
    if self.op.vm_capable == False:
4469
      (ipri, isec) = self.cfg.GetNodeInstances(self.op.node_name)
4470
      if ipri or isec:
4471
        raise errors.OpPrereqError("Node %s hosts instances, cannot unset"
4472
                                   " the vm_capable flag" % node.name,
4473
                                   errors.ECODE_STATE)
4474

    
4475
    if node.master_candidate and self.might_demote and not self.lock_all:
4476
      assert not self.op.auto_promote, "auto_promote set but lock_all not"
4477
      # check if after removing the current node, we're missing master
4478
      # candidates
4479
      (mc_remaining, mc_should, _) = \
4480
          self.cfg.GetMasterCandidateStats(exceptions=[node.name])
4481
      if mc_remaining < mc_should:
4482
        raise errors.OpPrereqError("Not enough master candidates, please"
4483
                                   " pass auto promote option to allow"
4484
                                   " promotion", errors.ECODE_STATE)
4485

    
4486
    self.old_flags = old_flags = (node.master_candidate,
4487
                                  node.drained, node.offline)
4488
    assert old_flags in self._F2R, "Un-handled old flags  %s" % str(old_flags)
4489
    self.old_role = old_role = self._F2R[old_flags]
4490

    
4491
    # Check for ineffective changes
4492
    for attr in self._FLAGS:
4493
      if (getattr(self.op, attr) == False and getattr(node, attr) == False):
4494
        self.LogInfo("Ignoring request to unset flag %s, already unset", attr)
4495
        setattr(self.op, attr, None)
4496

    
4497
    # Past this point, any flag change to False means a transition
4498
    # away from the respective state, as only real changes are kept
4499

    
4500
    # TODO: We might query the real power state if it supports OOB
4501
    if _SupportsOob(self.cfg, node):
4502
      if self.op.offline is False and not (node.powered or
4503
                                           self.op.powered == True):
4504
        raise errors.OpPrereqError(("Please power on node %s first before you"
4505
                                    " can reset offline state") %
4506
                                   self.op.node_name)
4507
    elif self.op.powered is not None:
4508
      raise errors.OpPrereqError(("Unable to change powered state for node %s"
4509
                                  " which does not support out-of-band"
4510
                                  " handling") % self.op.node_name)
4511

    
4512
    # If we're being deofflined/drained, we'll MC ourself if needed
4513
    if (self.op.drained == False or self.op.offline == False or
4514
        (self.op.master_capable and not node.master_capable)):
4515
      if _DecideSelfPromotion(self):
4516
        self.op.master_candidate = True
4517
        self.LogInfo("Auto-promoting node to master candidate")
4518

    
4519
    # If we're no longer master capable, we'll demote ourselves from MC
4520
    if self.op.master_capable == False and node.master_candidate:
4521
      self.LogInfo("Demoting from master candidate")
4522
      self.op.master_candidate = False
4523

    
4524
    # Compute new role
4525
    assert [getattr(self.op, attr) for attr in self._FLAGS].count(True) <= 1
4526
    if self.op.master_candidate:
4527
      new_role = self._ROLE_CANDIDATE
4528
    elif self.op.drained:
4529
      new_role = self._ROLE_DRAINED
4530
    elif self.op.offline:
4531
      new_role = self._ROLE_OFFLINE
4532
    elif False in [self.op.master_candidate, self.op.drained, self.op.offline]:
4533
      # False is still in new flags, which means we're un-setting (the
4534
      # only) True flag
4535
      new_role = self._ROLE_REGULAR
4536
    else: # no new flags, nothing, keep old role
4537
      new_role = old_role
4538

    
4539
    self.new_role = new_role
4540

    
4541
    if old_role == self._ROLE_OFFLINE and new_role != old_role:
4542
      # Trying to transition out of offline status
4543
      result = self.rpc.call_version([node.name])[node.name]
4544
      if result.fail_msg:
4545
        raise errors.OpPrereqError("Node %s is being de-offlined but fails"
4546
                                   " to report its version: %s" %
4547
                                   (node.name, result.fail_msg),
4548
                                   errors.ECODE_STATE)
4549
      else:
4550
        self.LogWarning("Transitioning node from offline to online state"
4551
                        " without using re-add. Please make sure the node"
4552
                        " is healthy!")
4553

    
4554
    if self.op.secondary_ip:
4555
      # Ok even without locking, because this can't be changed by any LU
4556
      master = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
4557
      master_singlehomed = master.secondary_ip == master.primary_ip
4558
      if master_singlehomed and self.op.secondary_ip:
4559
        raise errors.OpPrereqError("Cannot change the secondary ip on a single"
4560
                                   " homed cluster", errors.ECODE_INVAL)
4561

    
4562
      if node.offline:
4563
        if self.affected_instances:
4564
          raise errors.OpPrereqError("Cannot change secondary ip: offline"
4565
                                     " node has instances (%s) configured"
4566
                                     " to use it" % self.affected_instances)
4567
      else:
4568
        # On online nodes, check that no instances are running, and that
4569
        # the node has the new ip and we can reach it.
4570
        for instance in self.affected_instances:
4571
          _CheckInstanceDown(self, instance, "cannot change secondary ip")
4572

    
4573
        _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
4574
        if master.name != node.name:
4575
          # check reachability from master secondary ip to new secondary ip
4576
          if not netutils.TcpPing(self.op.secondary_ip,
4577
                                  constants.DEFAULT_NODED_PORT,
4578
                                  source=master.secondary_ip):
4579
            raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
4580
                                       " based ping to node daemon port",
4581
                                       errors.ECODE_ENVIRON)
4582

    
4583
    if self.op.ndparams:
4584
      new_ndparams = _GetUpdatedParams(self.node.ndparams, self.op.ndparams)
4585
      utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
4586
      self.new_ndparams = new_ndparams
4587

    
4588
  def Exec(self, feedback_fn):
4589
    """Modifies a node.
4590

4591
    """
4592
    node = self.node
4593
    old_role = self.old_role
4594
    new_role = self.new_role
4595

    
4596
    result = []
4597

    
4598
    if self.op.ndparams:
4599
      node.ndparams = self.new_ndparams
4600

    
4601
    if self.op.powered is not None:
4602
      node.powered = self.op.powered
4603

    
4604
    for attr in ["master_capable", "vm_capable"]:
4605
      val = getattr(self.op, attr)
4606
      if val is not None:
4607
        setattr(node, attr, val)
4608
        result.append((attr, str(val)))
4609

    
4610
    if new_role != old_role:
4611
      # Tell the node to demote itself, if no longer MC and not offline
4612
      if old_role == self._ROLE_CANDIDATE and new_role != self._ROLE_OFFLINE:
4613
        msg = self.rpc.call_node_demote_from_mc(node.name).fail_msg
4614
        if msg:
4615
          self.LogWarning("Node failed to demote itself: %s", msg)
4616

    
4617
      new_flags = self._R2F[new_role]
4618
      for of, nf, desc in zip(self.old_flags, new_flags, self._FLAGS):
4619
        if of != nf:
4620
          result.append((desc, str(nf)))
4621
      (node.master_candidate, node.drained, node.offline) = new_flags
4622

    
4623
      # we locked all nodes, we adjust the CP before updating this node
4624
      if self.lock_all:
4625
        _AdjustCandidatePool(self, [node.name])
4626

    
4627
    if self.op.secondary_ip:
4628
      node.secondary_ip = self.op.secondary_ip
4629
      result.append(("secondary_ip", self.op.secondary_ip))
4630

    
4631
    # this will trigger configuration file update, if needed
4632
    self.cfg.Update(node, feedback_fn)
4633

    
4634
    # this will trigger job queue propagation or cleanup if the mc
4635
    # flag changed
4636
    if [old_role, new_role].count(self._ROLE_CANDIDATE) == 1:
4637
      self.context.ReaddNode(node)
4638

    
4639
    return result
4640

    
4641

    
4642
class LUNodePowercycle(NoHooksLU):
4643
  """Powercycles a node.
4644

4645
  """
4646
  REQ_BGL = False
4647

    
4648
  def CheckArguments(self):
4649
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
4650
    if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
4651
      raise errors.OpPrereqError("The node is the master and the force"
4652
                                 " parameter was not set",
4653
                                 errors.ECODE_INVAL)
4654

    
4655
  def ExpandNames(self):
4656
    """Locking for PowercycleNode.
4657

4658
    This is a last-resort option and shouldn't block on other
4659
    jobs. Therefore, we grab no locks.
4660

4661
    """
4662
    self.needed_locks = {}
4663

    
4664
  def Exec(self, feedback_fn):
4665
    """Reboots a node.
4666

4667
    """
4668
    result = self.rpc.call_node_powercycle(self.op.node_name,
4669
                                           self.cfg.GetHypervisorType())
4670
    result.Raise("Failed to schedule the reboot")
4671
    return result.payload
4672

    
4673

    
4674
class LUClusterQuery(NoHooksLU):
4675
  """Query cluster configuration.
4676

4677
  """
4678
  REQ_BGL = False
4679

    
4680
  def ExpandNames(self):
4681
    self.needed_locks = {}
4682

    
4683
  def Exec(self, feedback_fn):
4684
    """Return cluster config.
4685

4686
    """
4687
    cluster = self.cfg.GetClusterInfo()
4688
    os_hvp = {}
4689

    
4690
    # Filter just for enabled hypervisors
4691
    for os_name, hv_dict in cluster.os_hvp.items():
4692
      os_hvp[os_name] = {}
4693
      for hv_name, hv_params in hv_dict.items():
4694
        if hv_name in cluster.enabled_hypervisors:
4695
          os_hvp[os_name][hv_name] = hv_params
4696

    
4697
    # Convert ip_family to ip_version
4698
    primary_ip_version = constants.IP4_VERSION
4699
    if cluster.primary_ip_family == netutils.IP6Address.family:
4700
      primary_ip_version = constants.IP6_VERSION
4701

    
4702
    result = {
4703
      "software_version": constants.RELEASE_VERSION,
4704
      "protocol_version": constants.PROTOCOL_VERSION,
4705
      "config_version": constants.CONFIG_VERSION,
4706
      "os_api_version": max(constants.OS_API_VERSIONS),
4707
      "export_version": constants.EXPORT_VERSION,
4708
      "architecture": (platform.architecture()[0], platform.machine()),
4709
      "name": cluster.cluster_name,
4710
      "master": cluster.master_node,
4711
      "default_hypervisor": cluster.enabled_hypervisors[0],
4712
      "enabled_hypervisors": cluster.enabled_hypervisors,
4713
      "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
4714
                        for hypervisor_name in cluster.enabled_hypervisors]),
4715
      "os_hvp": os_hvp,
4716
      "beparams": cluster.beparams,
4717
      "osparams": cluster.osparams,
4718
      "nicparams": cluster.nicparams,
4719
      "ndparams": cluster.ndparams,
4720
      "candidate_pool_size": cluster.candidate_pool_size,
4721
      "master_netdev": cluster.master_netdev,
4722
      "volume_group_name": cluster.volume_group_name,
4723
      "drbd_usermode_helper": cluster.drbd_usermode_helper,
4724
      "file_storage_dir": cluster.file_storage_dir,
4725
      "maintain_node_health": cluster.maintain_node_health,
4726
      "ctime": cluster.ctime,
4727
      "mtime": cluster.mtime,
4728
      "uuid": cluster.uuid,
4729
      "tags": list(cluster.GetTags()),
4730
      "uid_pool": cluster.uid_pool,
4731
      "default_iallocator": cluster.default_iallocator,
4732
      "reserved_lvs": cluster.reserved_lvs,
4733
      "primary_ip_version": primary_ip_version,
4734
      "prealloc_wipe_disks": cluster.prealloc_wipe_disks,
4735
      "hidden_os": cluster.hidden_os,
4736
      "blacklisted_os": cluster.blacklisted_os,
4737
      }
4738

    
4739
    return result
4740

    
4741

    
4742
class LUClusterConfigQuery(NoHooksLU):
4743
  """Return configuration values.
4744

4745
  """
4746
  REQ_BGL = False
4747
  _FIELDS_DYNAMIC = utils.FieldSet()
4748
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag",
4749
                                  "watcher_pause", "volume_group_name")
4750

    
4751
  def CheckArguments(self):
4752
    _CheckOutputFields(static=self._FIELDS_STATIC,
4753
                       dynamic=self._FIELDS_DYNAMIC,
4754
                       selected=self.op.output_fields)
4755

    
4756
  def ExpandNames(self):
4757
    self.needed_locks = {}
4758

    
4759
  def Exec(self, feedback_fn):
4760
    """Dump a representation of the cluster config to the standard output.
4761

4762
    """
4763
    values = []
4764
    for field in self.op.output_fields:
4765
      if field == "cluster_name":
4766
        entry = self.cfg.GetClusterName()
4767
      elif field == "master_node":
4768
        entry = self.cfg.GetMasterNode()
4769
      elif field == "drain_flag":
4770
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
4771
      elif field == "watcher_pause":
4772
        entry = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
4773
      elif field == "volume_group_name":
4774
        entry = self.cfg.GetVGName()
4775
      else:
4776
        raise errors.ParameterError(field)
4777
      values.append(entry)
4778
    return values
4779

    
4780

    
4781
class LUInstanceActivateDisks(NoHooksLU):
4782
  """Bring up an instance's disks.
4783

4784
  """
4785
  REQ_BGL = False
4786

    
4787
  def ExpandNames(self):
4788
    self._ExpandAndLockInstance()
4789
    self.needed_locks[locking.LEVEL_NODE] = []
4790
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4791

    
4792
  def DeclareLocks(self, level):
4793
    if level == locking.LEVEL_NODE:
4794
      self._LockInstancesNodes()
4795

    
4796
  def CheckPrereq(self):
4797
    """Check prerequisites.
4798

4799
    This checks that the instance is in the cluster.
4800

4801
    """
4802
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4803
    assert self.instance is not None, \
4804
      "Cannot retrieve locked instance %s" % self.op.instance_name
4805
    _CheckNodeOnline(self, self.instance.primary_node)
4806

    
4807
  def Exec(self, feedback_fn):
4808
    """Activate the disks.
4809

4810
    """
4811
    disks_ok, disks_info = \
4812
              _AssembleInstanceDisks(self, self.instance,
4813
                                     ignore_size=self.op.ignore_size)
4814
    if not disks_ok:
4815
      raise errors.OpExecError("Cannot activate block devices")
4816

    
4817
    return disks_info
4818

    
4819

    
4820
def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
4821
                           ignore_size=False):
4822
  """Prepare the block devices for an instance.
4823

4824
  This sets up the block devices on all nodes.
4825

4826
  @type lu: L{LogicalUnit}
4827
  @param lu: the logical unit on whose behalf we execute
4828
  @type instance: L{objects.Instance}
4829
  @param instance: the instance for whose disks we assemble
4830
  @type disks: list of L{objects.Disk} or None
4831
  @param disks: which disks to assemble (or all, if None)
4832
  @type ignore_secondaries: boolean
4833
  @param ignore_secondaries: if true, errors on secondary nodes
4834
      won't result in an error return from the function
4835
  @type ignore_size: boolean
4836
  @param ignore_size: if true, the current known size of the disk
4837
      will not be used during the disk activation, useful for cases
4838
      when the size is wrong
4839
  @return: False if the operation failed, otherwise a list of
4840
      (host, instance_visible_name, node_visible_name)
4841
      with the mapping from node devices to instance devices
4842

4843
  """
4844
  device_info = []
4845
  disks_ok = True
4846
  iname = instance.name
4847
  disks = _ExpandCheckDisks(instance, disks)
4848

    
4849
  # With the two passes mechanism we try to reduce the window of
4850
  # opportunity for the race condition of switching DRBD to primary
4851
  # before handshaking occured, but we do not eliminate it
4852

    
4853
  # The proper fix would be to wait (with some limits) until the
4854
  # connection has been made and drbd transitions from WFConnection
4855
  # into any other network-connected state (Connected, SyncTarget,
4856
  # SyncSource, etc.)
4857

    
4858
  # 1st pass, assemble on all nodes in secondary mode
4859
  for idx, inst_disk in enumerate(disks):
4860
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
4861
      if ignore_size:
4862
        node_disk = node_disk.Copy()
4863
        node_disk.UnsetSize()
4864
      lu.cfg.SetDiskID(node_disk, node)
4865
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False, idx)
4866
      msg = result.fail_msg
4867
      if msg:
4868
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
4869
                           " (is_primary=False, pass=1): %s",
4870
                           inst_disk.iv_name, node, msg)
4871
        if not ignore_secondaries:
4872
          disks_ok = False
4873

    
4874
  # FIXME: race condition on drbd migration to primary
4875

    
4876
  # 2nd pass, do only the primary node
4877
  for idx, inst_disk in enumerate(disks):
4878
    dev_path = None
4879

    
4880
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
4881
      if node != instance.primary_node:
4882
        continue
4883
      if ignore_size:
4884
        node_disk = node_disk.Copy()
4885
        node_disk.UnsetSize()
4886
      lu.cfg.SetDiskID(node_disk, node)
4887
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True, idx)
4888
      msg = result.fail_msg
4889
      if msg:
4890
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
4891
                           " (is_primary=True, pass=2): %s",
4892
                           inst_disk.iv_name, node, msg)
4893
        disks_ok = False
4894
      else:
4895
        dev_path = result.payload
4896

    
4897
    device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
4898

    
4899
  # leave the disks configured for the primary node
4900
  # this is a workaround that would be fixed better by
4901
  # improving the logical/physical id handling
4902
  for disk in disks:
4903
    lu.cfg.SetDiskID(disk, instance.primary_node)
4904

    
4905
  return disks_ok, device_info
4906

    
4907

    
4908
def _StartInstanceDisks(lu, instance, force):
4909
  """Start the disks of an instance.
4910

4911
  """
4912
  disks_ok, _ = _AssembleInstanceDisks(lu, instance,
4913
                                           ignore_secondaries=force)
4914
  if not disks_ok:
4915
    _ShutdownInstanceDisks(lu, instance)
4916
    if force is not None and not force:
4917
      lu.proc.LogWarning("", hint="If the message above refers to a"
4918
                         " secondary node,"
4919
                         " you can retry the operation using '--force'.")
4920
    raise errors.OpExecError("Disk consistency error")
4921

    
4922

    
4923
class LUInstanceDeactivateDisks(NoHooksLU):
4924
  """Shutdown an instance's disks.
4925

4926
  """
4927
  REQ_BGL = False
4928

    
4929
  def ExpandNames(self):
4930
    self._ExpandAndLockInstance()
4931
    self.needed_locks[locking.LEVEL_NODE] = []
4932
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4933

    
4934
  def DeclareLocks(self, level):
4935
    if level == locking.LEVEL_NODE:
4936
      self._LockInstancesNodes()
4937

    
4938
  def CheckPrereq(self):
4939
    """Check prerequisites.
4940

4941
    This checks that the instance is in the cluster.
4942

4943
    """
4944
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4945
    assert self.instance is not None, \
4946
      "Cannot retrieve locked instance %s" % self.op.instance_name
4947

    
4948
  def Exec(self, feedback_fn):
4949
    """Deactivate the disks
4950

4951
    """
4952
    instance = self.instance
4953
    if self.op.force:
4954
      _ShutdownInstanceDisks(self, instance)
4955
    else:
4956
      _SafeShutdownInstanceDisks(self, instance)
4957

    
4958

    
4959
def _SafeShutdownInstanceDisks(lu, instance, disks=None):
4960
  """Shutdown block devices of an instance.
4961

4962
  This function checks if an instance is running, before calling
4963
  _ShutdownInstanceDisks.
4964

4965
  """
4966
  _CheckInstanceDown(lu, instance, "cannot shutdown disks")
4967
  _ShutdownInstanceDisks(lu, instance, disks=disks)
4968

    
4969

    
4970
def _ExpandCheckDisks(instance, disks):
4971
  """Return the instance disks selected by the disks list
4972

4973
  @type disks: list of L{objects.Disk} or None
4974
  @param disks: selected disks
4975
  @rtype: list of L{objects.Disk}
4976
  @return: selected instance disks to act on
4977

4978
  """
4979
  if disks is None:
4980
    return instance.disks
4981
  else:
4982
    if not set(disks).issubset(instance.disks):
4983
      raise errors.ProgrammerError("Can only act on disks belonging to the"
4984
                                   " target instance")
4985
    return disks
4986

    
4987

    
4988
def _ShutdownInstanceDisks(lu, instance, disks=None, ignore_primary=False):
4989
  """Shutdown block devices of an instance.
4990

4991
  This does the shutdown on all nodes of the instance.
4992

4993
  If the ignore_primary is false, errors on the primary node are
4994
  ignored.
4995

4996
  """
4997
  all_result = True
4998
  disks = _ExpandCheckDisks(instance, disks)
4999

    
5000
  for disk in disks:
5001
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
5002
      lu.cfg.SetDiskID(top_disk, node)
5003
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
5004
      msg = result.fail_msg
5005
      if msg:
5006
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
5007
                      disk.iv_name, node, msg)
5008
        if ((node == instance.primary_node and not ignore_primary) or
5009
            (node != instance.primary_node and not result.offline)):
5010
          all_result = False
5011
  return all_result
5012

    
5013

    
5014
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
5015
  """Checks if a node has enough free memory.
5016

5017
  This function check if a given node has the needed amount of free
5018
  memory. In case the node has less memory or we cannot get the
5019
  information from the node, this function raise an OpPrereqError
5020
  exception.
5021

5022
  @type lu: C{LogicalUnit}
5023
  @param lu: a logical unit from which we get configuration data
5024
  @type node: C{str}
5025
  @param node: the node to check
5026
  @type reason: C{str}
5027
  @param reason: string to use in the error message
5028
  @type requested: C{int}
5029
  @param requested: the amount of memory in MiB to check for
5030
  @type hypervisor_name: C{str}
5031
  @param hypervisor_name: the hypervisor to ask for memory stats
5032
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
5033
      we cannot check the node
5034

5035
  """
5036
  nodeinfo = lu.rpc.call_node_info([node], None, hypervisor_name)
5037
  nodeinfo[node].Raise("Can't get data from node %s" % node,
5038
                       prereq=True, ecode=errors.ECODE_ENVIRON)
5039
  free_mem = nodeinfo[node].payload.get('memory_free', None)
5040
  if not isinstance(free_mem, int):
5041
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
5042
                               " was '%s'" % (node, free_mem),
5043
                               errors.ECODE_ENVIRON)
5044
  if requested > free_mem:
5045
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
5046
                               " needed %s MiB, available %s MiB" %
5047
                               (node, reason, requested, free_mem),
5048
                               errors.ECODE_NORES)
5049

    
5050

    
5051
def _CheckNodesFreeDiskPerVG(lu, nodenames, req_sizes):
5052
  """Checks if nodes have enough free disk space in the all VGs.
5053

5054
  This function check if all given nodes have the needed amount of
5055
  free disk. In case any node has less disk or we cannot get the
5056
  information from the node, this function raise an OpPrereqError
5057
  exception.
5058

5059
  @type lu: C{LogicalUnit}
5060
  @param lu: a logical unit from which we get configuration data
5061
  @type nodenames: C{list}
5062
  @param nodenames: the list of node names to check
5063
  @type req_sizes: C{dict}
5064
  @param req_sizes: the hash of vg and corresponding amount of disk in
5065
      MiB to check for
5066
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5067
      or we cannot check the node
5068

5069
  """
5070
  for vg, req_size in req_sizes.items():
5071
    _CheckNodesFreeDiskOnVG(lu, nodenames, vg, req_size)
5072

    
5073

    
5074
def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
5075
  """Checks if nodes have enough free disk space in the specified VG.
5076

5077
  This function check if all given nodes have the needed amount of
5078
  free disk. In case any node has less disk or we cannot get the
5079
  information from the node, this function raise an OpPrereqError
5080
  exception.
5081

5082
  @type lu: C{LogicalUnit}
5083
  @param lu: a logical unit from which we get configuration data
5084
  @type nodenames: C{list}
5085
  @param nodenames: the list of node names to check
5086
  @type vg: C{str}
5087
  @param vg: the volume group to check
5088
  @type requested: C{int}
5089
  @param requested: the amount of disk in MiB to check for
5090
  @raise errors.OpPrereqError: if the node doesn't have enough disk,
5091
      or we cannot check the node
5092

5093
  """
5094
  nodeinfo = lu.rpc.call_node_info(nodenames, vg, None)
5095
  for node in nodenames:
5096
    info = nodeinfo[node]
5097
    info.Raise("Cannot get current information from node %s" % node,
5098
               prereq=True, ecode=errors.ECODE_ENVIRON)
5099
    vg_free = info.payload.get("vg_free", None)
5100
    if not isinstance(vg_free, int):
5101
      raise errors.OpPrereqError("Can't compute free disk space on node"
5102
                                 " %s for vg %s, result was '%s'" %
5103
                                 (node, vg, vg_free), errors.ECODE_ENVIRON)
5104
    if requested > vg_free:
5105
      raise errors.OpPrereqError("Not enough disk space on target node %s"
5106
                                 " vg %s: required %d MiB, available %d MiB" %
5107
                                 (node, vg, requested, vg_free),
5108
                                 errors.ECODE_NORES)
5109

    
5110

    
5111
class LUInstanceStartup(LogicalUnit):
5112
  """Starts an instance.
5113

5114
  """
5115
  HPATH = "instance-start"
5116
  HTYPE = constants.HTYPE_INSTANCE
5117
  REQ_BGL = False
5118

    
5119
  def CheckArguments(self):
5120
    # extra beparams
5121
    if self.op.beparams:
5122
      # fill the beparams dict
5123
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
5124

    
5125
  def ExpandNames(self):
5126
    self._ExpandAndLockInstance()
5127

    
5128
  def BuildHooksEnv(self):
5129
    """Build hooks env.
5130

5131
    This runs on master, primary and secondary nodes of the instance.
5132

5133
    """
5134
    env = {
5135
      "FORCE": self.op.force,
5136
      }
5137
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5138
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5139
    return env, nl, nl
5140

    
5141
  def CheckPrereq(self):
5142
    """Check prerequisites.
5143

5144
    This checks that the instance is in the cluster.
5145

5146
    """
5147
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5148
    assert self.instance is not None, \
5149
      "Cannot retrieve locked instance %s" % self.op.instance_name
5150

    
5151
    # extra hvparams
5152
    if self.op.hvparams:
5153
      # check hypervisor parameter syntax (locally)
5154
      cluster = self.cfg.GetClusterInfo()
5155
      utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
5156
      filled_hvp = cluster.FillHV(instance)
5157
      filled_hvp.update(self.op.hvparams)
5158
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
5159
      hv_type.CheckParameterSyntax(filled_hvp)
5160
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
5161

    
5162
    self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
5163

    
5164
    if self.primary_offline and self.op.ignore_offline_nodes:
5165
      self.proc.LogWarning("Ignoring offline primary node")
5166

    
5167
      if self.op.hvparams or self.op.beparams:
5168
        self.proc.LogWarning("Overridden parameters are ignored")
5169
    else:
5170
      _CheckNodeOnline(self, instance.primary_node)
5171

    
5172
      bep = self.cfg.GetClusterInfo().FillBE(instance)
5173

    
5174
      # check bridges existence
5175
      _CheckInstanceBridgesExist(self, instance)
5176

    
5177
      remote_info = self.rpc.call_instance_info(instance.primary_node,
5178
                                                instance.name,
5179
                                                instance.hypervisor)
5180
      remote_info.Raise("Error checking node %s" % instance.primary_node,
5181
                        prereq=True, ecode=errors.ECODE_ENVIRON)
5182
      if not remote_info.payload: # not running already
5183
        _CheckNodeFreeMemory(self, instance.primary_node,
5184
                             "starting instance %s" % instance.name,
5185
                             bep[constants.BE_MEMORY], instance.hypervisor)
5186

    
5187
  def Exec(self, feedback_fn):
5188
    """Start the instance.
5189

5190
    """
5191
    instance = self.instance
5192
    force = self.op.force
5193

    
5194
    self.cfg.MarkInstanceUp(instance.name)
5195

    
5196
    if self.primary_offline:
5197
      assert self.op.ignore_offline_nodes
5198
      self.proc.LogInfo("Primary node offline, marked instance as started")
5199
    else:
5200
      node_current = instance.primary_node
5201

    
5202
      _StartInstanceDisks(self, instance, force)
5203

    
5204
      result = self.rpc.call_instance_start(node_current, instance,
5205
                                            self.op.hvparams, self.op.beparams)
5206
      msg = result.fail_msg
5207
      if msg:
5208
        _ShutdownInstanceDisks(self, instance)
5209
        raise errors.OpExecError("Could not start instance: %s" % msg)
5210

    
5211

    
5212
class LUInstanceReboot(LogicalUnit):
5213
  """Reboot an instance.
5214

5215
  """
5216
  HPATH = "instance-reboot"
5217
  HTYPE = constants.HTYPE_INSTANCE
5218
  REQ_BGL = False
5219

    
5220
  def ExpandNames(self):
5221
    self._ExpandAndLockInstance()
5222

    
5223
  def BuildHooksEnv(self):
5224
    """Build hooks env.
5225

5226
    This runs on master, primary and secondary nodes of the instance.
5227

5228
    """
5229
    env = {
5230
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
5231
      "REBOOT_TYPE": self.op.reboot_type,
5232
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5233
      }
5234
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5235
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5236
    return env, nl, nl
5237

    
5238
  def CheckPrereq(self):
5239
    """Check prerequisites.
5240

5241
    This checks that the instance is in the cluster.
5242

5243
    """
5244
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5245
    assert self.instance is not None, \
5246
      "Cannot retrieve locked instance %s" % self.op.instance_name
5247

    
5248
    _CheckNodeOnline(self, instance.primary_node)
5249

    
5250
    # check bridges existence
5251
    _CheckInstanceBridgesExist(self, instance)
5252

    
5253
  def Exec(self, feedback_fn):
5254
    """Reboot the instance.
5255

5256
    """
5257
    instance = self.instance
5258
    ignore_secondaries = self.op.ignore_secondaries
5259
    reboot_type = self.op.reboot_type
5260

    
5261
    node_current = instance.primary_node
5262

    
5263
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
5264
                       constants.INSTANCE_REBOOT_HARD]:
5265
      for disk in instance.disks:
5266
        self.cfg.SetDiskID(disk, node_current)
5267
      result = self.rpc.call_instance_reboot(node_current, instance,
5268
                                             reboot_type,
5269
                                             self.op.shutdown_timeout)
5270
      result.Raise("Could not reboot instance")
5271
    else:
5272
      result = self.rpc.call_instance_shutdown(node_current, instance,
5273
                                               self.op.shutdown_timeout)
5274
      result.Raise("Could not shutdown instance for full reboot")
5275
      _ShutdownInstanceDisks(self, instance)
5276
      _StartInstanceDisks(self, instance, ignore_secondaries)
5277
      result = self.rpc.call_instance_start(node_current, instance, None, None)
5278
      msg = result.fail_msg
5279
      if msg:
5280
        _ShutdownInstanceDisks(self, instance)
5281
        raise errors.OpExecError("Could not start instance for"
5282
                                 " full reboot: %s" % msg)
5283

    
5284
    self.cfg.MarkInstanceUp(instance.name)
5285

    
5286

    
5287
class LUInstanceShutdown(LogicalUnit):
5288
  """Shutdown an instance.
5289

5290
  """
5291
  HPATH = "instance-stop"
5292
  HTYPE = constants.HTYPE_INSTANCE
5293
  REQ_BGL = False
5294

    
5295
  def ExpandNames(self):
5296
    self._ExpandAndLockInstance()
5297

    
5298
  def BuildHooksEnv(self):
5299
    """Build hooks env.
5300

5301
    This runs on master, primary and secondary nodes of the instance.
5302

5303
    """
5304
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5305
    env["TIMEOUT"] = self.op.timeout
5306
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5307
    return env, nl, nl
5308

    
5309
  def CheckPrereq(self):
5310
    """Check prerequisites.
5311

5312
    This checks that the instance is in the cluster.
5313

5314
    """
5315
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5316
    assert self.instance is not None, \
5317
      "Cannot retrieve locked instance %s" % self.op.instance_name
5318

    
5319
    self.primary_offline = \
5320
      self.cfg.GetNodeInfo(self.instance.primary_node).offline
5321

    
5322
    if self.primary_offline and self.op.ignore_offline_nodes:
5323
      self.proc.LogWarning("Ignoring offline primary node")
5324
    else:
5325
      _CheckNodeOnline(self, self.instance.primary_node)
5326

    
5327
  def Exec(self, feedback_fn):
5328
    """Shutdown the instance.
5329

5330
    """
5331
    instance = self.instance
5332
    node_current = instance.primary_node
5333
    timeout = self.op.timeout
5334

    
5335
    self.cfg.MarkInstanceDown(instance.name)
5336

    
5337
    if self.primary_offline:
5338
      assert self.op.ignore_offline_nodes
5339
      self.proc.LogInfo("Primary node offline, marked instance as stopped")
5340
    else:
5341
      result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
5342
      msg = result.fail_msg
5343
      if msg:
5344
        self.proc.LogWarning("Could not shutdown instance: %s" % msg)
5345

    
5346
      _ShutdownInstanceDisks(self, instance)
5347

    
5348

    
5349
class LUInstanceReinstall(LogicalUnit):
5350
  """Reinstall an instance.
5351

5352
  """
5353
  HPATH = "instance-reinstall"
5354
  HTYPE = constants.HTYPE_INSTANCE
5355
  REQ_BGL = False
5356

    
5357
  def ExpandNames(self):
5358
    self._ExpandAndLockInstance()
5359

    
5360
  def BuildHooksEnv(self):
5361
    """Build hooks env.
5362

5363
    This runs on master, primary and secondary nodes of the instance.
5364

5365
    """
5366
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5367
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5368
    return env, nl, nl
5369

    
5370
  def CheckPrereq(self):
5371
    """Check prerequisites.
5372

5373
    This checks that the instance is in the cluster and is not running.
5374

5375
    """
5376
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5377
    assert instance is not None, \
5378
      "Cannot retrieve locked instance %s" % self.op.instance_name
5379
    _CheckNodeOnline(self, instance.primary_node, "Instance primary node"
5380
                     " offline, cannot reinstall")
5381
    for node in instance.secondary_nodes:
5382
      _CheckNodeOnline(self, node, "Instance secondary node offline,"
5383
                       " cannot reinstall")
5384

    
5385
    if instance.disk_template == constants.DT_DISKLESS:
5386
      raise errors.OpPrereqError("Instance '%s' has no disks" %
5387
                                 self.op.instance_name,
5388
                                 errors.ECODE_INVAL)
5389
    _CheckInstanceDown(self, instance, "cannot reinstall")
5390

    
5391
    if self.op.os_type is not None:
5392
      # OS verification
5393
      pnode = _ExpandNodeName(self.cfg, instance.primary_node)
5394
      _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
5395
      instance_os = self.op.os_type
5396
    else:
5397
      instance_os = instance.os
5398

    
5399
    nodelist = list(instance.all_nodes)
5400

    
5401
    if self.op.osparams:
5402
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
5403
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
5404
      self.os_inst = i_osdict # the new dict (without defaults)
5405
    else:
5406
      self.os_inst = None
5407

    
5408
    self.instance = instance
5409

    
5410
  def Exec(self, feedback_fn):
5411
    """Reinstall the instance.
5412

5413
    """
5414
    inst = self.instance
5415

    
5416
    if self.op.os_type is not None:
5417
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
5418
      inst.os = self.op.os_type
5419
      # Write to configuration
5420
      self.cfg.Update(inst, feedback_fn)
5421

    
5422
    _StartInstanceDisks(self, inst, None)
5423
    try:
5424
      feedback_fn("Running the instance OS create scripts...")
5425
      # FIXME: pass debug option from opcode to backend
5426
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True,
5427
                                             self.op.debug_level,
5428
                                             osparams=self.os_inst)
5429
      result.Raise("Could not install OS for instance %s on node %s" %
5430
                   (inst.name, inst.primary_node))
5431
    finally:
5432
      _ShutdownInstanceDisks(self, inst)
5433

    
5434

    
5435
class LUInstanceRecreateDisks(LogicalUnit):
5436
  """Recreate an instance's missing disks.
5437

5438
  """
5439
  HPATH = "instance-recreate-disks"
5440
  HTYPE = constants.HTYPE_INSTANCE
5441
  REQ_BGL = False
5442

    
5443
  def CheckArguments(self):
5444
    # normalise the disk list
5445
    self.op.disks = sorted(frozenset(self.op.disks))
5446

    
5447
  def ExpandNames(self):
5448
    self._ExpandAndLockInstance()
5449
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5450
    if self.op.nodes:
5451
      self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
5452
      self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
5453
    else:
5454
      self.needed_locks[locking.LEVEL_NODE] = []
5455

    
5456
  def DeclareLocks(self, level):
5457
    if level == locking.LEVEL_NODE:
5458
      # if we replace the nodes, we only need to lock the old primary,
5459
      # otherwise we need to lock all nodes for disk re-creation
5460
      primary_only = bool(self.op.nodes)
5461
      self._LockInstancesNodes(primary_only=primary_only)
5462

    
5463
  def BuildHooksEnv(self):
5464
    """Build hooks env.
5465

5466
    This runs on master, primary and secondary nodes of the instance.
5467

5468
    """
5469
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5470
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5471
    return env, nl, nl
5472

    
5473
  def CheckPrereq(self):
5474
    """Check prerequisites.
5475

5476
    This checks that the instance is in the cluster and is not running.
5477

5478
    """
5479
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5480
    assert instance is not None, \
5481
      "Cannot retrieve locked instance %s" % self.op.instance_name
5482
    if self.op.nodes:
5483
      if len(self.op.nodes) != len(instance.all_nodes):
5484
        raise errors.OpPrereqError("Instance %s currently has %d nodes, but"
5485
                                   " %d replacement nodes were specified" %
5486
                                   (instance.name, len(instance.all_nodes),
5487
                                    len(self.op.nodes)),
5488
                                   errors.ECODE_INVAL)
5489
      assert instance.disk_template != constants.DT_DRBD8 or \
5490
          len(self.op.nodes) == 2
5491
      assert instance.disk_template != constants.DT_PLAIN or \
5492
          len(self.op.nodes) == 1
5493
      primary_node = self.op.nodes[0]
5494
    else:
5495
      primary_node = instance.primary_node
5496
    _CheckNodeOnline(self, primary_node)
5497

    
5498
    if instance.disk_template == constants.DT_DISKLESS:
5499
      raise errors.OpPrereqError("Instance '%s' has no disks" %
5500
                                 self.op.instance_name, errors.ECODE_INVAL)
5501
    # if we replace nodes *and* the old primary is offline, we don't
5502
    # check
5503
    assert instance.primary_node in self.needed_locks[locking.LEVEL_NODE]
5504
    old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
5505
    if not (self.op.nodes and old_pnode.offline):
5506
      _CheckInstanceDown(self, instance, "cannot recreate disks")
5507

    
5508
    if not self.op.disks:
5509
      self.op.disks = range(len(instance.disks))
5510
    else:
5511
      for idx in self.op.disks:
5512
        if idx >= len(instance.disks):
5513
          raise errors.OpPrereqError("Invalid disk index passed '%s'" % idx,
5514
                                     errors.ECODE_INVAL)
5515
    if self.op.disks != range(len(instance.disks)) and self.op.nodes:
5516
      raise errors.OpPrereqError("Can't recreate disks partially and"
5517
                                 " change the nodes at the same time",
5518
                                 errors.ECODE_INVAL)
5519
    self.instance = instance
5520

    
5521
  def Exec(self, feedback_fn):
5522
    """Recreate the disks.
5523

5524
    """
5525
    # change primary node, if needed
5526
    if self.op.nodes:
5527
      self.instance.primary_node = self.op.nodes[0]
5528
      self.LogWarning("Changing the instance's nodes, you will have to"
5529
                      " remove any disks left on the older nodes manually")
5530

    
5531
    to_skip = []
5532
    for idx, disk in enumerate(self.instance.disks):
5533
      if idx not in self.op.disks: # disk idx has not been passed in
5534
        to_skip.append(idx)
5535
        continue
5536
      # update secondaries for disks, if needed
5537
      if self.op.nodes:
5538
        if disk.dev_type == constants.LD_DRBD8:
5539
          # need to update the nodes
5540
          assert len(self.op.nodes) == 2
5541
          logical_id = list(disk.logical_id)
5542
          logical_id[0] = self.op.nodes[0]
5543
          logical_id[1] = self.op.nodes[1]
5544
          disk.logical_id = tuple(logical_id)
5545

    
5546
    if self.op.nodes:
5547
      self.cfg.Update(self.instance, feedback_fn)
5548

    
5549
    _CreateDisks(self, self.instance, to_skip=to_skip)
5550

    
5551

    
5552
class LUInstanceRename(LogicalUnit):
5553
  """Rename an instance.
5554

5555
  """
5556
  HPATH = "instance-rename"
5557
  HTYPE = constants.HTYPE_INSTANCE
5558

    
5559
  def CheckArguments(self):
5560
    """Check arguments.
5561

5562
    """
5563
    if self.op.ip_check and not self.op.name_check:
5564
      # TODO: make the ip check more flexible and not depend on the name check
5565
      raise errors.OpPrereqError("Cannot do ip check without a name check",
5566
                                 errors.ECODE_INVAL)
5567

    
5568
  def BuildHooksEnv(self):
5569
    """Build hooks env.
5570

5571
    This runs on master, primary and secondary nodes of the instance.
5572

5573
    """
5574
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5575
    env["INSTANCE_NEW_NAME"] = self.op.new_name
5576
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5577
    return env, nl, nl
5578

    
5579
  def CheckPrereq(self):
5580
    """Check prerequisites.
5581

5582
    This checks that the instance is in the cluster and is not running.
5583

5584
    """
5585
    self.op.instance_name = _ExpandInstanceName(self.cfg,
5586
                                                self.op.instance_name)
5587
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5588
    assert instance is not None
5589
    _CheckNodeOnline(self, instance.primary_node)
5590
    _CheckInstanceDown(self, instance, "cannot rename")
5591
    self.instance = instance
5592

    
5593
    new_name = self.op.new_name
5594
    if self.op.name_check:
5595
      hostname = netutils.GetHostname(name=new_name)
5596
      if hostname != new_name:
5597
        self.LogInfo("Resolved given name '%s' to '%s'", new_name,
5598
                     hostname.name)
5599
      new_name = self.op.new_name = hostname.name
5600
      if (self.op.ip_check and
5601
          netutils.TcpPing(hostname.ip, constants.DEFAULT_NODED_PORT)):
5602
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
5603
                                   (hostname.ip, new_name),
5604
                                   errors.ECODE_NOTUNIQUE)
5605

    
5606
    instance_list = self.cfg.GetInstanceList()
5607
    if new_name in instance_list and new_name != instance.name:
5608
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
5609
                                 new_name, errors.ECODE_EXISTS)
5610

    
5611
  def Exec(self, feedback_fn):
5612
    """Rename the instance.
5613

5614
    """
5615
    inst = self.instance
5616
    old_name = inst.name
5617

    
5618
    rename_file_storage = False
5619
    if (inst.disk_template == constants.DT_FILE and
5620
        self.op.new_name != inst.name):
5621
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
5622
      rename_file_storage = True
5623

    
5624
    self.cfg.RenameInstance(inst.name, self.op.new_name)
5625
    # Change the instance lock. This is definitely safe while we hold the BGL
5626
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
5627
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
5628

    
5629
    # re-read the instance from the configuration after rename
5630
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
5631

    
5632
    if rename_file_storage:
5633
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
5634
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
5635
                                                     old_file_storage_dir,
5636
                                                     new_file_storage_dir)
5637
      result.Raise("Could not rename on node %s directory '%s' to '%s'"
5638
                   " (but the instance has been renamed in Ganeti)" %
5639
                   (inst.primary_node, old_file_storage_dir,
5640
                    new_file_storage_dir))
5641

    
5642
    _StartInstanceDisks(self, inst, None)
5643
    try:
5644
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
5645
                                                 old_name, self.op.debug_level)
5646
      msg = result.fail_msg
5647
      if msg:
5648
        msg = ("Could not run OS rename script for instance %s on node %s"
5649
               " (but the instance has been renamed in Ganeti): %s" %
5650
               (inst.name, inst.primary_node, msg))
5651
        self.proc.LogWarning(msg)
5652
    finally:
5653
      _ShutdownInstanceDisks(self, inst)
5654

    
5655
    return inst.name
5656

    
5657

    
5658
class LUInstanceRemove(LogicalUnit):
5659
  """Remove an instance.
5660

5661
  """
5662
  HPATH = "instance-remove"
5663
  HTYPE = constants.HTYPE_INSTANCE
5664
  REQ_BGL = False
5665

    
5666
  def ExpandNames(self):
5667
    self._ExpandAndLockInstance()
5668
    self.needed_locks[locking.LEVEL_NODE] = []
5669
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5670

    
5671
  def DeclareLocks(self, level):
5672
    if level == locking.LEVEL_NODE:
5673
      self._LockInstancesNodes()
5674

    
5675
  def BuildHooksEnv(self):
5676
    """Build hooks env.
5677

5678
    This runs on master, primary and secondary nodes of the instance.
5679

5680
    """
5681
    env = _BuildInstanceHookEnvByObject(self, self.instance)
5682
    env["SHUTDOWN_TIMEOUT"] = self.op.shutdown_timeout
5683
    nl = [self.cfg.GetMasterNode()]
5684
    nl_post = list(self.instance.all_nodes) + nl
5685
    return env, nl, nl_post
5686

    
5687
  def CheckPrereq(self):
5688
    """Check prerequisites.
5689

5690
    This checks that the instance is in the cluster.
5691

5692
    """
5693
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5694
    assert self.instance is not None, \
5695
      "Cannot retrieve locked instance %s" % self.op.instance_name
5696

    
5697
  def Exec(self, feedback_fn):
5698
    """Remove the instance.
5699

5700
    """
5701
    instance = self.instance
5702
    logging.info("Shutting down instance %s on node %s",
5703
                 instance.name, instance.primary_node)
5704

    
5705
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
5706
                                             self.op.shutdown_timeout)
5707
    msg = result.fail_msg
5708
    if msg:
5709
      if self.op.ignore_failures:
5710
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
5711
      else:
5712
        raise errors.OpExecError("Could not shutdown instance %s on"
5713
                                 " node %s: %s" %
5714
                                 (instance.name, instance.primary_node, msg))
5715

    
5716
    _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
5717

    
5718

    
5719
def _RemoveInstance(lu, feedback_fn, instance, ignore_failures):
5720
  """Utility function to remove an instance.
5721

5722
  """
5723
  logging.info("Removing block devices for instance %s", instance.name)
5724

    
5725
  if not _RemoveDisks(lu, instance):
5726
    if not ignore_failures:
5727
      raise errors.OpExecError("Can't remove instance's disks")
5728
    feedback_fn("Warning: can't remove instance's disks")
5729

    
5730
  logging.info("Removing instance %s out of cluster config", instance.name)
5731

    
5732
  lu.cfg.RemoveInstance(instance.name)
5733

    
5734
  assert not lu.remove_locks.get(locking.LEVEL_INSTANCE), \
5735
    "Instance lock removal conflict"
5736

    
5737
  # Remove lock for the instance
5738
  lu.remove_locks[locking.LEVEL_INSTANCE] = instance.name
5739

    
5740

    
5741
class LUInstanceQuery(NoHooksLU):
5742
  """Logical unit for querying instances.
5743

5744
  """
5745
  # pylint: disable-msg=W0142
5746
  REQ_BGL = False
5747

    
5748
  def CheckArguments(self):
5749
    self.iq = _InstanceQuery(self.op.names, self.op.output_fields,
5750
                             self.op.use_locking)
5751

    
5752
  def ExpandNames(self):
5753
    self.iq.ExpandNames(self)
5754

    
5755
  def DeclareLocks(self, level):
5756
    self.iq.DeclareLocks(self, level)
5757

    
5758
  def Exec(self, feedback_fn):
5759
    return self.iq.OldStyleQuery(self)
5760

    
5761

    
5762
class LUInstanceFailover(LogicalUnit):
5763
  """Failover an instance.
5764

5765
  """
5766
  HPATH = "instance-failover"
5767
  HTYPE = constants.HTYPE_INSTANCE
5768
  REQ_BGL = False
5769

    
5770
  def ExpandNames(self):
5771
    self._ExpandAndLockInstance()
5772
    self.needed_locks[locking.LEVEL_NODE] = []
5773
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5774

    
5775
  def DeclareLocks(self, level):
5776
    if level == locking.LEVEL_NODE:
5777
      self._LockInstancesNodes()
5778

    
5779
  def BuildHooksEnv(self):
5780
    """Build hooks env.
5781

5782
    This runs on master, primary and secondary nodes of the instance.
5783

5784
    """
5785
    instance = self.instance
5786
    source_node = instance.primary_node
5787
    target_node = instance.secondary_nodes[0]
5788
    env = {
5789
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
5790
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5791
      "OLD_PRIMARY": source_node,
5792
      "OLD_SECONDARY": target_node,
5793
      "NEW_PRIMARY": target_node,
5794
      "NEW_SECONDARY": source_node,
5795
      }
5796
    env.update(_BuildInstanceHookEnvByObject(self, instance))
5797
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
5798
    nl_post = list(nl)
5799
    nl_post.append(source_node)
5800
    return env, nl, nl_post
5801

    
5802
  def CheckPrereq(self):
5803
    """Check prerequisites.
5804

5805
    This checks that the instance is in the cluster.
5806

5807
    """
5808
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5809
    assert self.instance is not None, \
5810
      "Cannot retrieve locked instance %s" % self.op.instance_name
5811

    
5812
    bep = self.cfg.GetClusterInfo().FillBE(instance)
5813
    if instance.disk_template not in constants.DTS_NET_MIRROR:
5814
      raise errors.OpPrereqError("Instance's disk layout is not"
5815
                                 " network mirrored, cannot failover.",
5816
                                 errors.ECODE_STATE)
5817

    
5818
    secondary_nodes = instance.secondary_nodes
5819
    if not secondary_nodes:
5820
      raise errors.ProgrammerError("no secondary node but using "
5821
                                   "a mirrored disk template")
5822

    
5823
    target_node = secondary_nodes[0]
5824
    _CheckNodeOnline(self, target_node)
5825
    _CheckNodeNotDrained(self, target_node)
5826
    if instance.admin_up:
5827
      # check memory requirements on the secondary node
5828
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
5829
                           instance.name, bep[constants.BE_MEMORY],
5830
                           instance.hypervisor)
5831
    else:
5832
      self.LogInfo("Not checking memory on the secondary node as"
5833
                   " instance will not be started")
5834

    
5835
    # check bridge existance
5836
    _CheckInstanceBridgesExist(self, instance, node=target_node)
5837

    
5838
  def Exec(self, feedback_fn):
5839
    """Failover an instance.
5840

5841
    The failover is done by shutting it down on its present node and
5842
    starting it on the secondary.
5843

5844
    """
5845
    instance = self.instance
5846
    primary_node = self.cfg.GetNodeInfo(instance.primary_node)
5847

    
5848
    source_node = instance.primary_node
5849
    target_node = instance.secondary_nodes[0]
5850

    
5851
    if instance.admin_up:
5852
      feedback_fn("* checking disk consistency between source and target")
5853
      for dev in instance.disks:
5854
        # for drbd, these are drbd over lvm
5855
        if not _CheckDiskConsistency(self, dev, target_node, False):
5856
          if not self.op.ignore_consistency:
5857
            raise errors.OpExecError("Disk %s is degraded on target node,"
5858
                                     " aborting failover." % dev.iv_name)
5859
    else:
5860
      feedback_fn("* not checking disk consistency as instance is not running")
5861

    
5862
    feedback_fn("* shutting down instance on source node")
5863
    logging.info("Shutting down instance %s on node %s",
5864
                 instance.name, source_node)
5865

    
5866
    result = self.rpc.call_instance_shutdown(source_node, instance,
5867
                                             self.op.shutdown_timeout)
5868
    msg = result.fail_msg
5869
    if msg:
5870
      if self.op.ignore_consistency or primary_node.offline:
5871
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
5872
                             " Proceeding anyway. Please make sure node"
5873
                             " %s is down. Error details: %s",
5874
                             instance.name, source_node, source_node, msg)
5875
      else:
5876
        raise errors.OpExecError("Could not shutdown instance %s on"
5877
                                 " node %s: %s" %
5878
                                 (instance.name, source_node, msg))
5879

    
5880
    feedback_fn("* deactivating the instance's disks on source node")
5881
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
5882
      raise errors.OpExecError("Can't shut down the instance's disks.")
5883

    
5884
    instance.primary_node = target_node
5885
    # distribute new instance config to the other nodes
5886
    self.cfg.Update(instance, feedback_fn)
5887

    
5888
    # Only start the instance if it's marked as up
5889
    if instance.admin_up:
5890
      feedback_fn("* activating the instance's disks on target node")
5891
      logging.info("Starting instance %s on node %s",
5892
                   instance.name, target_node)
5893

    
5894
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
5895
                                           ignore_secondaries=True)
5896
      if not disks_ok:
5897
        _ShutdownInstanceDisks(self, instance)
5898
        raise errors.OpExecError("Can't activate the instance's disks")
5899

    
5900
      feedback_fn("* starting the instance on the target node")
5901
      result = self.rpc.call_instance_start(target_node, instance, None, None)
5902
      msg = result.fail_msg
5903
      if msg:
5904
        _ShutdownInstanceDisks(self, instance)
5905
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
5906
                                 (instance.name, target_node, msg))
5907

    
5908

    
5909
class LUInstanceMigrate(LogicalUnit):
5910
  """Migrate an instance.
5911

5912
  This is migration without shutting down, compared to the failover,
5913
  which is done with shutdown.
5914

5915
  """
5916
  HPATH = "instance-migrate"
5917
  HTYPE = constants.HTYPE_INSTANCE
5918
  REQ_BGL = False
5919

    
5920
  def ExpandNames(self):
5921
    self._ExpandAndLockInstance()
5922

    
5923
    self.needed_locks[locking.LEVEL_NODE] = []
5924
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5925

    
5926
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
5927
                                       self.op.cleanup)
5928
    self.tasklets = [self._migrater]
5929

    
5930
  def DeclareLocks(self, level):
5931
    if level == locking.LEVEL_NODE:
5932
      self._LockInstancesNodes()
5933

    
5934
  def BuildHooksEnv(self):
5935
    """Build hooks env.
5936

5937
    This runs on master, primary and secondary nodes of the instance.
5938

5939
    """
5940
    instance = self._migrater.instance
5941
    source_node = instance.primary_node
5942
    target_node = instance.secondary_nodes[0]
5943
    env = _BuildInstanceHookEnvByObject(self, instance)
5944
    env["MIGRATE_LIVE"] = self._migrater.live
5945
    env["MIGRATE_CLEANUP"] = self.op.cleanup
5946
    env.update({
5947
        "OLD_PRIMARY": source_node,
5948
        "OLD_SECONDARY": target_node,
5949
        "NEW_PRIMARY": target_node,
5950
        "NEW_SECONDARY": source_node,
5951
        })
5952
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
5953
    nl_post = list(nl)
5954
    nl_post.append(source_node)
5955
    return env, nl, nl_post
5956

    
5957

    
5958
class LUInstanceMove(LogicalUnit):
5959
  """Move an instance by data-copying.
5960

5961
  """
5962
  HPATH = "instance-move"
5963
  HTYPE = constants.HTYPE_INSTANCE
5964
  REQ_BGL = False
5965

    
5966
  def ExpandNames(self):
5967
    self._ExpandAndLockInstance()
5968
    target_node = _ExpandNodeName(self.cfg, self.op.target_node)
5969
    self.op.target_node = target_node
5970
    self.needed_locks[locking.LEVEL_NODE] = [target_node]
5971
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5972

    
5973
  def DeclareLocks(self, level):
5974
    if level == locking.LEVEL_NODE:
5975
      self._LockInstancesNodes(primary_only=True)
5976

    
5977
  def BuildHooksEnv(self):
5978
    """Build hooks env.
5979

5980
    This runs on master, primary and secondary nodes of the instance.
5981

5982
    """
5983
    env = {
5984
      "TARGET_NODE": self.op.target_node,
5985
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
5986
      }
5987
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5988
    nl = [self.cfg.GetMasterNode()] + [self.instance.primary_node,
5989
                                       self.op.target_node]
5990
    return env, nl, nl
5991

    
5992
  def CheckPrereq(self):
5993
    """Check prerequisites.
5994

5995
    This checks that the instance is in the cluster.
5996

5997
    """
5998
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5999
    assert self.instance is not None, \
6000
      "Cannot retrieve locked instance %s" % self.op.instance_name
6001

    
6002
    node = self.cfg.GetNodeInfo(self.op.target_node)
6003
    assert node is not None, \
6004
      "Cannot retrieve locked node %s" % self.op.target_node
6005

    
6006
    self.target_node = target_node = node.name
6007

    
6008
    if target_node == instance.primary_node:
6009
      raise errors.OpPrereqError("Instance %s is already on the node %s" %
6010
                                 (instance.name, target_node),
6011
                                 errors.ECODE_STATE)
6012

    
6013
    bep = self.cfg.GetClusterInfo().FillBE(instance)
6014

    
6015
    for idx, dsk in enumerate(instance.disks):
6016
      if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
6017
        raise errors.OpPrereqError("Instance disk %d has a complex layout,"
6018
                                   " cannot copy" % idx, errors.ECODE_STATE)
6019

    
6020
    _CheckNodeOnline(self, target_node)
6021
    _CheckNodeNotDrained(self, target_node)
6022
    _CheckNodeVmCapable(self, target_node)
6023

    
6024
    if instance.admin_up:
6025
      # check memory requirements on the secondary node
6026
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
6027
                           instance.name, bep[constants.BE_MEMORY],
6028
                           instance.hypervisor)
6029
    else:
6030
      self.LogInfo("Not checking memory on the secondary node as"
6031
                   " instance will not be started")
6032

    
6033
    # check bridge existance
6034
    _CheckInstanceBridgesExist(self, instance, node=target_node)
6035

    
6036
  def Exec(self, feedback_fn):
6037
    """Move an instance.
6038

6039
    The move is done by shutting it down on its present node, copying
6040
    the data over (slow) and starting it on the new node.
6041

6042
    """
6043
    instance = self.instance
6044

    
6045
    source_node = instance.primary_node
6046
    target_node = self.target_node
6047

    
6048
    self.LogInfo("Shutting down instance %s on source node %s",
6049
                 instance.name, source_node)
6050

    
6051
    result = self.rpc.call_instance_shutdown(source_node, instance,
6052
                                             self.op.shutdown_timeout)
6053
    msg = result.fail_msg
6054
    if msg:
6055
      if self.op.ignore_consistency:
6056
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
6057
                             " Proceeding anyway. Please make sure node"
6058
                             " %s is down. Error details: %s",
6059
                             instance.name, source_node, source_node, msg)
6060
      else:
6061
        raise errors.OpExecError("Could not shutdown instance %s on"
6062
                                 " node %s: %s" %
6063
                                 (instance.name, source_node, msg))
6064

    
6065
    # create the target disks
6066
    try:
6067
      _CreateDisks(self, instance, target_node=target_node)
6068
    except errors.OpExecError:
6069
      self.LogWarning("Device creation failed, reverting...")
6070
      try:
6071
        _RemoveDisks(self, instance, target_node=target_node)
6072
      finally:
6073
        self.cfg.ReleaseDRBDMinors(instance.name)
6074
        raise
6075

    
6076
    cluster_name = self.cfg.GetClusterInfo().cluster_name
6077

    
6078
    errs = []
6079
    # activate, get path, copy the data over
6080
    for idx, disk in enumerate(instance.disks):
6081
      self.LogInfo("Copying data for disk %d", idx)
6082
      result = self.rpc.call_blockdev_assemble(target_node, disk,
6083
                                               instance.name, True, idx)
6084
      if result.fail_msg:
6085
        self.LogWarning("Can't assemble newly created disk %d: %s",
6086
                        idx, result.fail_msg)
6087
        errs.append(result.fail_msg)
6088
        break
6089
      dev_path = result.payload
6090
      result = self.rpc.call_blockdev_export(source_node, disk,
6091
                                             target_node, dev_path,
6092
                                             cluster_name)
6093
      if result.fail_msg:
6094
        self.LogWarning("Can't copy data over for disk %d: %s",
6095
                        idx, result.fail_msg)
6096
        errs.append(result.fail_msg)
6097
        break
6098

    
6099
    if errs:
6100
      self.LogWarning("Some disks failed to copy, aborting")
6101
      try:
6102
        _RemoveDisks(self, instance, target_node=target_node)
6103
      finally:
6104
        self.cfg.ReleaseDRBDMinors(instance.name)
6105
        raise errors.OpExecError("Errors during disk copy: %s" %
6106
                                 (",".join(errs),))
6107

    
6108
    instance.primary_node = target_node
6109
    self.cfg.Update(instance, feedback_fn)
6110

    
6111
    self.LogInfo("Removing the disks on the original node")
6112
    _RemoveDisks(self, instance, target_node=source_node)
6113

    
6114
    # Only start the instance if it's marked as up
6115
    if instance.admin_up:
6116
      self.LogInfo("Starting instance %s on node %s",
6117
                   instance.name, target_node)
6118

    
6119
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
6120
                                           ignore_secondaries=True)
6121
      if not disks_ok:
6122
        _ShutdownInstanceDisks(self, instance)
6123
        raise errors.OpExecError("Can't activate the instance's disks")
6124

    
6125
      result = self.rpc.call_instance_start(target_node, instance, None, None)
6126
      msg = result.fail_msg
6127
      if msg:
6128
        _ShutdownInstanceDisks(self, instance)
6129
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
6130
                                 (instance.name, target_node, msg))
6131

    
6132

    
6133
class LUNodeMigrate(LogicalUnit):
6134
  """Migrate all instances from a node.
6135

6136
  """
6137
  HPATH = "node-migrate"
6138
  HTYPE = constants.HTYPE_NODE
6139
  REQ_BGL = False
6140

    
6141
  def ExpandNames(self):
6142
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6143

    
6144
    self.needed_locks = {
6145
      locking.LEVEL_NODE: [self.op.node_name],
6146
      }
6147

    
6148
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6149

    
6150
    # Create tasklets for migrating instances for all instances on this node
6151
    names = []
6152
    tasklets = []
6153

    
6154
    for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name):
6155
      logging.debug("Migrating instance %s", inst.name)
6156
      names.append(inst.name)
6157

    
6158
      tasklets.append(TLMigrateInstance(self, inst.name, False))
6159

    
6160
    self.tasklets = tasklets
6161

    
6162
    # Declare instance locks
6163
    self.needed_locks[locking.LEVEL_INSTANCE] = names
6164

    
6165
  def DeclareLocks(self, level):
6166
    if level == locking.LEVEL_NODE:
6167
      self._LockInstancesNodes()
6168

    
6169
  def BuildHooksEnv(self):
6170
    """Build hooks env.
6171

6172
    This runs on the master, the primary and all the secondaries.
6173

6174
    """
6175
    env = {
6176
      "NODE_NAME": self.op.node_name,
6177
      }
6178

    
6179
    nl = [self.cfg.GetMasterNode()]
6180

    
6181
    return (env, nl, nl)
6182

    
6183

    
6184
class TLMigrateInstance(Tasklet):
6185
  """Tasklet class for instance migration.
6186

6187
  @type live: boolean
6188
  @ivar live: whether the migration will be done live or non-live;
6189
      this variable is initalized only after CheckPrereq has run
6190

6191
  """
6192
  def __init__(self, lu, instance_name, cleanup):
6193
    """Initializes this class.
6194

6195
    """
6196
    Tasklet.__init__(self, lu)
6197

    
6198
    # Parameters
6199
    self.instance_name = instance_name
6200
    self.cleanup = cleanup
6201
    self.live = False # will be overridden later
6202

    
6203
  def CheckPrereq(self):
6204
    """Check prerequisites.
6205

6206
    This checks that the instance is in the cluster.
6207

6208
    """
6209
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
6210
    instance = self.cfg.GetInstanceInfo(instance_name)
6211
    assert instance is not None
6212

    
6213
    if instance.disk_template != constants.DT_DRBD8:
6214
      raise errors.OpPrereqError("Instance's disk layout is not"
6215
                                 " drbd8, cannot migrate.", errors.ECODE_STATE)
6216

    
6217
    secondary_nodes = instance.secondary_nodes
6218
    if not secondary_nodes:
6219
      raise errors.ConfigurationError("No secondary node but using"
6220
                                      " drbd8 disk template")
6221

    
6222
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
6223

    
6224
    target_node = secondary_nodes[0]
6225
    # check memory requirements on the secondary node
6226
    _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
6227
                         instance.name, i_be[constants.BE_MEMORY],
6228
                         instance.hypervisor)
6229

    
6230
    # check bridge existance
6231
    _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
6232

    
6233
    if not self.cleanup:
6234
      _CheckNodeNotDrained(self.lu, target_node)
6235
      result = self.rpc.call_instance_migratable(instance.primary_node,
6236
                                                 instance)
6237
      result.Raise("Can't migrate, please use failover",
6238
                   prereq=True, ecode=errors.ECODE_STATE)
6239

    
6240
    self.instance = instance
6241

    
6242
    if self.lu.op.live is not None and self.lu.op.mode is not None:
6243
      raise errors.OpPrereqError("Only one of the 'live' and 'mode'"
6244
                                 " parameters are accepted",
6245
                                 errors.ECODE_INVAL)
6246
    if self.lu.op.live is not None:
6247
      if self.lu.op.live:
6248
        self.lu.op.mode = constants.HT_MIGRATION_LIVE
6249
      else:
6250
        self.lu.op.mode = constants.HT_MIGRATION_NONLIVE
6251
      # reset the 'live' parameter to None so that repeated
6252
      # invocations of CheckPrereq do not raise an exception
6253
      self.lu.op.live = None
6254
    elif self.lu.op.mode is None:
6255
      # read the default value from the hypervisor
6256
      i_hv = self.cfg.GetClusterInfo().FillHV(instance, skip_globals=False)
6257
      self.lu.op.mode = i_hv[constants.HV_MIGRATION_MODE]
6258

    
6259
    self.live = self.lu.op.mode == constants.HT_MIGRATION_LIVE
6260

    
6261
  def _WaitUntilSync(self):
6262
    """Poll with custom rpc for disk sync.
6263

6264
    This uses our own step-based rpc call.
6265

6266
    """
6267
    self.feedback_fn("* wait until resync is done")
6268
    all_done = False
6269
    while not all_done:
6270
      all_done = True
6271
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
6272
                                            self.nodes_ip,
6273
                                            self.instance.disks)
6274
      min_percent = 100
6275
      for node, nres in result.items():
6276
        nres.Raise("Cannot resync disks on node %s" % node)
6277
        node_done, node_percent = nres.payload
6278
        all_done = all_done and node_done
6279
        if node_percent is not None:
6280
          min_percent = min(min_percent, node_percent)
6281
      if not all_done:
6282
        if min_percent < 100:
6283
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
6284
        time.sleep(2)
6285

    
6286
  def _EnsureSecondary(self, node):
6287
    """Demote a node to secondary.
6288

6289
    """
6290
    self.feedback_fn("* switching node %s to secondary mode" % node)
6291

    
6292
    for dev in self.instance.disks:
6293
      self.cfg.SetDiskID(dev, node)
6294

    
6295
    result = self.rpc.call_blockdev_close(node, self.instance.name,
6296
                                          self.instance.disks)
6297
    result.Raise("Cannot change disk to secondary on node %s" % node)
6298

    
6299
  def _GoStandalone(self):
6300
    """Disconnect from the network.
6301

6302
    """
6303
    self.feedback_fn("* changing into standalone mode")
6304
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
6305
                                               self.instance.disks)
6306
    for node, nres in result.items():
6307
      nres.Raise("Cannot disconnect disks node %s" % node)
6308

    
6309
  def _GoReconnect(self, multimaster):
6310
    """Reconnect to the network.
6311

6312
    """
6313
    if multimaster:
6314
      msg = "dual-master"
6315
    else:
6316
      msg = "single-master"
6317
    self.feedback_fn("* changing disks into %s mode" % msg)
6318
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
6319
                                           self.instance.disks,
6320
                                           self.instance.name, multimaster)
6321
    for node, nres in result.items():
6322
      nres.Raise("Cannot change disks config on node %s" % node)
6323

    
6324
  def _ExecCleanup(self):
6325
    """Try to cleanup after a failed migration.
6326

6327
    The cleanup is done by:
6328
      - check that the instance is running only on one node
6329
        (and update the config if needed)
6330
      - change disks on its secondary node to secondary
6331
      - wait until disks are fully synchronized
6332
      - disconnect from the network
6333
      - change disks into single-master mode
6334
      - wait again until disks are fully synchronized
6335

6336
    """
6337
    instance = self.instance
6338
    target_node = self.target_node
6339
    source_node = self.source_node
6340

    
6341
    # check running on only one node
6342
    self.feedback_fn("* checking where the instance actually runs"
6343
                     " (if this hangs, the hypervisor might be in"
6344
                     " a bad state)")
6345
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
6346
    for node, result in ins_l.items():
6347
      result.Raise("Can't contact node %s" % node)
6348

    
6349
    runningon_source = instance.name in ins_l[source_node].payload
6350
    runningon_target = instance.name in ins_l[target_node].payload
6351

    
6352
    if runningon_source and runningon_target:
6353
      raise errors.OpExecError("Instance seems to be running on two nodes,"
6354
                               " or the hypervisor is confused. You will have"
6355
                               " to ensure manually that it runs only on one"
6356
                               " and restart this operation.")
6357

    
6358
    if not (runningon_source or runningon_target):
6359
      raise errors.OpExecError("Instance does not seem to be running at all."
6360
                               " In this case, it's safer to repair by"
6361
                               " running 'gnt-instance stop' to ensure disk"
6362
                               " shutdown, and then restarting it.")
6363

    
6364
    if runningon_target:
6365
      # the migration has actually succeeded, we need to update the config
6366
      self.feedback_fn("* instance running on secondary node (%s),"
6367
                       " updating config" % target_node)
6368
      instance.primary_node = target_node
6369
      self.cfg.Update(instance, self.feedback_fn)
6370
      demoted_node = source_node
6371
    else:
6372
      self.feedback_fn("* instance confirmed to be running on its"
6373
                       " primary node (%s)" % source_node)
6374
      demoted_node = target_node
6375

    
6376
    self._EnsureSecondary(demoted_node)
6377
    try:
6378
      self._WaitUntilSync()
6379
    except errors.OpExecError:
6380
      # we ignore here errors, since if the device is standalone, it
6381
      # won't be able to sync
6382
      pass
6383
    self._GoStandalone()
6384
    self._GoReconnect(False)
6385
    self._WaitUntilSync()
6386

    
6387
    self.feedback_fn("* done")
6388

    
6389
  def _RevertDiskStatus(self):
6390
    """Try to revert the disk status after a failed migration.
6391

6392
    """
6393
    target_node = self.target_node
6394
    try:
6395
      self._EnsureSecondary(target_node)
6396
      self._GoStandalone()
6397
      self._GoReconnect(False)
6398
      self._WaitUntilSync()
6399
    except errors.OpExecError, err:
6400
      self.lu.LogWarning("Migration failed and I can't reconnect the"
6401
                         " drives: error '%s'\n"
6402
                         "Please look and recover the instance status" %
6403
                         str(err))
6404

    
6405
  def _AbortMigration(self):
6406
    """Call the hypervisor code to abort a started migration.
6407

6408
    """
6409
    instance = self.instance
6410
    target_node = self.target_node
6411
    migration_info = self.migration_info
6412

    
6413
    abort_result = self.rpc.call_finalize_migration(target_node,
6414
                                                    instance,
6415
                                                    migration_info,
6416
                                                    False)
6417
    abort_msg = abort_result.fail_msg
6418
    if abort_msg:
6419
      logging.error("Aborting migration failed on target node %s: %s",
6420
                    target_node, abort_msg)
6421
      # Don't raise an exception here, as we stil have to try to revert the
6422
      # disk status, even if this step failed.
6423

    
6424
  def _ExecMigration(self):
6425
    """Migrate an instance.
6426

6427
    The migrate is done by:
6428
      - change the disks into dual-master mode
6429
      - wait until disks are fully synchronized again
6430
      - migrate the instance
6431
      - change disks on the new secondary node (the old primary) to secondary
6432
      - wait until disks are fully synchronized
6433
      - change disks into single-master mode
6434

6435
    """
6436
    instance = self.instance
6437
    target_node = self.target_node
6438
    source_node = self.source_node
6439

    
6440
    self.feedback_fn("* checking disk consistency between source and target")
6441
    for dev in instance.disks:
6442
      if not _CheckDiskConsistency(self.lu, dev, target_node, False):
6443
        raise errors.OpExecError("Disk %s is degraded or not fully"
6444
                                 " synchronized on target node,"
6445
                                 " aborting migrate." % dev.iv_name)
6446

    
6447
    # First get the migration information from the remote node
6448
    result = self.rpc.call_migration_info(source_node, instance)
6449
    msg = result.fail_msg
6450
    if msg:
6451
      log_err = ("Failed fetching source migration information from %s: %s" %
6452
                 (source_node, msg))
6453
      logging.error(log_err)
6454
      raise errors.OpExecError(log_err)
6455

    
6456
    self.migration_info = migration_info = result.payload
6457

    
6458
    # Then switch the disks to master/master mode
6459
    self._EnsureSecondary(target_node)
6460
    self._GoStandalone()
6461
    self._GoReconnect(True)
6462
    self._WaitUntilSync()
6463

    
6464
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
6465
    result = self.rpc.call_accept_instance(target_node,
6466
                                           instance,
6467
                                           migration_info,
6468
                                           self.nodes_ip[target_node])
6469

    
6470
    msg = result.fail_msg
6471
    if msg:
6472
      logging.error("Instance pre-migration failed, trying to revert"
6473
                    " disk status: %s", msg)
6474
      self.feedback_fn("Pre-migration failed, aborting")
6475
      self._AbortMigration()
6476
      self._RevertDiskStatus()
6477
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
6478
                               (instance.name, msg))
6479

    
6480
    self.feedback_fn("* migrating instance to %s" % target_node)
6481
    time.sleep(10)
6482
    result = self.rpc.call_instance_migrate(source_node, instance,
6483
                                            self.nodes_ip[target_node],
6484
                                            self.live)
6485
    msg = result.fail_msg
6486
    if msg:
6487
      logging.error("Instance migration failed, trying to revert"
6488
                    " disk status: %s", msg)
6489
      self.feedback_fn("Migration failed, aborting")
6490
      self._AbortMigration()
6491
      self._RevertDiskStatus()
6492
      raise errors.OpExecError("Could not migrate instance %s: %s" %
6493
                               (instance.name, msg))
6494
    time.sleep(10)
6495

    
6496
    instance.primary_node = target_node
6497
    # distribute new instance config to the other nodes
6498
    self.cfg.Update(instance, self.feedback_fn)
6499

    
6500
    result = self.rpc.call_finalize_migration(target_node,
6501
                                              instance,
6502
                                              migration_info,
6503
                                              True)
6504
    msg = result.fail_msg
6505
    if msg:
6506
      logging.error("Instance migration succeeded, but finalization failed:"
6507
                    " %s", msg)
6508
      raise errors.OpExecError("Could not finalize instance migration: %s" %
6509
                               msg)
6510

    
6511
    self._EnsureSecondary(source_node)
6512
    self._WaitUntilSync()
6513
    self._GoStandalone()
6514
    self._GoReconnect(False)
6515
    self._WaitUntilSync()
6516

    
6517
    self.feedback_fn("* done")
6518

    
6519
  def Exec(self, feedback_fn):
6520
    """Perform the migration.
6521

6522
    """
6523
    feedback_fn("Migrating instance %s" % self.instance.name)
6524

    
6525
    self.feedback_fn = feedback_fn
6526

    
6527
    self.source_node = self.instance.primary_node
6528
    self.target_node = self.instance.secondary_nodes[0]
6529
    self.all_nodes = [self.source_node, self.target_node]
6530
    self.nodes_ip = {
6531
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
6532
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
6533
      }
6534

    
6535
    if self.cleanup:
6536
      return self._ExecCleanup()
6537
    else:
6538
      return self._ExecMigration()
6539

    
6540

    
6541
def _CreateBlockDev(lu, node, instance, device, force_create,
6542
                    info, force_open):
6543
  """Create a tree of block devices on a given node.
6544

6545
  If this device type has to be created on secondaries, create it and
6546
  all its children.
6547

6548
  If not, just recurse to children keeping the same 'force' value.
6549

6550
  @param lu: the lu on whose behalf we execute
6551
  @param node: the node on which to create the device
6552
  @type instance: L{objects.Instance}
6553
  @param instance: the instance which owns the device
6554
  @type device: L{objects.Disk}
6555
  @param device: the device to create
6556
  @type force_create: boolean
6557
  @param force_create: whether to force creation of this device; this
6558
      will be change to True whenever we find a device which has
6559
      CreateOnSecondary() attribute
6560
  @param info: the extra 'metadata' we should attach to the device
6561
      (this will be represented as a LVM tag)
6562
  @type force_open: boolean
6563
  @param force_open: this parameter will be passes to the
6564
      L{backend.BlockdevCreate} function where it specifies
6565
      whether we run on primary or not, and it affects both
6566
      the child assembly and the device own Open() execution
6567

6568
  """
6569
  if device.CreateOnSecondary():
6570
    force_create = True
6571

    
6572
  if device.children:
6573
    for child in device.children:
6574
      _CreateBlockDev(lu, node, instance, child, force_create,
6575
                      info, force_open)
6576

    
6577
  if not force_create:
6578
    return
6579

    
6580
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
6581

    
6582

    
6583
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
6584
  """Create a single block device on a given node.
6585

6586
  This will not recurse over children of the device, so they must be
6587
  created in advance.
6588

6589
  @param lu: the lu on whose behalf we execute
6590
  @param node: the node on which to create the device
6591
  @type instance: L{objects.Instance}
6592
  @param instance: the instance which owns the device
6593
  @type device: L{objects.Disk}
6594
  @param device: the device to create
6595
  @param info: the extra 'metadata' we should attach to the device
6596
      (this will be represented as a LVM tag)
6597
  @type force_open: boolean
6598
  @param force_open: this parameter will be passes to the
6599
      L{backend.BlockdevCreate} function where it specifies
6600
      whether we run on primary or not, and it affects both
6601
      the child assembly and the device own Open() execution
6602

6603
  """
6604
  lu.cfg.SetDiskID(device, node)
6605
  result = lu.rpc.call_blockdev_create(node, device, device.size,
6606
                                       instance.name, force_open, info)
6607
  result.Raise("Can't create block device %s on"
6608
               " node %s for instance %s" % (device, node, instance.name))
6609
  if device.physical_id is None:
6610
    device.physical_id = result.payload
6611

    
6612

    
6613
def _GenerateUniqueNames(lu, exts):
6614
  """Generate a suitable LV name.
6615

6616
  This will generate a logical volume name for the given instance.
6617

6618
  """
6619
  results = []
6620
  for val in exts:
6621
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
6622
    results.append("%s%s" % (new_id, val))
6623
  return results
6624

    
6625

    
6626
def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
6627
                         iv_name, p_minor, s_minor):
6628
  """Generate a drbd8 device complete with its children.
6629

6630
  """
6631
  assert len(vgnames) == len(names) == 2
6632
  port = lu.cfg.AllocatePort()
6633
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
6634
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
6635
                          logical_id=(vgnames[0], names[0]))
6636
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
6637
                          logical_id=(vgnames[1], names[1]))
6638
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
6639
                          logical_id=(primary, secondary, port,
6640
                                      p_minor, s_minor,
6641
                                      shared_secret),
6642
                          children=[dev_data, dev_meta],
6643
                          iv_name=iv_name)
6644
  return drbd_dev
6645

    
6646

    
6647
def _GenerateDiskTemplate(lu, template_name,
6648
                          instance_name, primary_node,
6649
                          secondary_nodes, disk_info,
6650
                          file_storage_dir, file_driver,
6651
                          base_index, feedback_fn):
6652
  """Generate the entire disk layout for a given template type.
6653

6654
  """
6655
  #TODO: compute space requirements
6656

    
6657
  vgname = lu.cfg.GetVGName()
6658
  disk_count = len(disk_info)
6659
  disks = []
6660
  if template_name == constants.DT_DISKLESS:
6661
    pass
6662
  elif template_name == constants.DT_PLAIN:
6663
    if len(secondary_nodes) != 0:
6664
      raise errors.ProgrammerError("Wrong template configuration")
6665

    
6666
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6667
                                      for i in range(disk_count)])
6668
    for idx, disk in enumerate(disk_info):
6669
      disk_index = idx + base_index
6670
      vg = disk.get("vg", vgname)
6671
      feedback_fn("* disk %i, vg %s, name %s" % (idx, vg, names[idx]))
6672
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
6673
                              logical_id=(vg, names[idx]),
6674
                              iv_name="disk/%d" % disk_index,
6675
                              mode=disk["mode"])
6676
      disks.append(disk_dev)
6677
  elif template_name == constants.DT_DRBD8:
6678
    if len(secondary_nodes) != 1:
6679
      raise errors.ProgrammerError("Wrong template configuration")
6680
    remote_node = secondary_nodes[0]
6681
    minors = lu.cfg.AllocateDRBDMinor(
6682
      [primary_node, remote_node] * len(disk_info), instance_name)
6683

    
6684
    names = []
6685
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
6686
                                               for i in range(disk_count)]):
6687
      names.append(lv_prefix + "_data")
6688
      names.append(lv_prefix + "_meta")
6689
    for idx, disk in enumerate(disk_info):
6690
      disk_index = idx + base_index
6691
      data_vg = disk.get("vg", vgname)
6692
      meta_vg = disk.get("metavg", data_vg)
6693
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
6694
                                      disk["size"], [data_vg, meta_vg],
6695
                                      names[idx*2:idx*2+2],
6696
                                      "disk/%d" % disk_index,
6697
                                      minors[idx*2], minors[idx*2+1])
6698
      disk_dev.mode = disk["mode"]
6699
      disks.append(disk_dev)
6700
  elif template_name == constants.DT_FILE:
6701
    if len(secondary_nodes) != 0:
6702
      raise errors.ProgrammerError("Wrong template configuration")
6703

    
6704
    opcodes.RequireFileStorage()
6705

    
6706
    for idx, disk in enumerate(disk_info):
6707
      disk_index = idx + base_index
6708
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
6709
                              iv_name="disk/%d" % disk_index,
6710
                              logical_id=(file_driver,
6711
                                          "%s/disk%d" % (file_storage_dir,
6712
                                                         disk_index)),
6713
                              mode=disk["mode"])
6714
      disks.append(disk_dev)
6715
  else:
6716
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
6717
  return disks
6718

    
6719

    
6720
def _GetInstanceInfoText(instance):
6721
  """Compute that text that should be added to the disk's metadata.
6722

6723
  """
6724
  return "originstname+%s" % instance.name
6725

    
6726

    
6727
def _CalcEta(time_taken, written, total_size):
6728
  """Calculates the ETA based on size written and total size.
6729

6730
  @param time_taken: The time taken so far
6731
  @param written: amount written so far
6732
  @param total_size: The total size of data to be written
6733
  @return: The remaining time in seconds
6734

6735
  """
6736
  avg_time = time_taken / float(written)
6737
  return (total_size - written) * avg_time
6738

    
6739

    
6740
def _WipeDisks(lu, instance):
6741
  """Wipes instance disks.
6742

6743
  @type lu: L{LogicalUnit}
6744
  @param lu: the logical unit on whose behalf we execute
6745
  @type instance: L{objects.Instance}
6746
  @param instance: the instance whose disks we should create
6747
  @return: the success of the wipe
6748

6749
  """
6750
  node = instance.primary_node
6751

    
6752
  for device in instance.disks:
6753
    lu.cfg.SetDiskID(device, node)
6754

    
6755
  logging.info("Pause sync of instance %s disks", instance.name)
6756
  result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, True)
6757

    
6758
  for idx, success in enumerate(result.payload):
6759
    if not success:
6760
      logging.warn("pause-sync of instance %s for disks %d failed",
6761
                   instance.name, idx)
6762

    
6763
  try:
6764
    for idx, device in enumerate(instance.disks):
6765
      # The wipe size is MIN_WIPE_CHUNK_PERCENT % of the instance disk but
6766
      # MAX_WIPE_CHUNK at max
6767
      wipe_chunk_size = min(constants.MAX_WIPE_CHUNK, device.size / 100.0 *
6768
                            constants.MIN_WIPE_CHUNK_PERCENT)
6769
      # we _must_ make this an int, otherwise rounding errors will
6770
      # occur
6771
      wipe_chunk_size = int(wipe_chunk_size)
6772

    
6773
      lu.LogInfo("* Wiping disk %d", idx)
6774
      logging.info("Wiping disk %d for instance %s, node %s using"
6775
                   " chunk size %s", idx, instance.name, node, wipe_chunk_size)
6776

    
6777
      offset = 0
6778
      size = device.size
6779
      last_output = 0
6780
      start_time = time.time()
6781

    
6782
      while offset < size:
6783
        wipe_size = min(wipe_chunk_size, size - offset)
6784
        logging.debug("Wiping disk %d, offset %s, chunk %s",
6785
                      idx, offset, wipe_size)
6786
        result = lu.rpc.call_blockdev_wipe(node, device, offset, wipe_size)
6787
        result.Raise("Could not wipe disk %d at offset %d for size %d" %
6788
                     (idx, offset, wipe_size))
6789
        now = time.time()
6790
        offset += wipe_size
6791
        if now - last_output >= 60:
6792
          eta = _CalcEta(now - start_time, offset, size)
6793
          lu.LogInfo(" - done: %.1f%% ETA: %s" %
6794
                     (offset / float(size) * 100, utils.FormatSeconds(eta)))
6795
          last_output = now
6796
  finally:
6797
    logging.info("Resume sync of instance %s disks", instance.name)
6798

    
6799
    result = lu.rpc.call_blockdev_pause_resume_sync(node, instance.disks, False)
6800

    
6801
    for idx, success in enumerate(result.payload):
6802
      if not success:
6803
        lu.LogWarning("Warning: Resume sync of disk %d failed. Please have a"
6804
                      " look at the status and troubleshoot the issue.", idx)
6805
        logging.warn("resume-sync of instance %s for disks %d failed",
6806
                     instance.name, idx)
6807

    
6808

    
6809
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
6810
  """Create all disks for an instance.
6811

6812
  This abstracts away some work from AddInstance.
6813

6814
  @type lu: L{LogicalUnit}
6815
  @param lu: the logical unit on whose behalf we execute
6816
  @type instance: L{objects.Instance}
6817
  @param instance: the instance whose disks we should create
6818
  @type to_skip: list
6819
  @param to_skip: list of indices to skip
6820
  @type target_node: string
6821
  @param target_node: if passed, overrides the target node for creation
6822
  @rtype: boolean
6823
  @return: the success of the creation
6824

6825
  """
6826
  info = _GetInstanceInfoText(instance)
6827
  if target_node is None:
6828
    pnode = instance.primary_node
6829
    all_nodes = instance.all_nodes
6830
  else:
6831
    pnode = target_node
6832
    all_nodes = [pnode]
6833

    
6834
  if instance.disk_template == constants.DT_FILE:
6835
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6836
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
6837

    
6838
    result.Raise("Failed to create directory '%s' on"
6839
                 " node %s" % (file_storage_dir, pnode))
6840

    
6841
  # Note: this needs to be kept in sync with adding of disks in
6842
  # LUInstanceSetParams
6843
  for idx, device in enumerate(instance.disks):
6844
    if to_skip and idx in to_skip:
6845
      continue
6846
    logging.info("Creating volume %s for instance %s",
6847
                 device.iv_name, instance.name)
6848
    #HARDCODE
6849
    for node in all_nodes:
6850
      f_create = node == pnode
6851
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
6852

    
6853

    
6854
def _RemoveDisks(lu, instance, target_node=None):
6855
  """Remove all disks for an instance.
6856

6857
  This abstracts away some work from `AddInstance()` and
6858
  `RemoveInstance()`. Note that in case some of the devices couldn't
6859
  be removed, the removal will continue with the other ones (compare
6860
  with `_CreateDisks()`).
6861

6862
  @type lu: L{LogicalUnit}
6863
  @param lu: the logical unit on whose behalf we execute
6864
  @type instance: L{objects.Instance}
6865
  @param instance: the instance whose disks we should remove
6866
  @type target_node: string
6867
  @param target_node: used to override the node on which to remove the disks
6868
  @rtype: boolean
6869
  @return: the success of the removal
6870

6871
  """
6872
  logging.info("Removing block devices for instance %s", instance.name)
6873

    
6874
  all_result = True
6875
  for device in instance.disks:
6876
    if target_node:
6877
      edata = [(target_node, device)]
6878
    else:
6879
      edata = device.ComputeNodeTree(instance.primary_node)
6880
    for node, disk in edata:
6881
      lu.cfg.SetDiskID(disk, node)
6882
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
6883
      if msg:
6884
        lu.LogWarning("Could not remove block device %s on node %s,"
6885
                      " continuing anyway: %s", device.iv_name, node, msg)
6886
        all_result = False
6887

    
6888
  if instance.disk_template == constants.DT_FILE:
6889
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
6890
    if target_node:
6891
      tgt = target_node
6892
    else:
6893
      tgt = instance.primary_node
6894
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
6895
    if result.fail_msg:
6896
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
6897
                    file_storage_dir, instance.primary_node, result.fail_msg)
6898
      all_result = False
6899

    
6900
  return all_result
6901

    
6902

    
6903
def _ComputeDiskSizePerVG(disk_template, disks):
6904
  """Compute disk size requirements in the volume group
6905

6906
  """
6907
  def _compute(disks, payload):
6908
    """Universal algorithm
6909

6910
    """
6911
    vgs = {}
6912
    for disk in disks:
6913
      vgs[disk["vg"]] = vgs.get("vg", 0) + disk["size"] + payload
6914

    
6915
    return vgs
6916

    
6917
  # Required free disk space as a function of disk and swap space
6918
  req_size_dict = {
6919
    constants.DT_DISKLESS: {},
6920
    constants.DT_PLAIN: _compute(disks, 0),
6921
    # 128 MB are added for drbd metadata for each disk
6922
    constants.DT_DRBD8: _compute(disks, 128),
6923
    constants.DT_FILE: {},
6924
  }
6925

    
6926
  if disk_template not in req_size_dict:
6927
    raise errors.ProgrammerError("Disk template '%s' size requirement"
6928
                                 " is unknown" %  disk_template)
6929

    
6930
  return req_size_dict[disk_template]
6931

    
6932

    
6933
def _ComputeDiskSize(disk_template, disks):
6934
  """Compute disk size requirements in the volume group
6935

6936
  """
6937
  # Required free disk space as a function of disk and swap space
6938
  req_size_dict = {
6939
    constants.DT_DISKLESS: None,
6940
    constants.DT_PLAIN: sum(d["size"] for d in disks),
6941
    # 128 MB are added for drbd metadata for each disk
6942
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
6943
    constants.DT_FILE: None,
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 _FilterVmNodes(lu, nodenames):
6954
  """Filters out non-vm_capable nodes from a list.
6955

6956
  @type lu: L{LogicalUnit}
6957
  @param lu: the logical unit for which we check
6958
  @type nodenames: list
6959
  @param nodenames: the list of nodes on which we should check
6960
  @rtype: list
6961
  @return: the list of vm-capable nodes
6962

6963
  """
6964
  vm_nodes = frozenset(lu.cfg.GetNonVmCapableNodeList())
6965
  return [name for name in nodenames if name not in vm_nodes]
6966

    
6967

    
6968
def _CheckHVParams(lu, nodenames, hvname, hvparams):
6969
  """Hypervisor parameter validation.
6970

6971
  This function abstract the hypervisor parameter validation to be
6972
  used in both instance create and instance modify.
6973

6974
  @type lu: L{LogicalUnit}
6975
  @param lu: the logical unit for which we check
6976
  @type nodenames: list
6977
  @param nodenames: the list of nodes on which we should check
6978
  @type hvname: string
6979
  @param hvname: the name of the hypervisor we should use
6980
  @type hvparams: dict
6981
  @param hvparams: the parameters which we need to check
6982
  @raise errors.OpPrereqError: if the parameters are not valid
6983

6984
  """
6985
  nodenames = _FilterVmNodes(lu, nodenames)
6986
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
6987
                                                  hvname,
6988
                                                  hvparams)
6989
  for node in nodenames:
6990
    info = hvinfo[node]
6991
    if info.offline:
6992
      continue
6993
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
6994

    
6995

    
6996
def _CheckOSParams(lu, required, nodenames, osname, osparams):
6997
  """OS parameters validation.
6998

6999
  @type lu: L{LogicalUnit}
7000
  @param lu: the logical unit for which we check
7001
  @type required: boolean
7002
  @param required: whether the validation should fail if the OS is not
7003
      found
7004
  @type nodenames: list
7005
  @param nodenames: the list of nodes on which we should check
7006
  @type osname: string
7007
  @param osname: the name of the hypervisor we should use
7008
  @type osparams: dict
7009
  @param osparams: the parameters which we need to check
7010
  @raise errors.OpPrereqError: if the parameters are not valid
7011

7012
  """
7013
  nodenames = _FilterVmNodes(lu, nodenames)
7014
  result = lu.rpc.call_os_validate(required, nodenames, osname,
7015
                                   [constants.OS_VALIDATE_PARAMETERS],
7016
                                   osparams)
7017
  for node, nres in result.items():
7018
    # we don't check for offline cases since this should be run only
7019
    # against the master node and/or an instance's nodes
7020
    nres.Raise("OS Parameters validation failed on node %s" % node)
7021
    if not nres.payload:
7022
      lu.LogInfo("OS %s not found on node %s, validation skipped",
7023
                 osname, node)
7024

    
7025

    
7026
class LUInstanceCreate(LogicalUnit):
7027
  """Create an instance.
7028

7029
  """
7030
  HPATH = "instance-add"
7031
  HTYPE = constants.HTYPE_INSTANCE
7032
  REQ_BGL = False
7033

    
7034
  def CheckArguments(self):
7035
    """Check arguments.
7036

7037
    """
7038
    # do not require name_check to ease forward/backward compatibility
7039
    # for tools
7040
    if self.op.no_install and self.op.start:
7041
      self.LogInfo("No-installation mode selected, disabling startup")
7042
      self.op.start = False
7043
    # validate/normalize the instance name
7044
    self.op.instance_name = \
7045
      netutils.Hostname.GetNormalizedName(self.op.instance_name)
7046

    
7047
    if self.op.ip_check and not self.op.name_check:
7048
      # TODO: make the ip check more flexible and not depend on the name check
7049
      raise errors.OpPrereqError("Cannot do ip check without a name check",
7050
                                 errors.ECODE_INVAL)
7051

    
7052
    # check nics' parameter names
7053
    for nic in self.op.nics:
7054
      utils.ForceDictType(nic, constants.INIC_PARAMS_TYPES)
7055

    
7056
    # check disks. parameter names and consistent adopt/no-adopt strategy
7057
    has_adopt = has_no_adopt = False
7058
    for disk in self.op.disks:
7059
      utils.ForceDictType(disk, constants.IDISK_PARAMS_TYPES)
7060
      if "adopt" in disk:
7061
        has_adopt = True
7062
      else:
7063
        has_no_adopt = True
7064
    if has_adopt and has_no_adopt:
7065
      raise errors.OpPrereqError("Either all disks are adopted or none is",
7066
                                 errors.ECODE_INVAL)
7067
    if has_adopt:
7068
      if self.op.disk_template not in constants.DTS_MAY_ADOPT:
7069
        raise errors.OpPrereqError("Disk adoption is not supported for the"
7070
                                   " '%s' disk template" %
7071
                                   self.op.disk_template,
7072
                                   errors.ECODE_INVAL)
7073
      if self.op.iallocator is not None:
7074
        raise errors.OpPrereqError("Disk adoption not allowed with an"
7075
                                   " iallocator script", errors.ECODE_INVAL)
7076
      if self.op.mode == constants.INSTANCE_IMPORT:
7077
        raise errors.OpPrereqError("Disk adoption not allowed for"
7078
                                   " instance import", errors.ECODE_INVAL)
7079

    
7080
    self.adopt_disks = has_adopt
7081

    
7082
    # instance name verification
7083
    if self.op.name_check:
7084
      self.hostname1 = netutils.GetHostname(name=self.op.instance_name)
7085
      self.op.instance_name = self.hostname1.name
7086
      # used in CheckPrereq for ip ping check
7087
      self.check_ip = self.hostname1.ip
7088
    else:
7089
      self.check_ip = None
7090

    
7091
    # file storage checks
7092
    if (self.op.file_driver and
7093
        not self.op.file_driver in constants.FILE_DRIVER):
7094
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
7095
                                 self.op.file_driver, errors.ECODE_INVAL)
7096

    
7097
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
7098
      raise errors.OpPrereqError("File storage directory path not absolute",
7099
                                 errors.ECODE_INVAL)
7100

    
7101
    ### Node/iallocator related checks
7102
    _CheckIAllocatorOrNode(self, "iallocator", "pnode")
7103

    
7104
    if self.op.pnode is not None:
7105
      if self.op.disk_template in constants.DTS_NET_MIRROR:
7106
        if self.op.snode is None:
7107
          raise errors.OpPrereqError("The networked disk templates need"
7108
                                     " a mirror node", errors.ECODE_INVAL)
7109
      elif self.op.snode:
7110
        self.LogWarning("Secondary node will be ignored on non-mirrored disk"
7111
                        " template")
7112
        self.op.snode = None
7113

    
7114
    self._cds = _GetClusterDomainSecret()
7115

    
7116
    if self.op.mode == constants.INSTANCE_IMPORT:
7117
      # On import force_variant must be True, because if we forced it at
7118
      # initial install, our only chance when importing it back is that it
7119
      # works again!
7120
      self.op.force_variant = True
7121

    
7122
      if self.op.no_install:
7123
        self.LogInfo("No-installation mode has no effect during import")
7124

    
7125
    elif self.op.mode == constants.INSTANCE_CREATE:
7126
      if self.op.os_type is None:
7127
        raise errors.OpPrereqError("No guest OS specified",
7128
                                   errors.ECODE_INVAL)
7129
      if self.op.os_type in self.cfg.GetClusterInfo().blacklisted_os:
7130
        raise errors.OpPrereqError("Guest OS '%s' is not allowed for"
7131
                                   " installation" % self.op.os_type,
7132
                                   errors.ECODE_STATE)
7133
      if self.op.disk_template is None:
7134
        raise errors.OpPrereqError("No disk template specified",
7135
                                   errors.ECODE_INVAL)
7136

    
7137
    elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7138
      # Check handshake to ensure both clusters have the same domain secret
7139
      src_handshake = self.op.source_handshake
7140
      if not src_handshake:
7141
        raise errors.OpPrereqError("Missing source handshake",
7142
                                   errors.ECODE_INVAL)
7143

    
7144
      errmsg = masterd.instance.CheckRemoteExportHandshake(self._cds,
7145
                                                           src_handshake)
7146
      if errmsg:
7147
        raise errors.OpPrereqError("Invalid handshake: %s" % errmsg,
7148
                                   errors.ECODE_INVAL)
7149

    
7150
      # Load and check source CA
7151
      self.source_x509_ca_pem = self.op.source_x509_ca
7152
      if not self.source_x509_ca_pem:
7153
        raise errors.OpPrereqError("Missing source X509 CA",
7154
                                   errors.ECODE_INVAL)
7155

    
7156
      try:
7157
        (cert, _) = utils.LoadSignedX509Certificate(self.source_x509_ca_pem,
7158
                                                    self._cds)
7159
      except OpenSSL.crypto.Error, err:
7160
        raise errors.OpPrereqError("Unable to load source X509 CA (%s)" %
7161
                                   (err, ), errors.ECODE_INVAL)
7162

    
7163
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
7164
      if errcode is not None:
7165
        raise errors.OpPrereqError("Invalid source X509 CA (%s)" % (msg, ),
7166
                                   errors.ECODE_INVAL)
7167

    
7168
      self.source_x509_ca = cert
7169

    
7170
      src_instance_name = self.op.source_instance_name
7171
      if not src_instance_name:
7172
        raise errors.OpPrereqError("Missing source instance name",
7173
                                   errors.ECODE_INVAL)
7174

    
7175
      self.source_instance_name = \
7176
          netutils.GetHostname(name=src_instance_name).name
7177

    
7178
    else:
7179
      raise errors.OpPrereqError("Invalid instance creation mode %r" %
7180
                                 self.op.mode, errors.ECODE_INVAL)
7181

    
7182
  def ExpandNames(self):
7183
    """ExpandNames for CreateInstance.
7184

7185
    Figure out the right locks for instance creation.
7186

7187
    """
7188
    self.needed_locks = {}
7189

    
7190
    instance_name = self.op.instance_name
7191
    # this is just a preventive check, but someone might still add this
7192
    # instance in the meantime, and creation will fail at lock-add time
7193
    if instance_name in self.cfg.GetInstanceList():
7194
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
7195
                                 instance_name, errors.ECODE_EXISTS)
7196

    
7197
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
7198

    
7199
    if self.op.iallocator:
7200
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7201
    else:
7202
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
7203
      nodelist = [self.op.pnode]
7204
      if self.op.snode is not None:
7205
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
7206
        nodelist.append(self.op.snode)
7207
      self.needed_locks[locking.LEVEL_NODE] = nodelist
7208

    
7209
    # in case of import lock the source node too
7210
    if self.op.mode == constants.INSTANCE_IMPORT:
7211
      src_node = self.op.src_node
7212
      src_path = self.op.src_path
7213

    
7214
      if src_path is None:
7215
        self.op.src_path = src_path = self.op.instance_name
7216

    
7217
      if src_node is None:
7218
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
7219
        self.op.src_node = None
7220
        if os.path.isabs(src_path):
7221
          raise errors.OpPrereqError("Importing an instance from an absolute"
7222
                                     " path requires a source node option.",
7223
                                     errors.ECODE_INVAL)
7224
      else:
7225
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
7226
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
7227
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
7228
        if not os.path.isabs(src_path):
7229
          self.op.src_path = src_path = \
7230
            utils.PathJoin(constants.EXPORT_DIR, src_path)
7231

    
7232
  def _RunAllocator(self):
7233
    """Run the allocator based on input opcode.
7234

7235
    """
7236
    nics = [n.ToDict() for n in self.nics]
7237
    ial = IAllocator(self.cfg, self.rpc,
7238
                     mode=constants.IALLOCATOR_MODE_ALLOC,
7239
                     name=self.op.instance_name,
7240
                     disk_template=self.op.disk_template,
7241
                     tags=[],
7242
                     os=self.op.os_type,
7243
                     vcpus=self.be_full[constants.BE_VCPUS],
7244
                     mem_size=self.be_full[constants.BE_MEMORY],
7245
                     disks=self.disks,
7246
                     nics=nics,
7247
                     hypervisor=self.op.hypervisor,
7248
                     )
7249

    
7250
    ial.Run(self.op.iallocator)
7251

    
7252
    if not ial.success:
7253
      raise errors.OpPrereqError("Can't compute nodes using"
7254
                                 " iallocator '%s': %s" %
7255
                                 (self.op.iallocator, ial.info),
7256
                                 errors.ECODE_NORES)
7257
    if len(ial.result) != ial.required_nodes:
7258
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
7259
                                 " of nodes (%s), required %s" %
7260
                                 (self.op.iallocator, len(ial.result),
7261
                                  ial.required_nodes), errors.ECODE_FAULT)
7262
    self.op.pnode = ial.result[0]
7263
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
7264
                 self.op.instance_name, self.op.iallocator,
7265
                 utils.CommaJoin(ial.result))
7266
    if ial.required_nodes == 2:
7267
      self.op.snode = ial.result[1]
7268

    
7269
  def BuildHooksEnv(self):
7270
    """Build hooks env.
7271

7272
    This runs on master, primary and secondary nodes of the instance.
7273

7274
    """
7275
    env = {
7276
      "ADD_MODE": self.op.mode,
7277
      }
7278
    if self.op.mode == constants.INSTANCE_IMPORT:
7279
      env["SRC_NODE"] = self.op.src_node
7280
      env["SRC_PATH"] = self.op.src_path
7281
      env["SRC_IMAGES"] = self.src_images
7282

    
7283
    env.update(_BuildInstanceHookEnv(
7284
      name=self.op.instance_name,
7285
      primary_node=self.op.pnode,
7286
      secondary_nodes=self.secondaries,
7287
      status=self.op.start,
7288
      os_type=self.op.os_type,
7289
      memory=self.be_full[constants.BE_MEMORY],
7290
      vcpus=self.be_full[constants.BE_VCPUS],
7291
      nics=_NICListToTuple(self, self.nics),
7292
      disk_template=self.op.disk_template,
7293
      disks=[(d["size"], d["mode"]) for d in self.disks],
7294
      bep=self.be_full,
7295
      hvp=self.hv_full,
7296
      hypervisor_name=self.op.hypervisor,
7297
    ))
7298

    
7299
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
7300
          self.secondaries)
7301
    return env, nl, nl
7302

    
7303
  def _ReadExportInfo(self):
7304
    """Reads the export information from disk.
7305

7306
    It will override the opcode source node and path with the actual
7307
    information, if these two were not specified before.
7308

7309
    @return: the export information
7310

7311
    """
7312
    assert self.op.mode == constants.INSTANCE_IMPORT
7313

    
7314
    src_node = self.op.src_node
7315
    src_path = self.op.src_path
7316

    
7317
    if src_node is None:
7318
      locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
7319
      exp_list = self.rpc.call_export_list(locked_nodes)
7320
      found = False
7321
      for node in exp_list:
7322
        if exp_list[node].fail_msg:
7323
          continue
7324
        if src_path in exp_list[node].payload:
7325
          found = True
7326
          self.op.src_node = src_node = node
7327
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
7328
                                                       src_path)
7329
          break
7330
      if not found:
7331
        raise errors.OpPrereqError("No export found for relative path %s" %
7332
                                    src_path, errors.ECODE_INVAL)
7333

    
7334
    _CheckNodeOnline(self, src_node)
7335
    result = self.rpc.call_export_info(src_node, src_path)
7336
    result.Raise("No export or invalid export found in dir %s" % src_path)
7337

    
7338
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
7339
    if not export_info.has_section(constants.INISECT_EXP):
7340
      raise errors.ProgrammerError("Corrupted export config",
7341
                                   errors.ECODE_ENVIRON)
7342

    
7343
    ei_version = export_info.get(constants.INISECT_EXP, "version")
7344
    if (int(ei_version) != constants.EXPORT_VERSION):
7345
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
7346
                                 (ei_version, constants.EXPORT_VERSION),
7347
                                 errors.ECODE_ENVIRON)
7348
    return export_info
7349

    
7350
  def _ReadExportParams(self, einfo):
7351
    """Use export parameters as defaults.
7352

7353
    In case the opcode doesn't specify (as in override) some instance
7354
    parameters, then try to use them from the export information, if
7355
    that declares them.
7356

7357
    """
7358
    self.op.os_type = einfo.get(constants.INISECT_EXP, "os")
7359

    
7360
    if self.op.disk_template is None:
7361
      if einfo.has_option(constants.INISECT_INS, "disk_template"):
7362
        self.op.disk_template = einfo.get(constants.INISECT_INS,
7363
                                          "disk_template")
7364
      else:
7365
        raise errors.OpPrereqError("No disk template specified and the export"
7366
                                   " is missing the disk_template information",
7367
                                   errors.ECODE_INVAL)
7368

    
7369
    if not self.op.disks:
7370
      if einfo.has_option(constants.INISECT_INS, "disk_count"):
7371
        disks = []
7372
        # TODO: import the disk iv_name too
7373
        for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
7374
          disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
7375
          disks.append({"size": disk_sz})
7376
        self.op.disks = disks
7377
      else:
7378
        raise errors.OpPrereqError("No disk info specified and the export"
7379
                                   " is missing the disk information",
7380
                                   errors.ECODE_INVAL)
7381

    
7382
    if (not self.op.nics and
7383
        einfo.has_option(constants.INISECT_INS, "nic_count")):
7384
      nics = []
7385
      for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
7386
        ndict = {}
7387
        for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
7388
          v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
7389
          ndict[name] = v
7390
        nics.append(ndict)
7391
      self.op.nics = nics
7392

    
7393
    if (self.op.hypervisor is None and
7394
        einfo.has_option(constants.INISECT_INS, "hypervisor")):
7395
      self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
7396
    if einfo.has_section(constants.INISECT_HYP):
7397
      # use the export parameters but do not override the ones
7398
      # specified by the user
7399
      for name, value in einfo.items(constants.INISECT_HYP):
7400
        if name not in self.op.hvparams:
7401
          self.op.hvparams[name] = value
7402

    
7403
    if einfo.has_section(constants.INISECT_BEP):
7404
      # use the parameters, without overriding
7405
      for name, value in einfo.items(constants.INISECT_BEP):
7406
        if name not in self.op.beparams:
7407
          self.op.beparams[name] = value
7408
    else:
7409
      # try to read the parameters old style, from the main section
7410
      for name in constants.BES_PARAMETERS:
7411
        if (name not in self.op.beparams and
7412
            einfo.has_option(constants.INISECT_INS, name)):
7413
          self.op.beparams[name] = einfo.get(constants.INISECT_INS, name)
7414

    
7415
    if einfo.has_section(constants.INISECT_OSP):
7416
      # use the parameters, without overriding
7417
      for name, value in einfo.items(constants.INISECT_OSP):
7418
        if name not in self.op.osparams:
7419
          self.op.osparams[name] = value
7420

    
7421
  def _RevertToDefaults(self, cluster):
7422
    """Revert the instance parameters to the default values.
7423

7424
    """
7425
    # hvparams
7426
    hv_defs = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type, {})
7427
    for name in self.op.hvparams.keys():
7428
      if name in hv_defs and hv_defs[name] == self.op.hvparams[name]:
7429
        del self.op.hvparams[name]
7430
    # beparams
7431
    be_defs = cluster.SimpleFillBE({})
7432
    for name in self.op.beparams.keys():
7433
      if name in be_defs and be_defs[name] == self.op.beparams[name]:
7434
        del self.op.beparams[name]
7435
    # nic params
7436
    nic_defs = cluster.SimpleFillNIC({})
7437
    for nic in self.op.nics:
7438
      for name in constants.NICS_PARAMETERS:
7439
        if name in nic and name in nic_defs and nic[name] == nic_defs[name]:
7440
          del nic[name]
7441
    # osparams
7442
    os_defs = cluster.SimpleFillOS(self.op.os_type, {})
7443
    for name in self.op.osparams.keys():
7444
      if name in os_defs and os_defs[name] == self.op.osparams[name]:
7445
        del self.op.osparams[name]
7446

    
7447
  def CheckPrereq(self):
7448
    """Check prerequisites.
7449

7450
    """
7451
    if self.op.mode == constants.INSTANCE_IMPORT:
7452
      export_info = self._ReadExportInfo()
7453
      self._ReadExportParams(export_info)
7454

    
7455
    if (not self.cfg.GetVGName() and
7456
        self.op.disk_template not in constants.DTS_NOT_LVM):
7457
      raise errors.OpPrereqError("Cluster does not support lvm-based"
7458
                                 " instances", errors.ECODE_STATE)
7459

    
7460
    if self.op.hypervisor is None:
7461
      self.op.hypervisor = self.cfg.GetHypervisorType()
7462

    
7463
    cluster = self.cfg.GetClusterInfo()
7464
    enabled_hvs = cluster.enabled_hypervisors
7465
    if self.op.hypervisor not in enabled_hvs:
7466
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
7467
                                 " cluster (%s)" % (self.op.hypervisor,
7468
                                  ",".join(enabled_hvs)),
7469
                                 errors.ECODE_STATE)
7470

    
7471
    # check hypervisor parameter syntax (locally)
7472
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
7473
    filled_hvp = cluster.SimpleFillHV(self.op.hypervisor, self.op.os_type,
7474
                                      self.op.hvparams)
7475
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
7476
    hv_type.CheckParameterSyntax(filled_hvp)
7477
    self.hv_full = filled_hvp
7478
    # check that we don't specify global parameters on an instance
7479
    _CheckGlobalHvParams(self.op.hvparams)
7480

    
7481
    # fill and remember the beparams dict
7482
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
7483
    self.be_full = cluster.SimpleFillBE(self.op.beparams)
7484

    
7485
    # build os parameters
7486
    self.os_full = cluster.SimpleFillOS(self.op.os_type, self.op.osparams)
7487

    
7488
    # now that hvp/bep are in final format, let's reset to defaults,
7489
    # if told to do so
7490
    if self.op.identify_defaults:
7491
      self._RevertToDefaults(cluster)
7492

    
7493
    # NIC buildup
7494
    self.nics = []
7495
    for idx, nic in enumerate(self.op.nics):
7496
      nic_mode_req = nic.get("mode", None)
7497
      nic_mode = nic_mode_req
7498
      if nic_mode is None:
7499
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
7500

    
7501
      # in routed mode, for the first nic, the default ip is 'auto'
7502
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
7503
        default_ip_mode = constants.VALUE_AUTO
7504
      else:
7505
        default_ip_mode = constants.VALUE_NONE
7506

    
7507
      # ip validity checks
7508
      ip = nic.get("ip", default_ip_mode)
7509
      if ip is None or ip.lower() == constants.VALUE_NONE:
7510
        nic_ip = None
7511
      elif ip.lower() == constants.VALUE_AUTO:
7512
        if not self.op.name_check:
7513
          raise errors.OpPrereqError("IP address set to auto but name checks"
7514
                                     " have been skipped",
7515
                                     errors.ECODE_INVAL)
7516
        nic_ip = self.hostname1.ip
7517
      else:
7518
        if not netutils.IPAddress.IsValid(ip):
7519
          raise errors.OpPrereqError("Invalid IP address '%s'" % ip,
7520
                                     errors.ECODE_INVAL)
7521
        nic_ip = ip
7522

    
7523
      # TODO: check the ip address for uniqueness
7524
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
7525
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
7526
                                   errors.ECODE_INVAL)
7527

    
7528
      # MAC address verification
7529
      mac = nic.get("mac", constants.VALUE_AUTO)
7530
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7531
        mac = utils.NormalizeAndValidateMac(mac)
7532

    
7533
        try:
7534
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
7535
        except errors.ReservationError:
7536
          raise errors.OpPrereqError("MAC address %s already in use"
7537
                                     " in cluster" % mac,
7538
                                     errors.ECODE_NOTUNIQUE)
7539

    
7540
      # bridge verification
7541
      bridge = nic.get("bridge", None)
7542
      link = nic.get("link", None)
7543
      if bridge and link:
7544
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
7545
                                   " at the same time", errors.ECODE_INVAL)
7546
      elif bridge and nic_mode == constants.NIC_MODE_ROUTED:
7547
        raise errors.OpPrereqError("Cannot pass 'bridge' on a routed nic",
7548
                                   errors.ECODE_INVAL)
7549
      elif bridge:
7550
        link = bridge
7551

    
7552
      nicparams = {}
7553
      if nic_mode_req:
7554
        nicparams[constants.NIC_MODE] = nic_mode_req
7555
      if link:
7556
        nicparams[constants.NIC_LINK] = link
7557

    
7558
      check_params = cluster.SimpleFillNIC(nicparams)
7559
      objects.NIC.CheckParameterSyntax(check_params)
7560
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
7561

    
7562
    # disk checks/pre-build
7563
    self.disks = []
7564
    for disk in self.op.disks:
7565
      mode = disk.get("mode", constants.DISK_RDWR)
7566
      if mode not in constants.DISK_ACCESS_SET:
7567
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
7568
                                   mode, errors.ECODE_INVAL)
7569
      size = disk.get("size", None)
7570
      if size is None:
7571
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
7572
      try:
7573
        size = int(size)
7574
      except (TypeError, ValueError):
7575
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
7576
                                   errors.ECODE_INVAL)
7577
      data_vg = disk.get("vg", self.cfg.GetVGName())
7578
      meta_vg = disk.get("metavg", data_vg)
7579
      new_disk = {"size": size, "mode": mode, "vg": data_vg, "metavg": meta_vg}
7580
      if "adopt" in disk:
7581
        new_disk["adopt"] = disk["adopt"]
7582
      self.disks.append(new_disk)
7583

    
7584
    if self.op.mode == constants.INSTANCE_IMPORT:
7585

    
7586
      # Check that the new instance doesn't have less disks than the export
7587
      instance_disks = len(self.disks)
7588
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
7589
      if instance_disks < export_disks:
7590
        raise errors.OpPrereqError("Not enough disks to import."
7591
                                   " (instance: %d, export: %d)" %
7592
                                   (instance_disks, export_disks),
7593
                                   errors.ECODE_INVAL)
7594

    
7595
      disk_images = []
7596
      for idx in range(export_disks):
7597
        option = 'disk%d_dump' % idx
7598
        if export_info.has_option(constants.INISECT_INS, option):
7599
          # FIXME: are the old os-es, disk sizes, etc. useful?
7600
          export_name = export_info.get(constants.INISECT_INS, option)
7601
          image = utils.PathJoin(self.op.src_path, export_name)
7602
          disk_images.append(image)
7603
        else:
7604
          disk_images.append(False)
7605

    
7606
      self.src_images = disk_images
7607

    
7608
      old_name = export_info.get(constants.INISECT_INS, 'name')
7609
      try:
7610
        exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
7611
      except (TypeError, ValueError), err:
7612
        raise errors.OpPrereqError("Invalid export file, nic_count is not"
7613
                                   " an integer: %s" % str(err),
7614
                                   errors.ECODE_STATE)
7615
      if self.op.instance_name == old_name:
7616
        for idx, nic in enumerate(self.nics):
7617
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
7618
            nic_mac_ini = 'nic%d_mac' % idx
7619
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
7620

    
7621
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
7622

    
7623
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
7624
    if self.op.ip_check:
7625
      if netutils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
7626
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
7627
                                   (self.check_ip, self.op.instance_name),
7628
                                   errors.ECODE_NOTUNIQUE)
7629

    
7630
    #### mac address generation
7631
    # By generating here the mac address both the allocator and the hooks get
7632
    # the real final mac address rather than the 'auto' or 'generate' value.
7633
    # There is a race condition between the generation and the instance object
7634
    # creation, which means that we know the mac is valid now, but we're not
7635
    # sure it will be when we actually add the instance. If things go bad
7636
    # adding the instance will abort because of a duplicate mac, and the
7637
    # creation job will fail.
7638
    for nic in self.nics:
7639
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
7640
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
7641

    
7642
    #### allocator run
7643

    
7644
    if self.op.iallocator is not None:
7645
      self._RunAllocator()
7646

    
7647
    #### node related checks
7648

    
7649
    # check primary node
7650
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
7651
    assert self.pnode is not None, \
7652
      "Cannot retrieve locked node %s" % self.op.pnode
7653
    if pnode.offline:
7654
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
7655
                                 pnode.name, errors.ECODE_STATE)
7656
    if pnode.drained:
7657
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
7658
                                 pnode.name, errors.ECODE_STATE)
7659
    if not pnode.vm_capable:
7660
      raise errors.OpPrereqError("Cannot use non-vm_capable primary node"
7661
                                 " '%s'" % pnode.name, errors.ECODE_STATE)
7662

    
7663
    self.secondaries = []
7664

    
7665
    # mirror node verification
7666
    if self.op.disk_template in constants.DTS_NET_MIRROR:
7667
      if self.op.snode == pnode.name:
7668
        raise errors.OpPrereqError("The secondary node cannot be the"
7669
                                   " primary node.", errors.ECODE_INVAL)
7670
      _CheckNodeOnline(self, self.op.snode)
7671
      _CheckNodeNotDrained(self, self.op.snode)
7672
      _CheckNodeVmCapable(self, self.op.snode)
7673
      self.secondaries.append(self.op.snode)
7674

    
7675
    nodenames = [pnode.name] + self.secondaries
7676

    
7677
    if not self.adopt_disks:
7678
      # Check lv size requirements, if not adopting
7679
      req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
7680
      _CheckNodesFreeDiskPerVG(self, nodenames, req_sizes)
7681

    
7682
    else: # instead, we must check the adoption data
7683
      all_lvs = set([i["vg"] + "/" + i["adopt"] for i in self.disks])
7684
      if len(all_lvs) != len(self.disks):
7685
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
7686
                                   errors.ECODE_INVAL)
7687
      for lv_name in all_lvs:
7688
        try:
7689
          # FIXME: lv_name here is "vg/lv" need to ensure that other calls
7690
          # to ReserveLV uses the same syntax
7691
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
7692
        except errors.ReservationError:
7693
          raise errors.OpPrereqError("LV named %s used by another instance" %
7694
                                     lv_name, errors.ECODE_NOTUNIQUE)
7695

    
7696
      vg_names = self.rpc.call_vg_list([pnode.name])[pnode.name]
7697
      vg_names.Raise("Cannot get VG information from node %s" % pnode.name)
7698

    
7699
      node_lvs = self.rpc.call_lv_list([pnode.name],
7700
                                       vg_names.payload.keys())[pnode.name]
7701
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
7702
      node_lvs = node_lvs.payload
7703

    
7704
      delta = all_lvs.difference(node_lvs.keys())
7705
      if delta:
7706
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
7707
                                   utils.CommaJoin(delta),
7708
                                   errors.ECODE_INVAL)
7709
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
7710
      if online_lvs:
7711
        raise errors.OpPrereqError("Online logical volumes found, cannot"
7712
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
7713
                                   errors.ECODE_STATE)
7714
      # update the size of disk based on what is found
7715
      for dsk in self.disks:
7716
        dsk["size"] = int(float(node_lvs[dsk["vg"] + "/" + dsk["adopt"]][0]))
7717

    
7718
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
7719

    
7720
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
7721
    # check OS parameters (remotely)
7722
    _CheckOSParams(self, True, nodenames, self.op.os_type, self.os_full)
7723

    
7724
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
7725

    
7726
    # memory check on primary node
7727
    if self.op.start:
7728
      _CheckNodeFreeMemory(self, self.pnode.name,
7729
                           "creating instance %s" % self.op.instance_name,
7730
                           self.be_full[constants.BE_MEMORY],
7731
                           self.op.hypervisor)
7732

    
7733
    self.dry_run_result = list(nodenames)
7734

    
7735
  def Exec(self, feedback_fn):
7736
    """Create and add the instance to the cluster.
7737

7738
    """
7739
    instance = self.op.instance_name
7740
    pnode_name = self.pnode.name
7741

    
7742
    ht_kind = self.op.hypervisor
7743
    if ht_kind in constants.HTS_REQ_PORT:
7744
      network_port = self.cfg.AllocatePort()
7745
    else:
7746
      network_port = None
7747

    
7748
    if constants.ENABLE_FILE_STORAGE:
7749
      # this is needed because os.path.join does not accept None arguments
7750
      if self.op.file_storage_dir is None:
7751
        string_file_storage_dir = ""
7752
      else:
7753
        string_file_storage_dir = self.op.file_storage_dir
7754

    
7755
      # build the full file storage dir path
7756
      file_storage_dir = utils.PathJoin(self.cfg.GetFileStorageDir(),
7757
                                        string_file_storage_dir, instance)
7758
    else:
7759
      file_storage_dir = ""
7760

    
7761
    disks = _GenerateDiskTemplate(self,
7762
                                  self.op.disk_template,
7763
                                  instance, pnode_name,
7764
                                  self.secondaries,
7765
                                  self.disks,
7766
                                  file_storage_dir,
7767
                                  self.op.file_driver,
7768
                                  0,
7769
                                  feedback_fn)
7770

    
7771
    iobj = objects.Instance(name=instance, os=self.op.os_type,
7772
                            primary_node=pnode_name,
7773
                            nics=self.nics, disks=disks,
7774
                            disk_template=self.op.disk_template,
7775
                            admin_up=False,
7776
                            network_port=network_port,
7777
                            beparams=self.op.beparams,
7778
                            hvparams=self.op.hvparams,
7779
                            hypervisor=self.op.hypervisor,
7780
                            osparams=self.op.osparams,
7781
                            )
7782

    
7783
    if self.adopt_disks:
7784
      # rename LVs to the newly-generated names; we need to construct
7785
      # 'fake' LV disks with the old data, plus the new unique_id
7786
      tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
7787
      rename_to = []
7788
      for t_dsk, a_dsk in zip (tmp_disks, self.disks):
7789
        rename_to.append(t_dsk.logical_id)
7790
        t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk["adopt"])
7791
        self.cfg.SetDiskID(t_dsk, pnode_name)
7792
      result = self.rpc.call_blockdev_rename(pnode_name,
7793
                                             zip(tmp_disks, rename_to))
7794
      result.Raise("Failed to rename adoped LVs")
7795
    else:
7796
      feedback_fn("* creating instance disks...")
7797
      try:
7798
        _CreateDisks(self, iobj)
7799
      except errors.OpExecError:
7800
        self.LogWarning("Device creation failed, reverting...")
7801
        try:
7802
          _RemoveDisks(self, iobj)
7803
        finally:
7804
          self.cfg.ReleaseDRBDMinors(instance)
7805
          raise
7806

    
7807
    feedback_fn("adding instance %s to cluster config" % instance)
7808

    
7809
    self.cfg.AddInstance(iobj, self.proc.GetECId())
7810

    
7811
    # Declare that we don't want to remove the instance lock anymore, as we've
7812
    # added the instance to the config
7813
    del self.remove_locks[locking.LEVEL_INSTANCE]
7814
    # Unlock all the nodes
7815
    if self.op.mode == constants.INSTANCE_IMPORT:
7816
      nodes_keep = [self.op.src_node]
7817
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
7818
                       if node != self.op.src_node]
7819
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
7820
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
7821
    else:
7822
      self.context.glm.release(locking.LEVEL_NODE)
7823
      del self.acquired_locks[locking.LEVEL_NODE]
7824

    
7825
    disk_abort = False
7826
    if not self.adopt_disks and self.cfg.GetClusterInfo().prealloc_wipe_disks:
7827
      feedback_fn("* wiping instance disks...")
7828
      try:
7829
        _WipeDisks(self, iobj)
7830
      except errors.OpExecError, err:
7831
        logging.exception("Wiping disks failed")
7832
        self.LogWarning("Wiping instance disks failed (%s)", err)
7833
        disk_abort = True
7834

    
7835
    if disk_abort:
7836
      # Something is already wrong with the disks, don't do anything else
7837
      pass
7838
    elif self.op.wait_for_sync:
7839
      disk_abort = not _WaitForSync(self, iobj)
7840
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
7841
      # make sure the disks are not degraded (still sync-ing is ok)
7842
      time.sleep(15)
7843
      feedback_fn("* checking mirrors status")
7844
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
7845
    else:
7846
      disk_abort = False
7847

    
7848
    if disk_abort:
7849
      _RemoveDisks(self, iobj)
7850
      self.cfg.RemoveInstance(iobj.name)
7851
      # Make sure the instance lock gets removed
7852
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
7853
      raise errors.OpExecError("There are some degraded disks for"
7854
                               " this instance")
7855

    
7856
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
7857
      if self.op.mode == constants.INSTANCE_CREATE:
7858
        if not self.op.no_install:
7859
          feedback_fn("* running the instance OS create scripts...")
7860
          # FIXME: pass debug option from opcode to backend
7861
          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
7862
                                                 self.op.debug_level)
7863
          result.Raise("Could not add os for instance %s"
7864
                       " on node %s" % (instance, pnode_name))
7865

    
7866
      elif self.op.mode == constants.INSTANCE_IMPORT:
7867
        feedback_fn("* running the instance OS import scripts...")
7868

    
7869
        transfers = []
7870

    
7871
        for idx, image in enumerate(self.src_images):
7872
          if not image:
7873
            continue
7874

    
7875
          # FIXME: pass debug option from opcode to backend
7876
          dt = masterd.instance.DiskTransfer("disk/%s" % idx,
7877
                                             constants.IEIO_FILE, (image, ),
7878
                                             constants.IEIO_SCRIPT,
7879
                                             (iobj.disks[idx], idx),
7880
                                             None)
7881
          transfers.append(dt)
7882

    
7883
        import_result = \
7884
          masterd.instance.TransferInstanceData(self, feedback_fn,
7885
                                                self.op.src_node, pnode_name,
7886
                                                self.pnode.secondary_ip,
7887
                                                iobj, transfers)
7888
        if not compat.all(import_result):
7889
          self.LogWarning("Some disks for instance %s on node %s were not"
7890
                          " imported successfully" % (instance, pnode_name))
7891

    
7892
      elif self.op.mode == constants.INSTANCE_REMOTE_IMPORT:
7893
        feedback_fn("* preparing remote import...")
7894
        # The source cluster will stop the instance before attempting to make a
7895
        # connection. In some cases stopping an instance can take a long time,
7896
        # hence the shutdown timeout is added to the connection timeout.
7897
        connect_timeout = (constants.RIE_CONNECT_TIMEOUT +
7898
                           self.op.source_shutdown_timeout)
7899
        timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
7900

    
7901
        assert iobj.primary_node == self.pnode.name
7902
        disk_results = \
7903
          masterd.instance.RemoteImport(self, feedback_fn, iobj, self.pnode,
7904
                                        self.source_x509_ca,
7905
                                        self._cds, timeouts)
7906
        if not compat.all(disk_results):
7907
          # TODO: Should the instance still be started, even if some disks
7908
          # failed to import (valid for local imports, too)?
7909
          self.LogWarning("Some disks for instance %s on node %s were not"
7910
                          " imported successfully" % (instance, pnode_name))
7911

    
7912
        # Run rename script on newly imported instance
7913
        assert iobj.name == instance
7914
        feedback_fn("Running rename script for %s" % instance)
7915
        result = self.rpc.call_instance_run_rename(pnode_name, iobj,
7916
                                                   self.source_instance_name,
7917
                                                   self.op.debug_level)
7918
        if result.fail_msg:
7919
          self.LogWarning("Failed to run rename script for %s on node"
7920
                          " %s: %s" % (instance, pnode_name, result.fail_msg))
7921

    
7922
      else:
7923
        # also checked in the prereq part
7924
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
7925
                                     % self.op.mode)
7926

    
7927
    if self.op.start:
7928
      iobj.admin_up = True
7929
      self.cfg.Update(iobj, feedback_fn)
7930
      logging.info("Starting instance %s on node %s", instance, pnode_name)
7931
      feedback_fn("* starting instance...")
7932
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
7933
      result.Raise("Could not start instance")
7934

    
7935
    return list(iobj.all_nodes)
7936

    
7937

    
7938
class LUInstanceConsole(NoHooksLU):
7939
  """Connect to an instance's console.
7940

7941
  This is somewhat special in that it returns the command line that
7942
  you need to run on the master node in order to connect to the
7943
  console.
7944

7945
  """
7946
  REQ_BGL = False
7947

    
7948
  def ExpandNames(self):
7949
    self._ExpandAndLockInstance()
7950

    
7951
  def CheckPrereq(self):
7952
    """Check prerequisites.
7953

7954
    This checks that the instance is in the cluster.
7955

7956
    """
7957
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7958
    assert self.instance is not None, \
7959
      "Cannot retrieve locked instance %s" % self.op.instance_name
7960
    _CheckNodeOnline(self, self.instance.primary_node)
7961

    
7962
  def Exec(self, feedback_fn):
7963
    """Connect to the console of an instance
7964

7965
    """
7966
    instance = self.instance
7967
    node = instance.primary_node
7968

    
7969
    node_insts = self.rpc.call_instance_list([node],
7970
                                             [instance.hypervisor])[node]
7971
    node_insts.Raise("Can't get node information from %s" % node)
7972

    
7973
    if instance.name not in node_insts.payload:
7974
      if instance.admin_up:
7975
        state = "ERROR_down"
7976
      else:
7977
        state = "ADMIN_down"
7978
      raise errors.OpExecError("Instance %s is not running (state %s)" %
7979
                               (instance.name, state))
7980

    
7981
    logging.debug("Connecting to console of %s on %s", instance.name, node)
7982

    
7983
    return _GetInstanceConsole(self.cfg.GetClusterInfo(), instance)
7984

    
7985

    
7986
def _GetInstanceConsole(cluster, instance):
7987
  """Returns console information for an instance.
7988

7989
  @type cluster: L{objects.Cluster}
7990
  @type instance: L{objects.Instance}
7991
  @rtype: dict
7992

7993
  """
7994
  hyper = hypervisor.GetHypervisor(instance.hypervisor)
7995
  # beparams and hvparams are passed separately, to avoid editing the
7996
  # instance and then saving the defaults in the instance itself.
7997
  hvparams = cluster.FillHV(instance)
7998
  beparams = cluster.FillBE(instance)
7999
  console = hyper.GetInstanceConsole(instance, hvparams, beparams)
8000

    
8001
  assert console.instance == instance.name
8002
  assert console.Validate()
8003

    
8004
  return console.ToDict()
8005

    
8006

    
8007
class LUInstanceReplaceDisks(LogicalUnit):
8008
  """Replace the disks of an instance.
8009

8010
  """
8011
  HPATH = "mirrors-replace"
8012
  HTYPE = constants.HTYPE_INSTANCE
8013
  REQ_BGL = False
8014

    
8015
  def CheckArguments(self):
8016
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
8017
                                  self.op.iallocator)
8018

    
8019
  def ExpandNames(self):
8020
    self._ExpandAndLockInstance()
8021

    
8022
    if self.op.iallocator is not None:
8023
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8024

    
8025
    elif self.op.remote_node is not None:
8026
      remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8027
      self.op.remote_node = remote_node
8028

    
8029
      # Warning: do not remove the locking of the new secondary here
8030
      # unless DRBD8.AddChildren is changed to work in parallel;
8031
      # currently it doesn't since parallel invocations of
8032
      # FindUnusedMinor will conflict
8033
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
8034
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
8035

    
8036
    else:
8037
      self.needed_locks[locking.LEVEL_NODE] = []
8038
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8039

    
8040
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
8041
                                   self.op.iallocator, self.op.remote_node,
8042
                                   self.op.disks, False, self.op.early_release)
8043

    
8044
    self.tasklets = [self.replacer]
8045

    
8046
  def DeclareLocks(self, level):
8047
    # If we're not already locking all nodes in the set we have to declare the
8048
    # instance's primary/secondary nodes.
8049
    if (level == locking.LEVEL_NODE and
8050
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
8051
      self._LockInstancesNodes()
8052

    
8053
  def BuildHooksEnv(self):
8054
    """Build hooks env.
8055

8056
    This runs on the master, the primary and all the secondaries.
8057

8058
    """
8059
    instance = self.replacer.instance
8060
    env = {
8061
      "MODE": self.op.mode,
8062
      "NEW_SECONDARY": self.op.remote_node,
8063
      "OLD_SECONDARY": instance.secondary_nodes[0],
8064
      }
8065
    env.update(_BuildInstanceHookEnvByObject(self, instance))
8066
    nl = [
8067
      self.cfg.GetMasterNode(),
8068
      instance.primary_node,
8069
      ]
8070
    if self.op.remote_node is not None:
8071
      nl.append(self.op.remote_node)
8072
    return env, nl, nl
8073

    
8074

    
8075
class TLReplaceDisks(Tasklet):
8076
  """Replaces disks for an instance.
8077

8078
  Note: Locking is not within the scope of this class.
8079

8080
  """
8081
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
8082
               disks, delay_iallocator, early_release):
8083
    """Initializes this class.
8084

8085
    """
8086
    Tasklet.__init__(self, lu)
8087

    
8088
    # Parameters
8089
    self.instance_name = instance_name
8090
    self.mode = mode
8091
    self.iallocator_name = iallocator_name
8092
    self.remote_node = remote_node
8093
    self.disks = disks
8094
    self.delay_iallocator = delay_iallocator
8095
    self.early_release = early_release
8096

    
8097
    # Runtime data
8098
    self.instance = None
8099
    self.new_node = None
8100
    self.target_node = None
8101
    self.other_node = None
8102
    self.remote_node_info = None
8103
    self.node_secondary_ip = None
8104

    
8105
  @staticmethod
8106
  def CheckArguments(mode, remote_node, iallocator):
8107
    """Helper function for users of this class.
8108

8109
    """
8110
    # check for valid parameter combination
8111
    if mode == constants.REPLACE_DISK_CHG:
8112
      if remote_node is None and iallocator is None:
8113
        raise errors.OpPrereqError("When changing the secondary either an"
8114
                                   " iallocator script must be used or the"
8115
                                   " new node given", errors.ECODE_INVAL)
8116

    
8117
      if remote_node is not None and iallocator is not None:
8118
        raise errors.OpPrereqError("Give either the iallocator or the new"
8119
                                   " secondary, not both", errors.ECODE_INVAL)
8120

    
8121
    elif remote_node is not None or iallocator is not None:
8122
      # Not replacing the secondary
8123
      raise errors.OpPrereqError("The iallocator and new node options can"
8124
                                 " only be used when changing the"
8125
                                 " secondary node", errors.ECODE_INVAL)
8126

    
8127
  @staticmethod
8128
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
8129
    """Compute a new secondary node using an IAllocator.
8130

8131
    """
8132
    ial = IAllocator(lu.cfg, lu.rpc,
8133
                     mode=constants.IALLOCATOR_MODE_RELOC,
8134
                     name=instance_name,
8135
                     relocate_from=relocate_from)
8136

    
8137
    ial.Run(iallocator_name)
8138

    
8139
    if not ial.success:
8140
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
8141
                                 " %s" % (iallocator_name, ial.info),
8142
                                 errors.ECODE_NORES)
8143

    
8144
    if len(ial.result) != ial.required_nodes:
8145
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
8146
                                 " of nodes (%s), required %s" %
8147
                                 (iallocator_name,
8148
                                  len(ial.result), ial.required_nodes),
8149
                                 errors.ECODE_FAULT)
8150

    
8151
    remote_node_name = ial.result[0]
8152

    
8153
    lu.LogInfo("Selected new secondary for instance '%s': %s",
8154
               instance_name, remote_node_name)
8155

    
8156
    return remote_node_name
8157

    
8158
  def _FindFaultyDisks(self, node_name):
8159
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
8160
                                    node_name, True)
8161

    
8162
  def CheckPrereq(self):
8163
    """Check prerequisites.
8164

8165
    This checks that the instance is in the cluster.
8166

8167
    """
8168
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
8169
    assert instance is not None, \
8170
      "Cannot retrieve locked instance %s" % self.instance_name
8171

    
8172
    if instance.disk_template != constants.DT_DRBD8:
8173
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
8174
                                 " instances", errors.ECODE_INVAL)
8175

    
8176
    if len(instance.secondary_nodes) != 1:
8177
      raise errors.OpPrereqError("The instance has a strange layout,"
8178
                                 " expected one secondary but found %d" %
8179
                                 len(instance.secondary_nodes),
8180
                                 errors.ECODE_FAULT)
8181

    
8182
    if not self.delay_iallocator:
8183
      self._CheckPrereq2()
8184

    
8185
  def _CheckPrereq2(self):
8186
    """Check prerequisites, second part.
8187

8188
    This function should always be part of CheckPrereq. It was separated and is
8189
    now called from Exec because during node evacuation iallocator was only
8190
    called with an unmodified cluster model, not taking planned changes into
8191
    account.
8192

8193
    """
8194
    instance = self.instance
8195
    secondary_node = instance.secondary_nodes[0]
8196

    
8197
    if self.iallocator_name is None:
8198
      remote_node = self.remote_node
8199
    else:
8200
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
8201
                                       instance.name, instance.secondary_nodes)
8202

    
8203
    if remote_node is not None:
8204
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
8205
      assert self.remote_node_info is not None, \
8206
        "Cannot retrieve locked node %s" % remote_node
8207
    else:
8208
      self.remote_node_info = None
8209

    
8210
    if remote_node == self.instance.primary_node:
8211
      raise errors.OpPrereqError("The specified node is the primary node of"
8212
                                 " the instance.", errors.ECODE_INVAL)
8213

    
8214
    if remote_node == secondary_node:
8215
      raise errors.OpPrereqError("The specified node is already the"
8216
                                 " secondary node of the instance.",
8217
                                 errors.ECODE_INVAL)
8218

    
8219
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
8220
                                    constants.REPLACE_DISK_CHG):
8221
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
8222
                                 errors.ECODE_INVAL)
8223

    
8224
    if self.mode == constants.REPLACE_DISK_AUTO:
8225
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
8226
      faulty_secondary = self._FindFaultyDisks(secondary_node)
8227

    
8228
      if faulty_primary and faulty_secondary:
8229
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
8230
                                   " one node and can not be repaired"
8231
                                   " automatically" % self.instance_name,
8232
                                   errors.ECODE_STATE)
8233

    
8234
      if faulty_primary:
8235
        self.disks = faulty_primary
8236
        self.target_node = instance.primary_node
8237
        self.other_node = secondary_node
8238
        check_nodes = [self.target_node, self.other_node]
8239
      elif faulty_secondary:
8240
        self.disks = faulty_secondary
8241
        self.target_node = secondary_node
8242
        self.other_node = instance.primary_node
8243
        check_nodes = [self.target_node, self.other_node]
8244
      else:
8245
        self.disks = []
8246
        check_nodes = []
8247

    
8248
    else:
8249
      # Non-automatic modes
8250
      if self.mode == constants.REPLACE_DISK_PRI:
8251
        self.target_node = instance.primary_node
8252
        self.other_node = secondary_node
8253
        check_nodes = [self.target_node, self.other_node]
8254

    
8255
      elif self.mode == constants.REPLACE_DISK_SEC:
8256
        self.target_node = secondary_node
8257
        self.other_node = instance.primary_node
8258
        check_nodes = [self.target_node, self.other_node]
8259

    
8260
      elif self.mode == constants.REPLACE_DISK_CHG:
8261
        self.new_node = remote_node
8262
        self.other_node = instance.primary_node
8263
        self.target_node = secondary_node
8264
        check_nodes = [self.new_node, self.other_node]
8265

    
8266
        _CheckNodeNotDrained(self.lu, remote_node)
8267
        _CheckNodeVmCapable(self.lu, remote_node)
8268

    
8269
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
8270
        assert old_node_info is not None
8271
        if old_node_info.offline and not self.early_release:
8272
          # doesn't make sense to delay the release
8273
          self.early_release = True
8274
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
8275
                          " early-release mode", secondary_node)
8276

    
8277
      else:
8278
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
8279
                                     self.mode)
8280

    
8281
      # If not specified all disks should be replaced
8282
      if not self.disks:
8283
        self.disks = range(len(self.instance.disks))
8284

    
8285
    for node in check_nodes:
8286
      _CheckNodeOnline(self.lu, node)
8287

    
8288
    touched_nodes = frozenset([self.new_node, self.other_node,
8289
                               self.target_node])
8290

    
8291
    if self.lu.needed_locks[locking.LEVEL_NODE] == locking.ALL_SET:
8292
      # Release unneeded node locks
8293
      for name in self.lu.acquired_locks[locking.LEVEL_NODE]:
8294
        if name not in touched_nodes:
8295
          self._ReleaseNodeLock(name)
8296

    
8297
    # Check whether disks are valid
8298
    for disk_idx in self.disks:
8299
      instance.FindDisk(disk_idx)
8300

    
8301
    # Get secondary node IP addresses
8302
    self.node_secondary_ip = \
8303
      dict((node_name, self.cfg.GetNodeInfo(node_name).secondary_ip)
8304
           for node_name in touched_nodes
8305
           if node_name is not None)
8306

    
8307
  def Exec(self, feedback_fn):
8308
    """Execute disk replacement.
8309

8310
    This dispatches the disk replacement to the appropriate handler.
8311

8312
    """
8313
    if self.delay_iallocator:
8314
      self._CheckPrereq2()
8315

    
8316
    if (self.lu.needed_locks[locking.LEVEL_NODE] == locking.ALL_SET and
8317
        __debug__):
8318
      # Verify owned locks before starting operation
8319
      owned_locks = self.lu.context.glm.list_owned(locking.LEVEL_NODE)
8320
      assert set(owned_locks) == set(self.node_secondary_ip), \
8321
          "Not owning the correct locks: %s" % (owned_locks, )
8322

    
8323
    if not self.disks:
8324
      feedback_fn("No disks need replacement")
8325
      return
8326

    
8327
    feedback_fn("Replacing disk(s) %s for %s" %
8328
                (utils.CommaJoin(self.disks), self.instance.name))
8329

    
8330
    activate_disks = (not self.instance.admin_up)
8331

    
8332
    # Activate the instance disks if we're replacing them on a down instance
8333
    if activate_disks:
8334
      _StartInstanceDisks(self.lu, self.instance, True)
8335

    
8336
    try:
8337
      # Should we replace the secondary node?
8338
      if self.new_node is not None:
8339
        fn = self._ExecDrbd8Secondary
8340
      else:
8341
        fn = self._ExecDrbd8DiskOnly
8342

    
8343
      return fn(feedback_fn)
8344

    
8345
    finally:
8346
      # Deactivate the instance disks if we're replacing them on a
8347
      # down instance
8348
      if activate_disks:
8349
        _SafeShutdownInstanceDisks(self.lu, self.instance)
8350

    
8351
      if __debug__:
8352
        # Verify owned locks
8353
        owned_locks = self.lu.context.glm.list_owned(locking.LEVEL_NODE)
8354
        assert ((self.early_release and not owned_locks) or
8355
                (not self.early_release and
8356
                 set(owned_locks) == set(self.node_secondary_ip))), \
8357
          ("Not owning the correct locks, early_release=%s, owned=%r" %
8358
           (self.early_release, owned_locks))
8359

    
8360
  def _CheckVolumeGroup(self, nodes):
8361
    self.lu.LogInfo("Checking volume groups")
8362

    
8363
    vgname = self.cfg.GetVGName()
8364

    
8365
    # Make sure volume group exists on all involved nodes
8366
    results = self.rpc.call_vg_list(nodes)
8367
    if not results:
8368
      raise errors.OpExecError("Can't list volume groups on the nodes")
8369

    
8370
    for node in nodes:
8371
      res = results[node]
8372
      res.Raise("Error checking node %s" % node)
8373
      if vgname not in res.payload:
8374
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
8375
                                 (vgname, node))
8376

    
8377
  def _CheckDisksExistence(self, nodes):
8378
    # Check disk existence
8379
    for idx, dev in enumerate(self.instance.disks):
8380
      if idx not in self.disks:
8381
        continue
8382

    
8383
      for node in nodes:
8384
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
8385
        self.cfg.SetDiskID(dev, node)
8386

    
8387
        result = self.rpc.call_blockdev_find(node, dev)
8388

    
8389
        msg = result.fail_msg
8390
        if msg or not result.payload:
8391
          if not msg:
8392
            msg = "disk not found"
8393
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
8394
                                   (idx, node, msg))
8395

    
8396
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
8397
    for idx, dev in enumerate(self.instance.disks):
8398
      if idx not in self.disks:
8399
        continue
8400

    
8401
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
8402
                      (idx, node_name))
8403

    
8404
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
8405
                                   ldisk=ldisk):
8406
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
8407
                                 " replace disks for instance %s" %
8408
                                 (node_name, self.instance.name))
8409

    
8410
  def _CreateNewStorage(self, node_name):
8411
    iv_names = {}
8412

    
8413
    for idx, dev in enumerate(self.instance.disks):
8414
      if idx not in self.disks:
8415
        continue
8416

    
8417
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
8418

    
8419
      self.cfg.SetDiskID(dev, node_name)
8420

    
8421
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
8422
      names = _GenerateUniqueNames(self.lu, lv_names)
8423

    
8424
      vg_data = dev.children[0].logical_id[0]
8425
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
8426
                             logical_id=(vg_data, names[0]))
8427
      vg_meta = dev.children[1].logical_id[0]
8428
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
8429
                             logical_id=(vg_meta, names[1]))
8430

    
8431
      new_lvs = [lv_data, lv_meta]
8432
      old_lvs = dev.children
8433
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
8434

    
8435
      # we pass force_create=True to force the LVM creation
8436
      for new_lv in new_lvs:
8437
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
8438
                        _GetInstanceInfoText(self.instance), False)
8439

    
8440
    return iv_names
8441

    
8442
  def _CheckDevices(self, node_name, iv_names):
8443
    for name, (dev, _, _) in iv_names.iteritems():
8444
      self.cfg.SetDiskID(dev, node_name)
8445

    
8446
      result = self.rpc.call_blockdev_find(node_name, dev)
8447

    
8448
      msg = result.fail_msg
8449
      if msg or not result.payload:
8450
        if not msg:
8451
          msg = "disk not found"
8452
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
8453
                                 (name, msg))
8454

    
8455
      if result.payload.is_degraded:
8456
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
8457

    
8458
  def _RemoveOldStorage(self, node_name, iv_names):
8459
    for name, (_, old_lvs, _) in iv_names.iteritems():
8460
      self.lu.LogInfo("Remove logical volumes for %s" % name)
8461

    
8462
      for lv in old_lvs:
8463
        self.cfg.SetDiskID(lv, node_name)
8464

    
8465
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
8466
        if msg:
8467
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
8468
                             hint="remove unused LVs manually")
8469

    
8470
  def _ReleaseNodeLock(self, node_name):
8471
    """Releases the lock for a given node."""
8472
    self.lu.context.glm.release(locking.LEVEL_NODE, node_name)
8473

    
8474
  def _ExecDrbd8DiskOnly(self, feedback_fn):
8475
    """Replace a disk on the primary or secondary for DRBD 8.
8476

8477
    The algorithm for replace is quite complicated:
8478

8479
      1. for each disk to be replaced:
8480

8481
        1. create new LVs on the target node with unique names
8482
        1. detach old LVs from the drbd device
8483
        1. rename old LVs to name_replaced.<time_t>
8484
        1. rename new LVs to old LVs
8485
        1. attach the new LVs (with the old names now) to the drbd device
8486

8487
      1. wait for sync across all devices
8488

8489
      1. for each modified disk:
8490

8491
        1. remove old LVs (which have the name name_replaces.<time_t>)
8492

8493
    Failures are not very well handled.
8494

8495
    """
8496
    steps_total = 6
8497

    
8498
    # Step: check device activation
8499
    self.lu.LogStep(1, steps_total, "Check device existence")
8500
    self._CheckDisksExistence([self.other_node, self.target_node])
8501
    self._CheckVolumeGroup([self.target_node, self.other_node])
8502

    
8503
    # Step: check other node consistency
8504
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8505
    self._CheckDisksConsistency(self.other_node,
8506
                                self.other_node == self.instance.primary_node,
8507
                                False)
8508

    
8509
    # Step: create new storage
8510
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8511
    iv_names = self._CreateNewStorage(self.target_node)
8512

    
8513
    # Step: for each lv, detach+rename*2+attach
8514
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8515
    for dev, old_lvs, new_lvs in iv_names.itervalues():
8516
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
8517

    
8518
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
8519
                                                     old_lvs)
8520
      result.Raise("Can't detach drbd from local storage on node"
8521
                   " %s for device %s" % (self.target_node, dev.iv_name))
8522
      #dev.children = []
8523
      #cfg.Update(instance)
8524

    
8525
      # ok, we created the new LVs, so now we know we have the needed
8526
      # storage; as such, we proceed on the target node to rename
8527
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
8528
      # using the assumption that logical_id == physical_id (which in
8529
      # turn is the unique_id on that node)
8530

    
8531
      # FIXME(iustin): use a better name for the replaced LVs
8532
      temp_suffix = int(time.time())
8533
      ren_fn = lambda d, suff: (d.physical_id[0],
8534
                                d.physical_id[1] + "_replaced-%s" % suff)
8535

    
8536
      # Build the rename list based on what LVs exist on the node
8537
      rename_old_to_new = []
8538
      for to_ren in old_lvs:
8539
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
8540
        if not result.fail_msg and result.payload:
8541
          # device exists
8542
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
8543

    
8544
      self.lu.LogInfo("Renaming the old LVs on the target node")
8545
      result = self.rpc.call_blockdev_rename(self.target_node,
8546
                                             rename_old_to_new)
8547
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
8548

    
8549
      # Now we rename the new LVs to the old LVs
8550
      self.lu.LogInfo("Renaming the new LVs on the target node")
8551
      rename_new_to_old = [(new, old.physical_id)
8552
                           for old, new in zip(old_lvs, new_lvs)]
8553
      result = self.rpc.call_blockdev_rename(self.target_node,
8554
                                             rename_new_to_old)
8555
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
8556

    
8557
      for old, new in zip(old_lvs, new_lvs):
8558
        new.logical_id = old.logical_id
8559
        self.cfg.SetDiskID(new, self.target_node)
8560

    
8561
      for disk in old_lvs:
8562
        disk.logical_id = ren_fn(disk, temp_suffix)
8563
        self.cfg.SetDiskID(disk, self.target_node)
8564

    
8565
      # Now that the new lvs have the old name, we can add them to the device
8566
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
8567
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
8568
                                                  new_lvs)
8569
      msg = result.fail_msg
8570
      if msg:
8571
        for new_lv in new_lvs:
8572
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
8573
                                               new_lv).fail_msg
8574
          if msg2:
8575
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
8576
                               hint=("cleanup manually the unused logical"
8577
                                     "volumes"))
8578
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
8579

    
8580
      dev.children = new_lvs
8581

    
8582
      self.cfg.Update(self.instance, feedback_fn)
8583

    
8584
    cstep = 5
8585
    if self.early_release:
8586
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8587
      cstep += 1
8588
      self._RemoveOldStorage(self.target_node, iv_names)
8589
      # WARNING: we release both node locks here, do not do other RPCs
8590
      # than WaitForSync to the primary node
8591
      self._ReleaseNodeLock([self.target_node, self.other_node])
8592

    
8593
    # Wait for sync
8594
    # This can fail as the old devices are degraded and _WaitForSync
8595
    # does a combined result over all disks, so we don't check its return value
8596
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8597
    cstep += 1
8598
    _WaitForSync(self.lu, self.instance)
8599

    
8600
    # Check all devices manually
8601
    self._CheckDevices(self.instance.primary_node, iv_names)
8602

    
8603
    # Step: remove old storage
8604
    if not self.early_release:
8605
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8606
      cstep += 1
8607
      self._RemoveOldStorage(self.target_node, iv_names)
8608

    
8609
  def _ExecDrbd8Secondary(self, feedback_fn):
8610
    """Replace the secondary node for DRBD 8.
8611

8612
    The algorithm for replace is quite complicated:
8613
      - for all disks of the instance:
8614
        - create new LVs on the new node with same names
8615
        - shutdown the drbd device on the old secondary
8616
        - disconnect the drbd network on the primary
8617
        - create the drbd device on the new secondary
8618
        - network attach the drbd on the primary, using an artifice:
8619
          the drbd code for Attach() will connect to the network if it
8620
          finds a device which is connected to the good local disks but
8621
          not network enabled
8622
      - wait for sync across all devices
8623
      - remove all disks from the old secondary
8624

8625
    Failures are not very well handled.
8626

8627
    """
8628
    steps_total = 6
8629

    
8630
    # Step: check device activation
8631
    self.lu.LogStep(1, steps_total, "Check device existence")
8632
    self._CheckDisksExistence([self.instance.primary_node])
8633
    self._CheckVolumeGroup([self.instance.primary_node])
8634

    
8635
    # Step: check other node consistency
8636
    self.lu.LogStep(2, steps_total, "Check peer consistency")
8637
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
8638

    
8639
    # Step: create new storage
8640
    self.lu.LogStep(3, steps_total, "Allocate new storage")
8641
    for idx, dev in enumerate(self.instance.disks):
8642
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
8643
                      (self.new_node, idx))
8644
      # we pass force_create=True to force LVM creation
8645
      for new_lv in dev.children:
8646
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
8647
                        _GetInstanceInfoText(self.instance), False)
8648

    
8649
    # Step 4: dbrd minors and drbd setups changes
8650
    # after this, we must manually remove the drbd minors on both the
8651
    # error and the success paths
8652
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
8653
    minors = self.cfg.AllocateDRBDMinor([self.new_node
8654
                                         for dev in self.instance.disks],
8655
                                        self.instance.name)
8656
    logging.debug("Allocated minors %r", minors)
8657

    
8658
    iv_names = {}
8659
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
8660
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
8661
                      (self.new_node, idx))
8662
      # create new devices on new_node; note that we create two IDs:
8663
      # one without port, so the drbd will be activated without
8664
      # networking information on the new node at this stage, and one
8665
      # with network, for the latter activation in step 4
8666
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
8667
      if self.instance.primary_node == o_node1:
8668
        p_minor = o_minor1
8669
      else:
8670
        assert self.instance.primary_node == o_node2, "Three-node instance?"
8671
        p_minor = o_minor2
8672

    
8673
      new_alone_id = (self.instance.primary_node, self.new_node, None,
8674
                      p_minor, new_minor, o_secret)
8675
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
8676
                    p_minor, new_minor, o_secret)
8677

    
8678
      iv_names[idx] = (dev, dev.children, new_net_id)
8679
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
8680
                    new_net_id)
8681
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
8682
                              logical_id=new_alone_id,
8683
                              children=dev.children,
8684
                              size=dev.size)
8685
      try:
8686
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
8687
                              _GetInstanceInfoText(self.instance), False)
8688
      except errors.GenericError:
8689
        self.cfg.ReleaseDRBDMinors(self.instance.name)
8690
        raise
8691

    
8692
    # We have new devices, shutdown the drbd on the old secondary
8693
    for idx, dev in enumerate(self.instance.disks):
8694
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
8695
      self.cfg.SetDiskID(dev, self.target_node)
8696
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
8697
      if msg:
8698
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
8699
                           "node: %s" % (idx, msg),
8700
                           hint=("Please cleanup this device manually as"
8701
                                 " soon as possible"))
8702

    
8703
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
8704
    result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
8705
                                               self.node_secondary_ip,
8706
                                               self.instance.disks)\
8707
                                              [self.instance.primary_node]
8708

    
8709
    msg = result.fail_msg
8710
    if msg:
8711
      # detaches didn't succeed (unlikely)
8712
      self.cfg.ReleaseDRBDMinors(self.instance.name)
8713
      raise errors.OpExecError("Can't detach the disks from the network on"
8714
                               " old node: %s" % (msg,))
8715

    
8716
    # if we managed to detach at least one, we update all the disks of
8717
    # the instance to point to the new secondary
8718
    self.lu.LogInfo("Updating instance configuration")
8719
    for dev, _, new_logical_id in iv_names.itervalues():
8720
      dev.logical_id = new_logical_id
8721
      self.cfg.SetDiskID(dev, self.instance.primary_node)
8722

    
8723
    self.cfg.Update(self.instance, feedback_fn)
8724

    
8725
    # and now perform the drbd attach
8726
    self.lu.LogInfo("Attaching primary drbds to new secondary"
8727
                    " (standalone => connected)")
8728
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
8729
                                            self.new_node],
8730
                                           self.node_secondary_ip,
8731
                                           self.instance.disks,
8732
                                           self.instance.name,
8733
                                           False)
8734
    for to_node, to_result in result.items():
8735
      msg = to_result.fail_msg
8736
      if msg:
8737
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
8738
                           to_node, msg,
8739
                           hint=("please do a gnt-instance info to see the"
8740
                                 " status of disks"))
8741
    cstep = 5
8742
    if self.early_release:
8743
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8744
      cstep += 1
8745
      self._RemoveOldStorage(self.target_node, iv_names)
8746
      # WARNING: we release all node locks here, do not do other RPCs
8747
      # than WaitForSync to the primary node
8748
      self._ReleaseNodeLock([self.instance.primary_node,
8749
                             self.target_node,
8750
                             self.new_node])
8751

    
8752
    # Wait for sync
8753
    # This can fail as the old devices are degraded and _WaitForSync
8754
    # does a combined result over all disks, so we don't check its return value
8755
    self.lu.LogStep(cstep, steps_total, "Sync devices")
8756
    cstep += 1
8757
    _WaitForSync(self.lu, self.instance)
8758

    
8759
    # Check all devices manually
8760
    self._CheckDevices(self.instance.primary_node, iv_names)
8761

    
8762
    # Step: remove old storage
8763
    if not self.early_release:
8764
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
8765
      self._RemoveOldStorage(self.target_node, iv_names)
8766

    
8767

    
8768
class LURepairNodeStorage(NoHooksLU):
8769
  """Repairs the volume group on a node.
8770

8771
  """
8772
  REQ_BGL = False
8773

    
8774
  def CheckArguments(self):
8775
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
8776

    
8777
    storage_type = self.op.storage_type
8778

    
8779
    if (constants.SO_FIX_CONSISTENCY not in
8780
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
8781
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
8782
                                 " repaired" % storage_type,
8783
                                 errors.ECODE_INVAL)
8784

    
8785
  def ExpandNames(self):
8786
    self.needed_locks = {
8787
      locking.LEVEL_NODE: [self.op.node_name],
8788
      }
8789

    
8790
  def _CheckFaultyDisks(self, instance, node_name):
8791
    """Ensure faulty disks abort the opcode or at least warn."""
8792
    try:
8793
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
8794
                                  node_name, True):
8795
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
8796
                                   " node '%s'" % (instance.name, node_name),
8797
                                   errors.ECODE_STATE)
8798
    except errors.OpPrereqError, err:
8799
      if self.op.ignore_consistency:
8800
        self.proc.LogWarning(str(err.args[0]))
8801
      else:
8802
        raise
8803

    
8804
  def CheckPrereq(self):
8805
    """Check prerequisites.
8806

8807
    """
8808
    # Check whether any instance on this node has faulty disks
8809
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
8810
      if not inst.admin_up:
8811
        continue
8812
      check_nodes = set(inst.all_nodes)
8813
      check_nodes.discard(self.op.node_name)
8814
      for inst_node_name in check_nodes:
8815
        self._CheckFaultyDisks(inst, inst_node_name)
8816

    
8817
  def Exec(self, feedback_fn):
8818
    feedback_fn("Repairing storage unit '%s' on %s ..." %
8819
                (self.op.name, self.op.node_name))
8820

    
8821
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
8822
    result = self.rpc.call_storage_execute(self.op.node_name,
8823
                                           self.op.storage_type, st_args,
8824
                                           self.op.name,
8825
                                           constants.SO_FIX_CONSISTENCY)
8826
    result.Raise("Failed to repair storage unit '%s' on %s" %
8827
                 (self.op.name, self.op.node_name))
8828

    
8829

    
8830
class LUNodeEvacStrategy(NoHooksLU):
8831
  """Computes the node evacuation strategy.
8832

8833
  """
8834
  REQ_BGL = False
8835

    
8836
  def CheckArguments(self):
8837
    _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
8838

    
8839
  def ExpandNames(self):
8840
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
8841
    self.needed_locks = locks = {}
8842
    if self.op.remote_node is None:
8843
      locks[locking.LEVEL_NODE] = locking.ALL_SET
8844
    else:
8845
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8846
      locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
8847

    
8848
  def Exec(self, feedback_fn):
8849
    if self.op.remote_node is not None:
8850
      instances = []
8851
      for node in self.op.nodes:
8852
        instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
8853
      result = []
8854
      for i in instances:
8855
        if i.primary_node == self.op.remote_node:
8856
          raise errors.OpPrereqError("Node %s is the primary node of"
8857
                                     " instance %s, cannot use it as"
8858
                                     " secondary" %
8859
                                     (self.op.remote_node, i.name),
8860
                                     errors.ECODE_INVAL)
8861
        result.append([i.name, self.op.remote_node])
8862
    else:
8863
      ial = IAllocator(self.cfg, self.rpc,
8864
                       mode=constants.IALLOCATOR_MODE_MEVAC,
8865
                       evac_nodes=self.op.nodes)
8866
      ial.Run(self.op.iallocator, validate=True)
8867
      if not ial.success:
8868
        raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
8869
                                 errors.ECODE_NORES)
8870
      result = ial.result
8871
    return result
8872

    
8873

    
8874
class LUInstanceGrowDisk(LogicalUnit):
8875
  """Grow a disk of an instance.
8876

8877
  """
8878
  HPATH = "disk-grow"
8879
  HTYPE = constants.HTYPE_INSTANCE
8880
  REQ_BGL = False
8881

    
8882
  def ExpandNames(self):
8883
    self._ExpandAndLockInstance()
8884
    self.needed_locks[locking.LEVEL_NODE] = []
8885
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8886

    
8887
  def DeclareLocks(self, level):
8888
    if level == locking.LEVEL_NODE:
8889
      self._LockInstancesNodes()
8890

    
8891
  def BuildHooksEnv(self):
8892
    """Build hooks env.
8893

8894
    This runs on the master, the primary and all the secondaries.
8895

8896
    """
8897
    env = {
8898
      "DISK": self.op.disk,
8899
      "AMOUNT": self.op.amount,
8900
      }
8901
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
8902
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
8903
    return env, nl, nl
8904

    
8905
  def CheckPrereq(self):
8906
    """Check prerequisites.
8907

8908
    This checks that the instance is in the cluster.
8909

8910
    """
8911
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8912
    assert instance is not None, \
8913
      "Cannot retrieve locked instance %s" % self.op.instance_name
8914
    nodenames = list(instance.all_nodes)
8915
    for node in nodenames:
8916
      _CheckNodeOnline(self, node)
8917

    
8918
    self.instance = instance
8919

    
8920
    if instance.disk_template not in constants.DTS_GROWABLE:
8921
      raise errors.OpPrereqError("Instance's disk layout does not support"
8922
                                 " growing.", errors.ECODE_INVAL)
8923

    
8924
    self.disk = instance.FindDisk(self.op.disk)
8925

    
8926
    if instance.disk_template != constants.DT_FILE:
8927
      # TODO: check the free disk space for file, when that feature
8928
      # will be supported
8929
      _CheckNodesFreeDiskPerVG(self, nodenames,
8930
                               self.disk.ComputeGrowth(self.op.amount))
8931

    
8932
  def Exec(self, feedback_fn):
8933
    """Execute disk grow.
8934

8935
    """
8936
    instance = self.instance
8937
    disk = self.disk
8938

    
8939
    disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
8940
    if not disks_ok:
8941
      raise errors.OpExecError("Cannot activate block device to grow")
8942

    
8943
    for node in instance.all_nodes:
8944
      self.cfg.SetDiskID(disk, node)
8945
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
8946
      result.Raise("Grow request failed to node %s" % node)
8947

    
8948
      # TODO: Rewrite code to work properly
8949
      # DRBD goes into sync mode for a short amount of time after executing the
8950
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
8951
      # calling "resize" in sync mode fails. Sleeping for a short amount of
8952
      # time is a work-around.
8953
      time.sleep(5)
8954

    
8955
    disk.RecordGrow(self.op.amount)
8956
    self.cfg.Update(instance, feedback_fn)
8957
    if self.op.wait_for_sync:
8958
      disk_abort = not _WaitForSync(self, instance, disks=[disk])
8959
      if disk_abort:
8960
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
8961
                             " status.\nPlease check the instance.")
8962
      if not instance.admin_up:
8963
        _SafeShutdownInstanceDisks(self, instance, disks=[disk])
8964
    elif not instance.admin_up:
8965
      self.proc.LogWarning("Not shutting down the disk even if the instance is"
8966
                           " not supposed to be running because no wait for"
8967
                           " sync mode was requested.")
8968

    
8969

    
8970
class LUInstanceQueryData(NoHooksLU):
8971
  """Query runtime instance data.
8972

8973
  """
8974
  REQ_BGL = False
8975

    
8976
  def ExpandNames(self):
8977
    self.needed_locks = {}
8978

    
8979
    # Use locking if requested or when non-static information is wanted
8980
    if not (self.op.static or self.op.use_locking):
8981
      self.LogWarning("Non-static data requested, locks need to be acquired")
8982
      self.op.use_locking = True
8983

    
8984
    if self.op.instances or not self.op.use_locking:
8985
      # Expand instance names right here
8986
      self.wanted_names = _GetWantedInstances(self, self.op.instances)
8987
    else:
8988
      # Will use acquired locks
8989
      self.wanted_names = None
8990

    
8991
    if self.op.use_locking:
8992
      self.share_locks = dict.fromkeys(locking.LEVELS, 1)
8993

    
8994
      if self.wanted_names is None:
8995
        self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
8996
      else:
8997
        self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
8998

    
8999
      self.needed_locks[locking.LEVEL_NODE] = []
9000
      self.share_locks = dict.fromkeys(locking.LEVELS, 1)
9001
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9002

    
9003
  def DeclareLocks(self, level):
9004
    if self.op.use_locking and level == locking.LEVEL_NODE:
9005
      self._LockInstancesNodes()
9006

    
9007
  def CheckPrereq(self):
9008
    """Check prerequisites.
9009

9010
    This only checks the optional instance list against the existing names.
9011

9012
    """
9013
    if self.wanted_names is None:
9014
      assert self.op.use_locking, "Locking was not used"
9015
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
9016

    
9017
    self.wanted_instances = [self.cfg.GetInstanceInfo(name)
9018
                             for name in self.wanted_names]
9019

    
9020
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
9021
    """Returns the status of a block device
9022

9023
    """
9024
    if self.op.static or not node:
9025
      return None
9026

    
9027
    self.cfg.SetDiskID(dev, node)
9028

    
9029
    result = self.rpc.call_blockdev_find(node, dev)
9030
    if result.offline:
9031
      return None
9032

    
9033
    result.Raise("Can't compute disk status for %s" % instance_name)
9034

    
9035
    status = result.payload
9036
    if status is None:
9037
      return None
9038

    
9039
    return (status.dev_path, status.major, status.minor,
9040
            status.sync_percent, status.estimated_time,
9041
            status.is_degraded, status.ldisk_status)
9042

    
9043
  def _ComputeDiskStatus(self, instance, snode, dev):
9044
    """Compute block device status.
9045

9046
    """
9047
    if dev.dev_type in constants.LDS_DRBD:
9048
      # we change the snode then (otherwise we use the one passed in)
9049
      if dev.logical_id[0] == instance.primary_node:
9050
        snode = dev.logical_id[1]
9051
      else:
9052
        snode = dev.logical_id[0]
9053

    
9054
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
9055
                                              instance.name, dev)
9056
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
9057

    
9058
    if dev.children:
9059
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
9060
                      for child in dev.children]
9061
    else:
9062
      dev_children = []
9063

    
9064
    return {
9065
      "iv_name": dev.iv_name,
9066
      "dev_type": dev.dev_type,
9067
      "logical_id": dev.logical_id,
9068
      "physical_id": dev.physical_id,
9069
      "pstatus": dev_pstatus,
9070
      "sstatus": dev_sstatus,
9071
      "children": dev_children,
9072
      "mode": dev.mode,
9073
      "size": dev.size,
9074
      }
9075

    
9076
  def Exec(self, feedback_fn):
9077
    """Gather and return data"""
9078
    result = {}
9079

    
9080
    cluster = self.cfg.GetClusterInfo()
9081

    
9082
    for instance in self.wanted_instances:
9083
      if not self.op.static:
9084
        remote_info = self.rpc.call_instance_info(instance.primary_node,
9085
                                                  instance.name,
9086
                                                  instance.hypervisor)
9087
        remote_info.Raise("Error checking node %s" % instance.primary_node)
9088
        remote_info = remote_info.payload
9089
        if remote_info and "state" in remote_info:
9090
          remote_state = "up"
9091
        else:
9092
          remote_state = "down"
9093
      else:
9094
        remote_state = None
9095
      if instance.admin_up:
9096
        config_state = "up"
9097
      else:
9098
        config_state = "down"
9099

    
9100
      disks = [self._ComputeDiskStatus(instance, None, device)
9101
               for device in instance.disks]
9102

    
9103
      result[instance.name] = {
9104
        "name": instance.name,
9105
        "config_state": config_state,
9106
        "run_state": remote_state,
9107
        "pnode": instance.primary_node,
9108
        "snodes": instance.secondary_nodes,
9109
        "os": instance.os,
9110
        # this happens to be the same format used for hooks
9111
        "nics": _NICListToTuple(self, instance.nics),
9112
        "disk_template": instance.disk_template,
9113
        "disks": disks,
9114
        "hypervisor": instance.hypervisor,
9115
        "network_port": instance.network_port,
9116
        "hv_instance": instance.hvparams,
9117
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
9118
        "be_instance": instance.beparams,
9119
        "be_actual": cluster.FillBE(instance),
9120
        "os_instance": instance.osparams,
9121
        "os_actual": cluster.SimpleFillOS(instance.os, instance.osparams),
9122
        "serial_no": instance.serial_no,
9123
        "mtime": instance.mtime,
9124
        "ctime": instance.ctime,
9125
        "uuid": instance.uuid,
9126
        }
9127

    
9128
    return result
9129

    
9130

    
9131
class LUInstanceSetParams(LogicalUnit):
9132
  """Modifies an instances's parameters.
9133

9134
  """
9135
  HPATH = "instance-modify"
9136
  HTYPE = constants.HTYPE_INSTANCE
9137
  REQ_BGL = False
9138

    
9139
  def CheckArguments(self):
9140
    if not (self.op.nics or self.op.disks or self.op.disk_template or
9141
            self.op.hvparams or self.op.beparams or self.op.os_name):
9142
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
9143

    
9144
    if self.op.hvparams:
9145
      _CheckGlobalHvParams(self.op.hvparams)
9146

    
9147
    # Disk validation
9148
    disk_addremove = 0
9149
    for disk_op, disk_dict in self.op.disks:
9150
      utils.ForceDictType(disk_dict, constants.IDISK_PARAMS_TYPES)
9151
      if disk_op == constants.DDM_REMOVE:
9152
        disk_addremove += 1
9153
        continue
9154
      elif disk_op == constants.DDM_ADD:
9155
        disk_addremove += 1
9156
      else:
9157
        if not isinstance(disk_op, int):
9158
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
9159
        if not isinstance(disk_dict, dict):
9160
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
9161
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9162

    
9163
      if disk_op == constants.DDM_ADD:
9164
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
9165
        if mode not in constants.DISK_ACCESS_SET:
9166
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
9167
                                     errors.ECODE_INVAL)
9168
        size = disk_dict.get('size', None)
9169
        if size is None:
9170
          raise errors.OpPrereqError("Required disk parameter size missing",
9171
                                     errors.ECODE_INVAL)
9172
        try:
9173
          size = int(size)
9174
        except (TypeError, ValueError), err:
9175
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
9176
                                     str(err), errors.ECODE_INVAL)
9177
        disk_dict['size'] = size
9178
      else:
9179
        # modification of disk
9180
        if 'size' in disk_dict:
9181
          raise errors.OpPrereqError("Disk size change not possible, use"
9182
                                     " grow-disk", errors.ECODE_INVAL)
9183

    
9184
    if disk_addremove > 1:
9185
      raise errors.OpPrereqError("Only one disk add or remove operation"
9186
                                 " supported at a time", errors.ECODE_INVAL)
9187

    
9188
    if self.op.disks and self.op.disk_template is not None:
9189
      raise errors.OpPrereqError("Disk template conversion and other disk"
9190
                                 " changes not supported at the same time",
9191
                                 errors.ECODE_INVAL)
9192

    
9193
    if (self.op.disk_template and
9194
        self.op.disk_template in constants.DTS_NET_MIRROR and
9195
        self.op.remote_node is None):
9196
      raise errors.OpPrereqError("Changing the disk template to a mirrored"
9197
                                 " one requires specifying a secondary node",
9198
                                 errors.ECODE_INVAL)
9199

    
9200
    # NIC validation
9201
    nic_addremove = 0
9202
    for nic_op, nic_dict in self.op.nics:
9203
      utils.ForceDictType(nic_dict, constants.INIC_PARAMS_TYPES)
9204
      if nic_op == constants.DDM_REMOVE:
9205
        nic_addremove += 1
9206
        continue
9207
      elif nic_op == constants.DDM_ADD:
9208
        nic_addremove += 1
9209
      else:
9210
        if not isinstance(nic_op, int):
9211
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
9212
        if not isinstance(nic_dict, dict):
9213
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
9214
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
9215

    
9216
      # nic_dict should be a dict
9217
      nic_ip = nic_dict.get('ip', None)
9218
      if nic_ip is not None:
9219
        if nic_ip.lower() == constants.VALUE_NONE:
9220
          nic_dict['ip'] = None
9221
        else:
9222
          if not netutils.IPAddress.IsValid(nic_ip):
9223
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
9224
                                       errors.ECODE_INVAL)
9225

    
9226
      nic_bridge = nic_dict.get('bridge', None)
9227
      nic_link = nic_dict.get('link', None)
9228
      if nic_bridge and nic_link:
9229
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
9230
                                   " at the same time", errors.ECODE_INVAL)
9231
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
9232
        nic_dict['bridge'] = None
9233
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
9234
        nic_dict['link'] = None
9235

    
9236
      if nic_op == constants.DDM_ADD:
9237
        nic_mac = nic_dict.get('mac', None)
9238
        if nic_mac is None:
9239
          nic_dict['mac'] = constants.VALUE_AUTO
9240

    
9241
      if 'mac' in nic_dict:
9242
        nic_mac = nic_dict['mac']
9243
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9244
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
9245

    
9246
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
9247
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
9248
                                     " modifying an existing nic",
9249
                                     errors.ECODE_INVAL)
9250

    
9251
    if nic_addremove > 1:
9252
      raise errors.OpPrereqError("Only one NIC add or remove operation"
9253
                                 " supported at a time", errors.ECODE_INVAL)
9254

    
9255
  def ExpandNames(self):
9256
    self._ExpandAndLockInstance()
9257
    self.needed_locks[locking.LEVEL_NODE] = []
9258
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
9259

    
9260
  def DeclareLocks(self, level):
9261
    if level == locking.LEVEL_NODE:
9262
      self._LockInstancesNodes()
9263
      if self.op.disk_template and self.op.remote_node:
9264
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
9265
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
9266

    
9267
  def BuildHooksEnv(self):
9268
    """Build hooks env.
9269

9270
    This runs on the master, primary and secondaries.
9271

9272
    """
9273
    args = dict()
9274
    if constants.BE_MEMORY in self.be_new:
9275
      args['memory'] = self.be_new[constants.BE_MEMORY]
9276
    if constants.BE_VCPUS in self.be_new:
9277
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
9278
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
9279
    # information at all.
9280
    if self.op.nics:
9281
      args['nics'] = []
9282
      nic_override = dict(self.op.nics)
9283
      for idx, nic in enumerate(self.instance.nics):
9284
        if idx in nic_override:
9285
          this_nic_override = nic_override[idx]
9286
        else:
9287
          this_nic_override = {}
9288
        if 'ip' in this_nic_override:
9289
          ip = this_nic_override['ip']
9290
        else:
9291
          ip = nic.ip
9292
        if 'mac' in this_nic_override:
9293
          mac = this_nic_override['mac']
9294
        else:
9295
          mac = nic.mac
9296
        if idx in self.nic_pnew:
9297
          nicparams = self.nic_pnew[idx]
9298
        else:
9299
          nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
9300
        mode = nicparams[constants.NIC_MODE]
9301
        link = nicparams[constants.NIC_LINK]
9302
        args['nics'].append((ip, mac, mode, link))
9303
      if constants.DDM_ADD in nic_override:
9304
        ip = nic_override[constants.DDM_ADD].get('ip', None)
9305
        mac = nic_override[constants.DDM_ADD]['mac']
9306
        nicparams = self.nic_pnew[constants.DDM_ADD]
9307
        mode = nicparams[constants.NIC_MODE]
9308
        link = nicparams[constants.NIC_LINK]
9309
        args['nics'].append((ip, mac, mode, link))
9310
      elif constants.DDM_REMOVE in nic_override:
9311
        del args['nics'][-1]
9312

    
9313
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
9314
    if self.op.disk_template:
9315
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
9316
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
9317
    return env, nl, nl
9318

    
9319
  def CheckPrereq(self):
9320
    """Check prerequisites.
9321

9322
    This only checks the instance list against the existing names.
9323

9324
    """
9325
    # checking the new params on the primary/secondary nodes
9326

    
9327
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
9328
    cluster = self.cluster = self.cfg.GetClusterInfo()
9329
    assert self.instance is not None, \
9330
      "Cannot retrieve locked instance %s" % self.op.instance_name
9331
    pnode = instance.primary_node
9332
    nodelist = list(instance.all_nodes)
9333

    
9334
    # OS change
9335
    if self.op.os_name and not self.op.force:
9336
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
9337
                      self.op.force_variant)
9338
      instance_os = self.op.os_name
9339
    else:
9340
      instance_os = instance.os
9341

    
9342
    if self.op.disk_template:
9343
      if instance.disk_template == self.op.disk_template:
9344
        raise errors.OpPrereqError("Instance already has disk template %s" %
9345
                                   instance.disk_template, errors.ECODE_INVAL)
9346

    
9347
      if (instance.disk_template,
9348
          self.op.disk_template) not in self._DISK_CONVERSIONS:
9349
        raise errors.OpPrereqError("Unsupported disk template conversion from"
9350
                                   " %s to %s" % (instance.disk_template,
9351
                                                  self.op.disk_template),
9352
                                   errors.ECODE_INVAL)
9353
      _CheckInstanceDown(self, instance, "cannot change disk template")
9354
      if self.op.disk_template in constants.DTS_NET_MIRROR:
9355
        if self.op.remote_node == pnode:
9356
          raise errors.OpPrereqError("Given new secondary node %s is the same"
9357
                                     " as the primary node of the instance" %
9358
                                     self.op.remote_node, errors.ECODE_STATE)
9359
        _CheckNodeOnline(self, self.op.remote_node)
9360
        _CheckNodeNotDrained(self, self.op.remote_node)
9361
        # FIXME: here we assume that the old instance type is DT_PLAIN
9362
        assert instance.disk_template == constants.DT_PLAIN
9363
        disks = [{"size": d.size, "vg": d.logical_id[0]}
9364
                 for d in instance.disks]
9365
        required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
9366
        _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
9367

    
9368
    # hvparams processing
9369
    if self.op.hvparams:
9370
      hv_type = instance.hypervisor
9371
      i_hvdict = _GetUpdatedParams(instance.hvparams, self.op.hvparams)
9372
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
9373
      hv_new = cluster.SimpleFillHV(hv_type, instance.os, i_hvdict)
9374

    
9375
      # local check
9376
      hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
9377
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
9378
      self.hv_new = hv_new # the new actual values
9379
      self.hv_inst = i_hvdict # the new dict (without defaults)
9380
    else:
9381
      self.hv_new = self.hv_inst = {}
9382

    
9383
    # beparams processing
9384
    if self.op.beparams:
9385
      i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
9386
                                   use_none=True)
9387
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
9388
      be_new = cluster.SimpleFillBE(i_bedict)
9389
      self.be_new = be_new # the new actual values
9390
      self.be_inst = i_bedict # the new dict (without defaults)
9391
    else:
9392
      self.be_new = self.be_inst = {}
9393
    be_old = cluster.FillBE(instance)
9394

    
9395
    # osparams processing
9396
    if self.op.osparams:
9397
      i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
9398
      _CheckOSParams(self, True, nodelist, instance_os, i_osdict)
9399
      self.os_inst = i_osdict # the new dict (without defaults)
9400
    else:
9401
      self.os_inst = {}
9402

    
9403
    self.warn = []
9404

    
9405
    if (constants.BE_MEMORY in self.op.beparams and not self.op.force and
9406
        be_new[constants.BE_MEMORY] > be_old[constants.BE_MEMORY]):
9407
      mem_check_list = [pnode]
9408
      if be_new[constants.BE_AUTO_BALANCE]:
9409
        # either we changed auto_balance to yes or it was from before
9410
        mem_check_list.extend(instance.secondary_nodes)
9411
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
9412
                                                  instance.hypervisor)
9413
      nodeinfo = self.rpc.call_node_info(mem_check_list, None,
9414
                                         instance.hypervisor)
9415
      pninfo = nodeinfo[pnode]
9416
      msg = pninfo.fail_msg
9417
      if msg:
9418
        # Assume the primary node is unreachable and go ahead
9419
        self.warn.append("Can't get info from primary node %s: %s" %
9420
                         (pnode,  msg))
9421
      elif not isinstance(pninfo.payload.get('memory_free', None), int):
9422
        self.warn.append("Node data from primary node %s doesn't contain"
9423
                         " free memory information" % pnode)
9424
      elif instance_info.fail_msg:
9425
        self.warn.append("Can't get instance runtime information: %s" %
9426
                        instance_info.fail_msg)
9427
      else:
9428
        if instance_info.payload:
9429
          current_mem = int(instance_info.payload['memory'])
9430
        else:
9431
          # Assume instance not running
9432
          # (there is a slight race condition here, but it's not very probable,
9433
          # and we have no other way to check)
9434
          current_mem = 0
9435
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
9436
                    pninfo.payload['memory_free'])
9437
        if miss_mem > 0:
9438
          raise errors.OpPrereqError("This change will prevent the instance"
9439
                                     " from starting, due to %d MB of memory"
9440
                                     " missing on its primary node" % miss_mem,
9441
                                     errors.ECODE_NORES)
9442

    
9443
      if be_new[constants.BE_AUTO_BALANCE]:
9444
        for node, nres in nodeinfo.items():
9445
          if node not in instance.secondary_nodes:
9446
            continue
9447
          nres.Raise("Can't get info from secondary node %s" % node,
9448
                     prereq=True, ecode=errors.ECODE_STATE)
9449
          if not isinstance(nres.payload.get('memory_free', None), int):
9450
            raise errors.OpPrereqError("Secondary node %s didn't return free"
9451
                                       " memory information" % node,
9452
                                       errors.ECODE_STATE)
9453
          elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
9454
            raise errors.OpPrereqError("This change will prevent the instance"
9455
                                       " from failover to its secondary node"
9456
                                       " %s, due to not enough memory" % node,
9457
                                       errors.ECODE_STATE)
9458

    
9459
    # NIC processing
9460
    self.nic_pnew = {}
9461
    self.nic_pinst = {}
9462
    for nic_op, nic_dict in self.op.nics:
9463
      if nic_op == constants.DDM_REMOVE:
9464
        if not instance.nics:
9465
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
9466
                                     errors.ECODE_INVAL)
9467
        continue
9468
      if nic_op != constants.DDM_ADD:
9469
        # an existing nic
9470
        if not instance.nics:
9471
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
9472
                                     " no NICs" % nic_op,
9473
                                     errors.ECODE_INVAL)
9474
        if nic_op < 0 or nic_op >= len(instance.nics):
9475
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
9476
                                     " are 0 to %d" %
9477
                                     (nic_op, len(instance.nics) - 1),
9478
                                     errors.ECODE_INVAL)
9479
        old_nic_params = instance.nics[nic_op].nicparams
9480
        old_nic_ip = instance.nics[nic_op].ip
9481
      else:
9482
        old_nic_params = {}
9483
        old_nic_ip = None
9484

    
9485
      update_params_dict = dict([(key, nic_dict[key])
9486
                                 for key in constants.NICS_PARAMETERS
9487
                                 if key in nic_dict])
9488

    
9489
      if 'bridge' in nic_dict:
9490
        update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
9491

    
9492
      new_nic_params = _GetUpdatedParams(old_nic_params,
9493
                                         update_params_dict)
9494
      utils.ForceDictType(new_nic_params, constants.NICS_PARAMETER_TYPES)
9495
      new_filled_nic_params = cluster.SimpleFillNIC(new_nic_params)
9496
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
9497
      self.nic_pinst[nic_op] = new_nic_params
9498
      self.nic_pnew[nic_op] = new_filled_nic_params
9499
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
9500

    
9501
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
9502
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
9503
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
9504
        if msg:
9505
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
9506
          if self.op.force:
9507
            self.warn.append(msg)
9508
          else:
9509
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
9510
      if new_nic_mode == constants.NIC_MODE_ROUTED:
9511
        if 'ip' in nic_dict:
9512
          nic_ip = nic_dict['ip']
9513
        else:
9514
          nic_ip = old_nic_ip
9515
        if nic_ip is None:
9516
          raise errors.OpPrereqError('Cannot set the nic ip to None'
9517
                                     ' on a routed nic', errors.ECODE_INVAL)
9518
      if 'mac' in nic_dict:
9519
        nic_mac = nic_dict['mac']
9520
        if nic_mac is None:
9521
          raise errors.OpPrereqError('Cannot set the nic mac to None',
9522
                                     errors.ECODE_INVAL)
9523
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
9524
          # otherwise generate the mac
9525
          nic_dict['mac'] = self.cfg.GenerateMAC(self.proc.GetECId())
9526
        else:
9527
          # or validate/reserve the current one
9528
          try:
9529
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
9530
          except errors.ReservationError:
9531
            raise errors.OpPrereqError("MAC address %s already in use"
9532
                                       " in cluster" % nic_mac,
9533
                                       errors.ECODE_NOTUNIQUE)
9534

    
9535
    # DISK processing
9536
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
9537
      raise errors.OpPrereqError("Disk operations not supported for"
9538
                                 " diskless instances",
9539
                                 errors.ECODE_INVAL)
9540
    for disk_op, _ in self.op.disks:
9541
      if disk_op == constants.DDM_REMOVE:
9542
        if len(instance.disks) == 1:
9543
          raise errors.OpPrereqError("Cannot remove the last disk of"
9544
                                     " an instance", errors.ECODE_INVAL)
9545
        _CheckInstanceDown(self, instance, "cannot remove disks")
9546

    
9547
      if (disk_op == constants.DDM_ADD and
9548
          len(instance.disks) >= constants.MAX_DISKS):
9549
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
9550
                                   " add more" % constants.MAX_DISKS,
9551
                                   errors.ECODE_STATE)
9552
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
9553
        # an existing disk
9554
        if disk_op < 0 or disk_op >= len(instance.disks):
9555
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
9556
                                     " are 0 to %d" %
9557
                                     (disk_op, len(instance.disks)),
9558
                                     errors.ECODE_INVAL)
9559

    
9560
    return
9561

    
9562
  def _ConvertPlainToDrbd(self, feedback_fn):
9563
    """Converts an instance from plain to drbd.
9564

9565
    """
9566
    feedback_fn("Converting template to drbd")
9567
    instance = self.instance
9568
    pnode = instance.primary_node
9569
    snode = self.op.remote_node
9570

    
9571
    # create a fake disk info for _GenerateDiskTemplate
9572
    disk_info = [{"size": d.size, "mode": d.mode,
9573
                  "vg": d.logical_id[0]} for d in instance.disks]
9574
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
9575
                                      instance.name, pnode, [snode],
9576
                                      disk_info, None, None, 0, feedback_fn)
9577
    info = _GetInstanceInfoText(instance)
9578
    feedback_fn("Creating aditional volumes...")
9579
    # first, create the missing data and meta devices
9580
    for disk in new_disks:
9581
      # unfortunately this is... not too nice
9582
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
9583
                            info, True)
9584
      for child in disk.children:
9585
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
9586
    # at this stage, all new LVs have been created, we can rename the
9587
    # old ones
9588
    feedback_fn("Renaming original volumes...")
9589
    rename_list = [(o, n.children[0].logical_id)
9590
                   for (o, n) in zip(instance.disks, new_disks)]
9591
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
9592
    result.Raise("Failed to rename original LVs")
9593

    
9594
    feedback_fn("Initializing DRBD devices...")
9595
    # all child devices are in place, we can now create the DRBD devices
9596
    for disk in new_disks:
9597
      for node in [pnode, snode]:
9598
        f_create = node == pnode
9599
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
9600

    
9601
    # at this point, the instance has been modified
9602
    instance.disk_template = constants.DT_DRBD8
9603
    instance.disks = new_disks
9604
    self.cfg.Update(instance, feedback_fn)
9605

    
9606
    # disks are created, waiting for sync
9607
    disk_abort = not _WaitForSync(self, instance,
9608
                                  oneshot=not self.op.wait_for_sync)
9609
    if disk_abort:
9610
      raise errors.OpExecError("There are some degraded disks for"
9611
                               " this instance, please cleanup manually")
9612

    
9613
  def _ConvertDrbdToPlain(self, feedback_fn):
9614
    """Converts an instance from drbd to plain.
9615

9616
    """
9617
    instance = self.instance
9618
    assert len(instance.secondary_nodes) == 1
9619
    pnode = instance.primary_node
9620
    snode = instance.secondary_nodes[0]
9621
    feedback_fn("Converting template to plain")
9622

    
9623
    old_disks = instance.disks
9624
    new_disks = [d.children[0] for d in old_disks]
9625

    
9626
    # copy over size and mode
9627
    for parent, child in zip(old_disks, new_disks):
9628
      child.size = parent.size
9629
      child.mode = parent.mode
9630

    
9631
    # update instance structure
9632
    instance.disks = new_disks
9633
    instance.disk_template = constants.DT_PLAIN
9634
    self.cfg.Update(instance, feedback_fn)
9635

    
9636
    feedback_fn("Removing volumes on the secondary node...")
9637
    for disk in old_disks:
9638
      self.cfg.SetDiskID(disk, snode)
9639
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
9640
      if msg:
9641
        self.LogWarning("Could not remove block device %s on node %s,"
9642
                        " continuing anyway: %s", disk.iv_name, snode, msg)
9643

    
9644
    feedback_fn("Removing unneeded volumes on the primary node...")
9645
    for idx, disk in enumerate(old_disks):
9646
      meta = disk.children[1]
9647
      self.cfg.SetDiskID(meta, pnode)
9648
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
9649
      if msg:
9650
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
9651
                        " continuing anyway: %s", idx, pnode, msg)
9652

    
9653
  def Exec(self, feedback_fn):
9654
    """Modifies an instance.
9655

9656
    All parameters take effect only at the next restart of the instance.
9657

9658
    """
9659
    # Process here the warnings from CheckPrereq, as we don't have a
9660
    # feedback_fn there.
9661
    for warn in self.warn:
9662
      feedback_fn("WARNING: %s" % warn)
9663

    
9664
    result = []
9665
    instance = self.instance
9666
    # disk changes
9667
    for disk_op, disk_dict in self.op.disks:
9668
      if disk_op == constants.DDM_REMOVE:
9669
        # remove the last disk
9670
        device = instance.disks.pop()
9671
        device_idx = len(instance.disks)
9672
        for node, disk in device.ComputeNodeTree(instance.primary_node):
9673
          self.cfg.SetDiskID(disk, node)
9674
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
9675
          if msg:
9676
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
9677
                            " continuing anyway", device_idx, node, msg)
9678
        result.append(("disk/%d" % device_idx, "remove"))
9679
      elif disk_op == constants.DDM_ADD:
9680
        # add a new disk
9681
        if instance.disk_template == constants.DT_FILE:
9682
          file_driver, file_path = instance.disks[0].logical_id
9683
          file_path = os.path.dirname(file_path)
9684
        else:
9685
          file_driver = file_path = None
9686
        disk_idx_base = len(instance.disks)
9687
        new_disk = _GenerateDiskTemplate(self,
9688
                                         instance.disk_template,
9689
                                         instance.name, instance.primary_node,
9690
                                         instance.secondary_nodes,
9691
                                         [disk_dict],
9692
                                         file_path,
9693
                                         file_driver,
9694
                                         disk_idx_base, feedback_fn)[0]
9695
        instance.disks.append(new_disk)
9696
        info = _GetInstanceInfoText(instance)
9697

    
9698
        logging.info("Creating volume %s for instance %s",
9699
                     new_disk.iv_name, instance.name)
9700
        # Note: this needs to be kept in sync with _CreateDisks
9701
        #HARDCODE
9702
        for node in instance.all_nodes:
9703
          f_create = node == instance.primary_node
9704
          try:
9705
            _CreateBlockDev(self, node, instance, new_disk,
9706
                            f_create, info, f_create)
9707
          except errors.OpExecError, err:
9708
            self.LogWarning("Failed to create volume %s (%s) on"
9709
                            " node %s: %s",
9710
                            new_disk.iv_name, new_disk, node, err)
9711
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
9712
                       (new_disk.size, new_disk.mode)))
9713
      else:
9714
        # change a given disk
9715
        instance.disks[disk_op].mode = disk_dict['mode']
9716
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
9717

    
9718
    if self.op.disk_template:
9719
      r_shut = _ShutdownInstanceDisks(self, instance)
9720
      if not r_shut:
9721
        raise errors.OpExecError("Cannot shutdown instance disks, unable to"
9722
                                 " proceed with disk template conversion")
9723
      mode = (instance.disk_template, self.op.disk_template)
9724
      try:
9725
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
9726
      except:
9727
        self.cfg.ReleaseDRBDMinors(instance.name)
9728
        raise
9729
      result.append(("disk_template", self.op.disk_template))
9730

    
9731
    # NIC changes
9732
    for nic_op, nic_dict in self.op.nics:
9733
      if nic_op == constants.DDM_REMOVE:
9734
        # remove the last nic
9735
        del instance.nics[-1]
9736
        result.append(("nic.%d" % len(instance.nics), "remove"))
9737
      elif nic_op == constants.DDM_ADD:
9738
        # mac and bridge should be set, by now
9739
        mac = nic_dict['mac']
9740
        ip = nic_dict.get('ip', None)
9741
        nicparams = self.nic_pinst[constants.DDM_ADD]
9742
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
9743
        instance.nics.append(new_nic)
9744
        result.append(("nic.%d" % (len(instance.nics) - 1),
9745
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
9746
                       (new_nic.mac, new_nic.ip,
9747
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
9748
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
9749
                       )))
9750
      else:
9751
        for key in 'mac', 'ip':
9752
          if key in nic_dict:
9753
            setattr(instance.nics[nic_op], key, nic_dict[key])
9754
        if nic_op in self.nic_pinst:
9755
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
9756
        for key, val in nic_dict.iteritems():
9757
          result.append(("nic.%s/%d" % (key, nic_op), val))
9758

    
9759
    # hvparams changes
9760
    if self.op.hvparams:
9761
      instance.hvparams = self.hv_inst
9762
      for key, val in self.op.hvparams.iteritems():
9763
        result.append(("hv/%s" % key, val))
9764

    
9765
    # beparams changes
9766
    if self.op.beparams:
9767
      instance.beparams = self.be_inst
9768
      for key, val in self.op.beparams.iteritems():
9769
        result.append(("be/%s" % key, val))
9770

    
9771
    # OS change
9772
    if self.op.os_name:
9773
      instance.os = self.op.os_name
9774

    
9775
    # osparams changes
9776
    if self.op.osparams:
9777
      instance.osparams = self.os_inst
9778
      for key, val in self.op.osparams.iteritems():
9779
        result.append(("os/%s" % key, val))
9780

    
9781
    self.cfg.Update(instance, feedback_fn)
9782

    
9783
    return result
9784

    
9785
  _DISK_CONVERSIONS = {
9786
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
9787
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
9788
    }
9789

    
9790

    
9791
class LUBackupQuery(NoHooksLU):
9792
  """Query the exports list
9793

9794
  """
9795
  REQ_BGL = False
9796

    
9797
  def ExpandNames(self):
9798
    self.needed_locks = {}
9799
    self.share_locks[locking.LEVEL_NODE] = 1
9800
    if not self.op.nodes:
9801
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9802
    else:
9803
      self.needed_locks[locking.LEVEL_NODE] = \
9804
        _GetWantedNodes(self, self.op.nodes)
9805

    
9806
  def Exec(self, feedback_fn):
9807
    """Compute the list of all the exported system images.
9808

9809
    @rtype: dict
9810
    @return: a dictionary with the structure node->(export-list)
9811
        where export-list is a list of the instances exported on
9812
        that node.
9813

9814
    """
9815
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
9816
    rpcresult = self.rpc.call_export_list(self.nodes)
9817
    result = {}
9818
    for node in rpcresult:
9819
      if rpcresult[node].fail_msg:
9820
        result[node] = False
9821
      else:
9822
        result[node] = rpcresult[node].payload
9823

    
9824
    return result
9825

    
9826

    
9827
class LUBackupPrepare(NoHooksLU):
9828
  """Prepares an instance for an export and returns useful information.
9829

9830
  """
9831
  REQ_BGL = False
9832

    
9833
  def ExpandNames(self):
9834
    self._ExpandAndLockInstance()
9835

    
9836
  def CheckPrereq(self):
9837
    """Check prerequisites.
9838

9839
    """
9840
    instance_name = self.op.instance_name
9841

    
9842
    self.instance = self.cfg.GetInstanceInfo(instance_name)
9843
    assert self.instance is not None, \
9844
          "Cannot retrieve locked instance %s" % self.op.instance_name
9845
    _CheckNodeOnline(self, self.instance.primary_node)
9846

    
9847
    self._cds = _GetClusterDomainSecret()
9848

    
9849
  def Exec(self, feedback_fn):
9850
    """Prepares an instance for an export.
9851

9852
    """
9853
    instance = self.instance
9854

    
9855
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
9856
      salt = utils.GenerateSecret(8)
9857

    
9858
      feedback_fn("Generating X509 certificate on %s" % instance.primary_node)
9859
      result = self.rpc.call_x509_cert_create(instance.primary_node,
9860
                                              constants.RIE_CERT_VALIDITY)
9861
      result.Raise("Can't create X509 key and certificate on %s" % result.node)
9862

    
9863
      (name, cert_pem) = result.payload
9864

    
9865
      cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
9866
                                             cert_pem)
9867

    
9868
      return {
9869
        "handshake": masterd.instance.ComputeRemoteExportHandshake(self._cds),
9870
        "x509_key_name": (name, utils.Sha1Hmac(self._cds, name, salt=salt),
9871
                          salt),
9872
        "x509_ca": utils.SignX509Certificate(cert, self._cds, salt),
9873
        }
9874

    
9875
    return None
9876

    
9877

    
9878
class LUBackupExport(LogicalUnit):
9879
  """Export an instance to an image in the cluster.
9880

9881
  """
9882
  HPATH = "instance-export"
9883
  HTYPE = constants.HTYPE_INSTANCE
9884
  REQ_BGL = False
9885

    
9886
  def CheckArguments(self):
9887
    """Check the arguments.
9888

9889
    """
9890
    self.x509_key_name = self.op.x509_key_name
9891
    self.dest_x509_ca_pem = self.op.destination_x509_ca
9892

    
9893
    if self.op.mode == constants.EXPORT_MODE_REMOTE:
9894
      if not self.x509_key_name:
9895
        raise errors.OpPrereqError("Missing X509 key name for encryption",
9896
                                   errors.ECODE_INVAL)
9897

    
9898
      if not self.dest_x509_ca_pem:
9899
        raise errors.OpPrereqError("Missing destination X509 CA",
9900
                                   errors.ECODE_INVAL)
9901

    
9902
  def ExpandNames(self):
9903
    self._ExpandAndLockInstance()
9904

    
9905
    # Lock all nodes for local exports
9906
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9907
      # FIXME: lock only instance primary and destination node
9908
      #
9909
      # Sad but true, for now we have do lock all nodes, as we don't know where
9910
      # the previous export might be, and in this LU we search for it and
9911
      # remove it from its current node. In the future we could fix this by:
9912
      #  - making a tasklet to search (share-lock all), then create the
9913
      #    new one, then one to remove, after
9914
      #  - removing the removal operation altogether
9915
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
9916

    
9917
  def DeclareLocks(self, level):
9918
    """Last minute lock declaration."""
9919
    # All nodes are locked anyway, so nothing to do here.
9920

    
9921
  def BuildHooksEnv(self):
9922
    """Build hooks env.
9923

9924
    This will run on the master, primary node and target node.
9925

9926
    """
9927
    env = {
9928
      "EXPORT_MODE": self.op.mode,
9929
      "EXPORT_NODE": self.op.target_node,
9930
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
9931
      "SHUTDOWN_TIMEOUT": self.op.shutdown_timeout,
9932
      # TODO: Generic function for boolean env variables
9933
      "REMOVE_INSTANCE": str(bool(self.op.remove_instance)),
9934
      }
9935

    
9936
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
9937

    
9938
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node]
9939

    
9940
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9941
      nl.append(self.op.target_node)
9942

    
9943
    return env, nl, nl
9944

    
9945
  def CheckPrereq(self):
9946
    """Check prerequisites.
9947

9948
    This checks that the instance and node names are valid.
9949

9950
    """
9951
    instance_name = self.op.instance_name
9952

    
9953
    self.instance = self.cfg.GetInstanceInfo(instance_name)
9954
    assert self.instance is not None, \
9955
          "Cannot retrieve locked instance %s" % self.op.instance_name
9956
    _CheckNodeOnline(self, self.instance.primary_node)
9957

    
9958
    if (self.op.remove_instance and self.instance.admin_up and
9959
        not self.op.shutdown):
9960
      raise errors.OpPrereqError("Can not remove instance without shutting it"
9961
                                 " down before")
9962

    
9963
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
9964
      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
9965
      self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
9966
      assert self.dst_node is not None
9967

    
9968
      _CheckNodeOnline(self, self.dst_node.name)
9969
      _CheckNodeNotDrained(self, self.dst_node.name)
9970

    
9971
      self._cds = None
9972
      self.dest_disk_info = None
9973
      self.dest_x509_ca = None
9974

    
9975
    elif self.op.mode == constants.EXPORT_MODE_REMOTE:
9976
      self.dst_node = None
9977

    
9978
      if len(self.op.target_node) != len(self.instance.disks):
9979
        raise errors.OpPrereqError(("Received destination information for %s"
9980
                                    " disks, but instance %s has %s disks") %
9981
                                   (len(self.op.target_node), instance_name,
9982
                                    len(self.instance.disks)),
9983
                                   errors.ECODE_INVAL)
9984

    
9985
      cds = _GetClusterDomainSecret()
9986

    
9987
      # Check X509 key name
9988
      try:
9989
        (key_name, hmac_digest, hmac_salt) = self.x509_key_name
9990
      except (TypeError, ValueError), err:
9991
        raise errors.OpPrereqError("Invalid data for X509 key name: %s" % err)
9992

    
9993
      if not utils.VerifySha1Hmac(cds, key_name, hmac_digest, salt=hmac_salt):
9994
        raise errors.OpPrereqError("HMAC for X509 key name is wrong",
9995
                                   errors.ECODE_INVAL)
9996

    
9997
      # Load and verify CA
9998
      try:
9999
        (cert, _) = utils.LoadSignedX509Certificate(self.dest_x509_ca_pem, cds)
10000
      except OpenSSL.crypto.Error, err:
10001
        raise errors.OpPrereqError("Unable to load destination X509 CA (%s)" %
10002
                                   (err, ), errors.ECODE_INVAL)
10003

    
10004
      (errcode, msg) = utils.VerifyX509Certificate(cert, None, None)
10005
      if errcode is not None:
10006
        raise errors.OpPrereqError("Invalid destination X509 CA (%s)" %
10007
                                   (msg, ), errors.ECODE_INVAL)
10008

    
10009
      self.dest_x509_ca = cert
10010

    
10011
      # Verify target information
10012
      disk_info = []
10013
      for idx, disk_data in enumerate(self.op.target_node):
10014
        try:
10015
          (host, port, magic) = \
10016
            masterd.instance.CheckRemoteExportDiskInfo(cds, idx, disk_data)
10017
        except errors.GenericError, err:
10018
          raise errors.OpPrereqError("Target info for disk %s: %s" %
10019
                                     (idx, err), errors.ECODE_INVAL)
10020

    
10021
        disk_info.append((host, port, magic))
10022

    
10023
      assert len(disk_info) == len(self.op.target_node)
10024
      self.dest_disk_info = disk_info
10025

    
10026
    else:
10027
      raise errors.ProgrammerError("Unhandled export mode %r" %
10028
                                   self.op.mode)
10029

    
10030
    # instance disk type verification
10031
    # TODO: Implement export support for file-based disks
10032
    for disk in self.instance.disks:
10033
      if disk.dev_type == constants.LD_FILE:
10034
        raise errors.OpPrereqError("Export not supported for instances with"
10035
                                   " file-based disks", errors.ECODE_INVAL)
10036

    
10037
  def _CleanupExports(self, feedback_fn):
10038
    """Removes exports of current instance from all other nodes.
10039

10040
    If an instance in a cluster with nodes A..D was exported to node C, its
10041
    exports will be removed from the nodes A, B and D.
10042

10043
    """
10044
    assert self.op.mode != constants.EXPORT_MODE_REMOTE
10045

    
10046
    nodelist = self.cfg.GetNodeList()
10047
    nodelist.remove(self.dst_node.name)
10048

    
10049
    # on one-node clusters nodelist will be empty after the removal
10050
    # if we proceed the backup would be removed because OpBackupQuery
10051
    # substitutes an empty list with the full cluster node list.
10052
    iname = self.instance.name
10053
    if nodelist:
10054
      feedback_fn("Removing old exports for instance %s" % iname)
10055
      exportlist = self.rpc.call_export_list(nodelist)
10056
      for node in exportlist:
10057
        if exportlist[node].fail_msg:
10058
          continue
10059
        if iname in exportlist[node].payload:
10060
          msg = self.rpc.call_export_remove(node, iname).fail_msg
10061
          if msg:
10062
            self.LogWarning("Could not remove older export for instance %s"
10063
                            " on node %s: %s", iname, node, msg)
10064

    
10065
  def Exec(self, feedback_fn):
10066
    """Export an instance to an image in the cluster.
10067

10068
    """
10069
    assert self.op.mode in constants.EXPORT_MODES
10070

    
10071
    instance = self.instance
10072
    src_node = instance.primary_node
10073

    
10074
    if self.op.shutdown:
10075
      # shutdown the instance, but not the disks
10076
      feedback_fn("Shutting down instance %s" % instance.name)
10077
      result = self.rpc.call_instance_shutdown(src_node, instance,
10078
                                               self.op.shutdown_timeout)
10079
      # TODO: Maybe ignore failures if ignore_remove_failures is set
10080
      result.Raise("Could not shutdown instance %s on"
10081
                   " node %s" % (instance.name, src_node))
10082

    
10083
    # set the disks ID correctly since call_instance_start needs the
10084
    # correct drbd minor to create the symlinks
10085
    for disk in instance.disks:
10086
      self.cfg.SetDiskID(disk, src_node)
10087

    
10088
    activate_disks = (not instance.admin_up)
10089

    
10090
    if activate_disks:
10091
      # Activate the instance disks if we'exporting a stopped instance
10092
      feedback_fn("Activating disks for %s" % instance.name)
10093
      _StartInstanceDisks(self, instance, None)
10094

    
10095
    try:
10096
      helper = masterd.instance.ExportInstanceHelper(self, feedback_fn,
10097
                                                     instance)
10098

    
10099
      helper.CreateSnapshots()
10100
      try:
10101
        if (self.op.shutdown and instance.admin_up and
10102
            not self.op.remove_instance):
10103
          assert not activate_disks
10104
          feedback_fn("Starting instance %s" % instance.name)
10105
          result = self.rpc.call_instance_start(src_node, instance, None, None)
10106
          msg = result.fail_msg
10107
          if msg:
10108
            feedback_fn("Failed to start instance: %s" % msg)
10109
            _ShutdownInstanceDisks(self, instance)
10110
            raise errors.OpExecError("Could not start instance: %s" % msg)
10111

    
10112
        if self.op.mode == constants.EXPORT_MODE_LOCAL:
10113
          (fin_resu, dresults) = helper.LocalExport(self.dst_node)
10114
        elif self.op.mode == constants.EXPORT_MODE_REMOTE:
10115
          connect_timeout = constants.RIE_CONNECT_TIMEOUT
10116
          timeouts = masterd.instance.ImportExportTimeouts(connect_timeout)
10117

    
10118
          (key_name, _, _) = self.x509_key_name
10119

    
10120
          dest_ca_pem = \
10121
            OpenSSL.crypto.dump_certificate(OpenSSL.crypto.FILETYPE_PEM,
10122
                                            self.dest_x509_ca)
10123

    
10124
          (fin_resu, dresults) = helper.RemoteExport(self.dest_disk_info,
10125
                                                     key_name, dest_ca_pem,
10126
                                                     timeouts)
10127
      finally:
10128
        helper.Cleanup()
10129

    
10130
      # Check for backwards compatibility
10131
      assert len(dresults) == len(instance.disks)
10132
      assert compat.all(isinstance(i, bool) for i in dresults), \
10133
             "Not all results are boolean: %r" % dresults
10134

    
10135
    finally:
10136
      if activate_disks:
10137
        feedback_fn("Deactivating disks for %s" % instance.name)
10138
        _ShutdownInstanceDisks(self, instance)
10139

    
10140
    if not (compat.all(dresults) and fin_resu):
10141
      failures = []
10142
      if not fin_resu:
10143
        failures.append("export finalization")
10144
      if not compat.all(dresults):
10145
        fdsk = utils.CommaJoin(idx for (idx, dsk) in enumerate(dresults)
10146
                               if not dsk)
10147
        failures.append("disk export: disk(s) %s" % fdsk)
10148

    
10149
      raise errors.OpExecError("Export failed, errors in %s" %
10150
                               utils.CommaJoin(failures))
10151

    
10152
    # At this point, the export was successful, we can cleanup/finish
10153

    
10154
    # Remove instance if requested
10155
    if self.op.remove_instance:
10156
      feedback_fn("Removing instance %s" % instance.name)
10157
      _RemoveInstance(self, feedback_fn, instance,
10158
                      self.op.ignore_remove_failures)
10159

    
10160
    if self.op.mode == constants.EXPORT_MODE_LOCAL:
10161
      self._CleanupExports(feedback_fn)
10162

    
10163
    return fin_resu, dresults
10164

    
10165

    
10166
class LUBackupRemove(NoHooksLU):
10167
  """Remove exports related to the named instance.
10168

10169
  """
10170
  REQ_BGL = False
10171

    
10172
  def ExpandNames(self):
10173
    self.needed_locks = {}
10174
    # We need all nodes to be locked in order for RemoveExport to work, but we
10175
    # don't need to lock the instance itself, as nothing will happen to it (and
10176
    # we can remove exports also for a removed instance)
10177
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
10178

    
10179
  def Exec(self, feedback_fn):
10180
    """Remove any export.
10181

10182
    """
10183
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
10184
    # If the instance was not found we'll try with the name that was passed in.
10185
    # This will only work if it was an FQDN, though.
10186
    fqdn_warn = False
10187
    if not instance_name:
10188
      fqdn_warn = True
10189
      instance_name = self.op.instance_name
10190

    
10191
    locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
10192
    exportlist = self.rpc.call_export_list(locked_nodes)
10193
    found = False
10194
    for node in exportlist:
10195
      msg = exportlist[node].fail_msg
10196
      if msg:
10197
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
10198
        continue
10199
      if instance_name in exportlist[node].payload:
10200
        found = True
10201
        result = self.rpc.call_export_remove(node, instance_name)
10202
        msg = result.fail_msg
10203
        if msg:
10204
          logging.error("Could not remove export for instance %s"
10205
                        " on node %s: %s", instance_name, node, msg)
10206

    
10207
    if fqdn_warn and not found:
10208
      feedback_fn("Export not found. If trying to remove an export belonging"
10209
                  " to a deleted instance please use its Fully Qualified"
10210
                  " Domain Name.")
10211

    
10212

    
10213
class LUGroupAdd(LogicalUnit):
10214
  """Logical unit for creating node groups.
10215

10216
  """
10217
  HPATH = "group-add"
10218
  HTYPE = constants.HTYPE_GROUP
10219
  REQ_BGL = False
10220

    
10221
  def ExpandNames(self):
10222
    # We need the new group's UUID here so that we can create and acquire the
10223
    # corresponding lock. Later, in Exec(), we'll indicate to cfg.AddNodeGroup
10224
    # that it should not check whether the UUID exists in the configuration.
10225
    self.group_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
10226
    self.needed_locks = {}
10227
    self.add_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10228

    
10229
  def CheckPrereq(self):
10230
    """Check prerequisites.
10231

10232
    This checks that the given group name is not an existing node group
10233
    already.
10234

10235
    """
10236
    try:
10237
      existing_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10238
    except errors.OpPrereqError:
10239
      pass
10240
    else:
10241
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
10242
                                 " node group (UUID: %s)" %
10243
                                 (self.op.group_name, existing_uuid),
10244
                                 errors.ECODE_EXISTS)
10245

    
10246
    if self.op.ndparams:
10247
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10248

    
10249
  def BuildHooksEnv(self):
10250
    """Build hooks env.
10251

10252
    """
10253
    env = {
10254
      "GROUP_NAME": self.op.group_name,
10255
      }
10256
    mn = self.cfg.GetMasterNode()
10257
    return env, [mn], [mn]
10258

    
10259
  def Exec(self, feedback_fn):
10260
    """Add the node group to the cluster.
10261

10262
    """
10263
    group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
10264
                                  uuid=self.group_uuid,
10265
                                  alloc_policy=self.op.alloc_policy,
10266
                                  ndparams=self.op.ndparams)
10267

    
10268
    self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
10269
    del self.remove_locks[locking.LEVEL_NODEGROUP]
10270

    
10271

    
10272
class LUGroupAssignNodes(NoHooksLU):
10273
  """Logical unit for assigning nodes to groups.
10274

10275
  """
10276
  REQ_BGL = False
10277

    
10278
  def ExpandNames(self):
10279
    # These raise errors.OpPrereqError on their own:
10280
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10281
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
10282

    
10283
    # We want to lock all the affected nodes and groups. We have readily
10284
    # available the list of nodes, and the *destination* group. To gather the
10285
    # list of "source" groups, we need to fetch node information later on.
10286
    self.needed_locks = {
10287
      locking.LEVEL_NODEGROUP: set([self.group_uuid]),
10288
      locking.LEVEL_NODE: self.op.nodes,
10289
      }
10290

    
10291
  def DeclareLocks(self, level):
10292
    if level == locking.LEVEL_NODEGROUP:
10293
      assert len(self.needed_locks[locking.LEVEL_NODEGROUP]) == 1
10294

    
10295
      # Try to get all affected nodes' groups without having the group or node
10296
      # lock yet. Needs verification later in the code flow.
10297
      groups = self.cfg.GetNodeGroupsFromNodes(self.op.nodes)
10298

    
10299
      self.needed_locks[locking.LEVEL_NODEGROUP].update(groups)
10300

    
10301
  def CheckPrereq(self):
10302
    """Check prerequisites.
10303

10304
    """
10305
    assert self.needed_locks[locking.LEVEL_NODEGROUP]
10306
    assert (frozenset(self.acquired_locks[locking.LEVEL_NODE]) ==
10307
            frozenset(self.op.nodes))
10308

    
10309
    expected_locks = (set([self.group_uuid]) |
10310
                      self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
10311
    actual_locks = self.acquired_locks[locking.LEVEL_NODEGROUP]
10312
    if actual_locks != expected_locks:
10313
      raise errors.OpExecError("Nodes changed groups since locks were acquired,"
10314
                               " current groups are '%s', used to be '%s'" %
10315
                               (utils.CommaJoin(expected_locks),
10316
                                utils.CommaJoin(actual_locks)))
10317

    
10318
    self.node_data = self.cfg.GetAllNodesInfo()
10319
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10320
    instance_data = self.cfg.GetAllInstancesInfo()
10321

    
10322
    if self.group is None:
10323
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10324
                               (self.op.group_name, self.group_uuid))
10325

    
10326
    (new_splits, previous_splits) = \
10327
      self.CheckAssignmentForSplitInstances([(node, self.group_uuid)
10328
                                             for node in self.op.nodes],
10329
                                            self.node_data, instance_data)
10330

    
10331
    if new_splits:
10332
      fmt_new_splits = utils.CommaJoin(utils.NiceSort(new_splits))
10333

    
10334
      if not self.op.force:
10335
        raise errors.OpExecError("The following instances get split by this"
10336
                                 " change and --force was not given: %s" %
10337
                                 fmt_new_splits)
10338
      else:
10339
        self.LogWarning("This operation will split the following instances: %s",
10340
                        fmt_new_splits)
10341

    
10342
        if previous_splits:
10343
          self.LogWarning("In addition, these already-split instances continue"
10344
                          " to be split across groups: %s",
10345
                          utils.CommaJoin(utils.NiceSort(previous_splits)))
10346

    
10347
  def Exec(self, feedback_fn):
10348
    """Assign nodes to a new group.
10349

10350
    """
10351
    for node in self.op.nodes:
10352
      self.node_data[node].group = self.group_uuid
10353

    
10354
    # FIXME: Depends on side-effects of modifying the result of
10355
    # C{cfg.GetAllNodesInfo}
10356

    
10357
    self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
10358

    
10359
  @staticmethod
10360
  def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
10361
    """Check for split instances after a node assignment.
10362

10363
    This method considers a series of node assignments as an atomic operation,
10364
    and returns information about split instances after applying the set of
10365
    changes.
10366

10367
    In particular, it returns information about newly split instances, and
10368
    instances that were already split, and remain so after the change.
10369

10370
    Only instances whose disk template is listed in constants.DTS_NET_MIRROR are
10371
    considered.
10372

10373
    @type changes: list of (node_name, new_group_uuid) pairs.
10374
    @param changes: list of node assignments to consider.
10375
    @param node_data: a dict with data for all nodes
10376
    @param instance_data: a dict with all instances to consider
10377
    @rtype: a two-tuple
10378
    @return: a list of instances that were previously okay and result split as a
10379
      consequence of this change, and a list of instances that were previously
10380
      split and this change does not fix.
10381

10382
    """
10383
    changed_nodes = dict((node, group) for node, group in changes
10384
                         if node_data[node].group != group)
10385

    
10386
    all_split_instances = set()
10387
    previously_split_instances = set()
10388

    
10389
    def InstanceNodes(instance):
10390
      return [instance.primary_node] + list(instance.secondary_nodes)
10391

    
10392
    for inst in instance_data.values():
10393
      if inst.disk_template not in constants.DTS_NET_MIRROR:
10394
        continue
10395

    
10396
      instance_nodes = InstanceNodes(inst)
10397

    
10398
      if len(set(node_data[node].group for node in instance_nodes)) > 1:
10399
        previously_split_instances.add(inst.name)
10400

    
10401
      if len(set(changed_nodes.get(node, node_data[node].group)
10402
                 for node in instance_nodes)) > 1:
10403
        all_split_instances.add(inst.name)
10404

    
10405
    return (list(all_split_instances - previously_split_instances),
10406
            list(previously_split_instances & all_split_instances))
10407

    
10408

    
10409
class _GroupQuery(_QueryBase):
10410

    
10411
  FIELDS = query.GROUP_FIELDS
10412

    
10413
  def ExpandNames(self, lu):
10414
    lu.needed_locks = {}
10415

    
10416
    self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
10417
    name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
10418

    
10419
    if not self.names:
10420
      self.wanted = [name_to_uuid[name]
10421
                     for name in utils.NiceSort(name_to_uuid.keys())]
10422
    else:
10423
      # Accept names to be either names or UUIDs.
10424
      missing = []
10425
      self.wanted = []
10426
      all_uuid = frozenset(self._all_groups.keys())
10427

    
10428
      for name in self.names:
10429
        if name in all_uuid:
10430
          self.wanted.append(name)
10431
        elif name in name_to_uuid:
10432
          self.wanted.append(name_to_uuid[name])
10433
        else:
10434
          missing.append(name)
10435

    
10436
      if missing:
10437
        raise errors.OpPrereqError("Some groups do not exist: %s" %
10438
                                   utils.CommaJoin(missing),
10439
                                   errors.ECODE_NOENT)
10440

    
10441
  def DeclareLocks(self, lu, level):
10442
    pass
10443

    
10444
  def _GetQueryData(self, lu):
10445
    """Computes the list of node groups and their attributes.
10446

10447
    """
10448
    do_nodes = query.GQ_NODE in self.requested_data
10449
    do_instances = query.GQ_INST in self.requested_data
10450

    
10451
    group_to_nodes = None
10452
    group_to_instances = None
10453

    
10454
    # For GQ_NODE, we need to map group->[nodes], and group->[instances] for
10455
    # GQ_INST. The former is attainable with just GetAllNodesInfo(), but for the
10456
    # latter GetAllInstancesInfo() is not enough, for we have to go through
10457
    # instance->node. Hence, we will need to process nodes even if we only need
10458
    # instance information.
10459
    if do_nodes or do_instances:
10460
      all_nodes = lu.cfg.GetAllNodesInfo()
10461
      group_to_nodes = dict((uuid, []) for uuid in self.wanted)
10462
      node_to_group = {}
10463

    
10464
      for node in all_nodes.values():
10465
        if node.group in group_to_nodes:
10466
          group_to_nodes[node.group].append(node.name)
10467
          node_to_group[node.name] = node.group
10468

    
10469
      if do_instances:
10470
        all_instances = lu.cfg.GetAllInstancesInfo()
10471
        group_to_instances = dict((uuid, []) for uuid in self.wanted)
10472

    
10473
        for instance in all_instances.values():
10474
          node = instance.primary_node
10475
          if node in node_to_group:
10476
            group_to_instances[node_to_group[node]].append(instance.name)
10477

    
10478
        if not do_nodes:
10479
          # Do not pass on node information if it was not requested.
10480
          group_to_nodes = None
10481

    
10482
    return query.GroupQueryData([self._all_groups[uuid]
10483
                                 for uuid in self.wanted],
10484
                                group_to_nodes, group_to_instances)
10485

    
10486

    
10487
class LUGroupQuery(NoHooksLU):
10488
  """Logical unit for querying node groups.
10489

10490
  """
10491
  REQ_BGL = False
10492

    
10493
  def CheckArguments(self):
10494
    self.gq = _GroupQuery(self.op.names, self.op.output_fields, False)
10495

    
10496
  def ExpandNames(self):
10497
    self.gq.ExpandNames(self)
10498

    
10499
  def Exec(self, feedback_fn):
10500
    return self.gq.OldStyleQuery(self)
10501

    
10502

    
10503
class LUGroupSetParams(LogicalUnit):
10504
  """Modifies the parameters of a node group.
10505

10506
  """
10507
  HPATH = "group-modify"
10508
  HTYPE = constants.HTYPE_GROUP
10509
  REQ_BGL = False
10510

    
10511
  def CheckArguments(self):
10512
    all_changes = [
10513
      self.op.ndparams,
10514
      self.op.alloc_policy,
10515
      ]
10516

    
10517
    if all_changes.count(None) == len(all_changes):
10518
      raise errors.OpPrereqError("Please pass at least one modification",
10519
                                 errors.ECODE_INVAL)
10520

    
10521
  def ExpandNames(self):
10522
    # This raises errors.OpPrereqError on its own:
10523
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10524

    
10525
    self.needed_locks = {
10526
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10527
      }
10528

    
10529
  def CheckPrereq(self):
10530
    """Check prerequisites.
10531

10532
    """
10533
    self.group = self.cfg.GetNodeGroup(self.group_uuid)
10534

    
10535
    if self.group is None:
10536
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10537
                               (self.op.group_name, self.group_uuid))
10538

    
10539
    if self.op.ndparams:
10540
      new_ndparams = _GetUpdatedParams(self.group.ndparams, self.op.ndparams)
10541
      utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
10542
      self.new_ndparams = new_ndparams
10543

    
10544
  def BuildHooksEnv(self):
10545
    """Build hooks env.
10546

10547
    """
10548
    env = {
10549
      "GROUP_NAME": self.op.group_name,
10550
      "NEW_ALLOC_POLICY": self.op.alloc_policy,
10551
      }
10552
    mn = self.cfg.GetMasterNode()
10553
    return env, [mn], [mn]
10554

    
10555
  def Exec(self, feedback_fn):
10556
    """Modifies the node group.
10557

10558
    """
10559
    result = []
10560

    
10561
    if self.op.ndparams:
10562
      self.group.ndparams = self.new_ndparams
10563
      result.append(("ndparams", str(self.group.ndparams)))
10564

    
10565
    if self.op.alloc_policy:
10566
      self.group.alloc_policy = self.op.alloc_policy
10567

    
10568
    self.cfg.Update(self.group, feedback_fn)
10569
    return result
10570

    
10571

    
10572

    
10573
class LUGroupRemove(LogicalUnit):
10574
  HPATH = "group-remove"
10575
  HTYPE = constants.HTYPE_GROUP
10576
  REQ_BGL = False
10577

    
10578
  def ExpandNames(self):
10579
    # This will raises errors.OpPrereqError on its own:
10580
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
10581
    self.needed_locks = {
10582
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10583
      }
10584

    
10585
  def CheckPrereq(self):
10586
    """Check prerequisites.
10587

10588
    This checks that the given group name exists as a node group, that is
10589
    empty (i.e., contains no nodes), and that is not the last group of the
10590
    cluster.
10591

10592
    """
10593
    # Verify that the group is empty.
10594
    group_nodes = [node.name
10595
                   for node in self.cfg.GetAllNodesInfo().values()
10596
                   if node.group == self.group_uuid]
10597

    
10598
    if group_nodes:
10599
      raise errors.OpPrereqError("Group '%s' not empty, has the following"
10600
                                 " nodes: %s" %
10601
                                 (self.op.group_name,
10602
                                  utils.CommaJoin(utils.NiceSort(group_nodes))),
10603
                                 errors.ECODE_STATE)
10604

    
10605
    # Verify the cluster would not be left group-less.
10606
    if len(self.cfg.GetNodeGroupList()) == 1:
10607
      raise errors.OpPrereqError("Group '%s' is the only group,"
10608
                                 " cannot be removed" %
10609
                                 self.op.group_name,
10610
                                 errors.ECODE_STATE)
10611

    
10612
  def BuildHooksEnv(self):
10613
    """Build hooks env.
10614

10615
    """
10616
    env = {
10617
      "GROUP_NAME": self.op.group_name,
10618
      }
10619
    mn = self.cfg.GetMasterNode()
10620
    return env, [mn], [mn]
10621

    
10622
  def Exec(self, feedback_fn):
10623
    """Remove the node group.
10624

10625
    """
10626
    try:
10627
      self.cfg.RemoveNodeGroup(self.group_uuid)
10628
    except errors.ConfigurationError:
10629
      raise errors.OpExecError("Group '%s' with UUID %s disappeared" %
10630
                               (self.op.group_name, self.group_uuid))
10631

    
10632
    self.remove_locks[locking.LEVEL_NODEGROUP] = self.group_uuid
10633

    
10634

    
10635
class LUGroupRename(LogicalUnit):
10636
  HPATH = "group-rename"
10637
  HTYPE = constants.HTYPE_GROUP
10638
  REQ_BGL = False
10639

    
10640
  def ExpandNames(self):
10641
    # This raises errors.OpPrereqError on its own:
10642
    self.group_uuid = self.cfg.LookupNodeGroup(self.op.old_name)
10643

    
10644
    self.needed_locks = {
10645
      locking.LEVEL_NODEGROUP: [self.group_uuid],
10646
      }
10647

    
10648
  def CheckPrereq(self):
10649
    """Check prerequisites.
10650

10651
    This checks that the given old_name exists as a node group, and that
10652
    new_name doesn't.
10653

10654
    """
10655
    try:
10656
      new_name_uuid = self.cfg.LookupNodeGroup(self.op.new_name)
10657
    except errors.OpPrereqError:
10658
      pass
10659
    else:
10660
      raise errors.OpPrereqError("Desired new name '%s' clashes with existing"
10661
                                 " node group (UUID: %s)" %
10662
                                 (self.op.new_name, new_name_uuid),
10663
                                 errors.ECODE_EXISTS)
10664

    
10665
  def BuildHooksEnv(self):
10666
    """Build hooks env.
10667

10668
    """
10669
    env = {
10670
      "OLD_NAME": self.op.old_name,
10671
      "NEW_NAME": self.op.new_name,
10672
      }
10673

    
10674
    mn = self.cfg.GetMasterNode()
10675
    all_nodes = self.cfg.GetAllNodesInfo()
10676
    run_nodes = [mn]
10677
    all_nodes.pop(mn, None)
10678

    
10679
    for node in all_nodes.values():
10680
      if node.group == self.group_uuid:
10681
        run_nodes.append(node.name)
10682

    
10683
    return env, run_nodes, run_nodes
10684

    
10685
  def Exec(self, feedback_fn):
10686
    """Rename the node group.
10687

10688
    """
10689
    group = self.cfg.GetNodeGroup(self.group_uuid)
10690

    
10691
    if group is None:
10692
      raise errors.OpExecError("Could not retrieve group '%s' (UUID: %s)" %
10693
                               (self.op.old_name, self.group_uuid))
10694

    
10695
    group.name = self.op.new_name
10696
    self.cfg.Update(group, feedback_fn)
10697

    
10698
    return self.op.new_name
10699

    
10700

    
10701
class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
10702
  """Generic tags LU.
10703

10704
  This is an abstract class which is the parent of all the other tags LUs.
10705

10706
  """
10707

    
10708
  def ExpandNames(self):
10709
    self.needed_locks = {}
10710
    if self.op.kind == constants.TAG_NODE:
10711
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
10712
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
10713
    elif self.op.kind == constants.TAG_INSTANCE:
10714
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
10715
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
10716

    
10717
    # FIXME: Acquire BGL for cluster tag operations (as of this writing it's
10718
    # not possible to acquire the BGL based on opcode parameters)
10719

    
10720
  def CheckPrereq(self):
10721
    """Check prerequisites.
10722

10723
    """
10724
    if self.op.kind == constants.TAG_CLUSTER:
10725
      self.target = self.cfg.GetClusterInfo()
10726
    elif self.op.kind == constants.TAG_NODE:
10727
      self.target = self.cfg.GetNodeInfo(self.op.name)
10728
    elif self.op.kind == constants.TAG_INSTANCE:
10729
      self.target = self.cfg.GetInstanceInfo(self.op.name)
10730
    else:
10731
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
10732
                                 str(self.op.kind), errors.ECODE_INVAL)
10733

    
10734

    
10735
class LUTagsGet(TagsLU):
10736
  """Returns the tags of a given object.
10737

10738
  """
10739
  REQ_BGL = False
10740

    
10741
  def ExpandNames(self):
10742
    TagsLU.ExpandNames(self)
10743

    
10744
    # Share locks as this is only a read operation
10745
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
10746

    
10747
  def Exec(self, feedback_fn):
10748
    """Returns the tag list.
10749

10750
    """
10751
    return list(self.target.GetTags())
10752

    
10753

    
10754
class LUTagsSearch(NoHooksLU):
10755
  """Searches the tags for a given pattern.
10756

10757
  """
10758
  REQ_BGL = False
10759

    
10760
  def ExpandNames(self):
10761
    self.needed_locks = {}
10762

    
10763
  def CheckPrereq(self):
10764
    """Check prerequisites.
10765

10766
    This checks the pattern passed for validity by compiling it.
10767

10768
    """
10769
    try:
10770
      self.re = re.compile(self.op.pattern)
10771
    except re.error, err:
10772
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
10773
                                 (self.op.pattern, err), errors.ECODE_INVAL)
10774

    
10775
  def Exec(self, feedback_fn):
10776
    """Returns the tag list.
10777

10778
    """
10779
    cfg = self.cfg
10780
    tgts = [("/cluster", cfg.GetClusterInfo())]
10781
    ilist = cfg.GetAllInstancesInfo().values()
10782
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
10783
    nlist = cfg.GetAllNodesInfo().values()
10784
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
10785
    results = []
10786
    for path, target in tgts:
10787
      for tag in target.GetTags():
10788
        if self.re.search(tag):
10789
          results.append((path, tag))
10790
    return results
10791

    
10792

    
10793
class LUTagsSet(TagsLU):
10794
  """Sets a tag on a given object.
10795

10796
  """
10797
  REQ_BGL = False
10798

    
10799
  def CheckPrereq(self):
10800
    """Check prerequisites.
10801

10802
    This checks the type and length of the tag name and value.
10803

10804
    """
10805
    TagsLU.CheckPrereq(self)
10806
    for tag in self.op.tags:
10807
      objects.TaggableObject.ValidateTag(tag)
10808

    
10809
  def Exec(self, feedback_fn):
10810
    """Sets the tag.
10811

10812
    """
10813
    try:
10814
      for tag in self.op.tags:
10815
        self.target.AddTag(tag)
10816
    except errors.TagError, err:
10817
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
10818
    self.cfg.Update(self.target, feedback_fn)
10819

    
10820

    
10821
class LUTagsDel(TagsLU):
10822
  """Delete a list of tags from a given object.
10823

10824
  """
10825
  REQ_BGL = False
10826

    
10827
  def CheckPrereq(self):
10828
    """Check prerequisites.
10829

10830
    This checks that we have the given tag.
10831

10832
    """
10833
    TagsLU.CheckPrereq(self)
10834
    for tag in self.op.tags:
10835
      objects.TaggableObject.ValidateTag(tag)
10836
    del_tags = frozenset(self.op.tags)
10837
    cur_tags = self.target.GetTags()
10838

    
10839
    diff_tags = del_tags - cur_tags
10840
    if diff_tags:
10841
      diff_names = ("'%s'" % i for i in sorted(diff_tags))
10842
      raise errors.OpPrereqError("Tag(s) %s not found" %
10843
                                 (utils.CommaJoin(diff_names), ),
10844
                                 errors.ECODE_NOENT)
10845

    
10846
  def Exec(self, feedback_fn):
10847
    """Remove the tag from the object.
10848

10849
    """
10850
    for tag in self.op.tags:
10851
      self.target.RemoveTag(tag)
10852
    self.cfg.Update(self.target, feedback_fn)
10853

    
10854

    
10855
class LUTestDelay(NoHooksLU):
10856
  """Sleep for a specified amount of time.
10857

10858
  This LU sleeps on the master and/or nodes for a specified amount of
10859
  time.
10860

10861
  """
10862
  REQ_BGL = False
10863

    
10864
  def ExpandNames(self):
10865
    """Expand names and set required locks.
10866

10867
    This expands the node list, if any.
10868

10869
    """
10870
    self.needed_locks = {}
10871
    if self.op.on_nodes:
10872
      # _GetWantedNodes can be used here, but is not always appropriate to use
10873
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
10874
      # more information.
10875
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
10876
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
10877

    
10878
  def _TestDelay(self):
10879
    """Do the actual sleep.
10880

10881
    """
10882
    if self.op.on_master:
10883
      if not utils.TestDelay(self.op.duration):
10884
        raise errors.OpExecError("Error during master delay test")
10885
    if self.op.on_nodes:
10886
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
10887
      for node, node_result in result.items():
10888
        node_result.Raise("Failure during rpc call to node %s" % node)
10889

    
10890
  def Exec(self, feedback_fn):
10891
    """Execute the test delay opcode, with the wanted repetitions.
10892

10893
    """
10894
    if self.op.repeat == 0:
10895
      self._TestDelay()
10896
    else:
10897
      top_value = self.op.repeat - 1
10898
      for i in range(self.op.repeat):
10899
        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
10900
        self._TestDelay()
10901

    
10902

    
10903
class LUTestJqueue(NoHooksLU):
10904
  """Utility LU to test some aspects of the job queue.
10905

10906
  """
10907
  REQ_BGL = False
10908

    
10909
  # Must be lower than default timeout for WaitForJobChange to see whether it
10910
  # notices changed jobs
10911
  _CLIENT_CONNECT_TIMEOUT = 20.0
10912
  _CLIENT_CONFIRM_TIMEOUT = 60.0
10913

    
10914
  @classmethod
10915
  def _NotifyUsingSocket(cls, cb, errcls):
10916
    """Opens a Unix socket and waits for another program to connect.
10917

10918
    @type cb: callable
10919
    @param cb: Callback to send socket name to client
10920
    @type errcls: class
10921
    @param errcls: Exception class to use for errors
10922

10923
    """
10924
    # Using a temporary directory as there's no easy way to create temporary
10925
    # sockets without writing a custom loop around tempfile.mktemp and
10926
    # socket.bind
10927
    tmpdir = tempfile.mkdtemp()
10928
    try:
10929
      tmpsock = utils.PathJoin(tmpdir, "sock")
10930

    
10931
      logging.debug("Creating temporary socket at %s", tmpsock)
10932
      sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
10933
      try:
10934
        sock.bind(tmpsock)
10935
        sock.listen(1)
10936

    
10937
        # Send details to client
10938
        cb(tmpsock)
10939

    
10940
        # Wait for client to connect before continuing
10941
        sock.settimeout(cls._CLIENT_CONNECT_TIMEOUT)
10942
        try:
10943
          (conn, _) = sock.accept()
10944
        except socket.error, err:
10945
          raise errcls("Client didn't connect in time (%s)" % err)
10946
      finally:
10947
        sock.close()
10948
    finally:
10949
      # Remove as soon as client is connected
10950
      shutil.rmtree(tmpdir)
10951

    
10952
    # Wait for client to close
10953
    try:
10954
      try:
10955
        # pylint: disable-msg=E1101
10956
        # Instance of '_socketobject' has no ... member
10957
        conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
10958
        conn.recv(1)
10959
      except socket.error, err:
10960
        raise errcls("Client failed to confirm notification (%s)" % err)
10961
    finally:
10962
      conn.close()
10963

    
10964
  def _SendNotification(self, test, arg, sockname):
10965
    """Sends a notification to the client.
10966

10967
    @type test: string
10968
    @param test: Test name
10969
    @param arg: Test argument (depends on test)
10970
    @type sockname: string
10971
    @param sockname: Socket path
10972

10973
    """
10974
    self.Log(constants.ELOG_JQUEUE_TEST, (sockname, test, arg))
10975

    
10976
  def _Notify(self, prereq, test, arg):
10977
    """Notifies the client of a test.
10978

10979
    @type prereq: bool
10980
    @param prereq: Whether this is a prereq-phase test
10981
    @type test: string
10982
    @param test: Test name
10983
    @param arg: Test argument (depends on test)
10984

10985
    """
10986
    if prereq:
10987
      errcls = errors.OpPrereqError
10988
    else:
10989
      errcls = errors.OpExecError
10990

    
10991
    return self._NotifyUsingSocket(compat.partial(self._SendNotification,
10992
                                                  test, arg),
10993
                                   errcls)
10994

    
10995
  def CheckArguments(self):
10996
    self.checkargs_calls = getattr(self, "checkargs_calls", 0) + 1
10997
    self.expandnames_calls = 0
10998

    
10999
  def ExpandNames(self):
11000
    checkargs_calls = getattr(self, "checkargs_calls", 0)
11001
    if checkargs_calls < 1:
11002
      raise errors.ProgrammerError("CheckArguments was not called")
11003

    
11004
    self.expandnames_calls += 1
11005

    
11006
    if self.op.notify_waitlock:
11007
      self._Notify(True, constants.JQT_EXPANDNAMES, None)
11008

    
11009
    self.LogInfo("Expanding names")
11010

    
11011
    # Get lock on master node (just to get a lock, not for a particular reason)
11012
    self.needed_locks = {
11013
      locking.LEVEL_NODE: self.cfg.GetMasterNode(),
11014
      }
11015

    
11016
  def Exec(self, feedback_fn):
11017
    if self.expandnames_calls < 1:
11018
      raise errors.ProgrammerError("ExpandNames was not called")
11019

    
11020
    if self.op.notify_exec:
11021
      self._Notify(False, constants.JQT_EXEC, None)
11022

    
11023
    self.LogInfo("Executing")
11024

    
11025
    if self.op.log_messages:
11026
      self._Notify(False, constants.JQT_STARTMSG, len(self.op.log_messages))
11027
      for idx, msg in enumerate(self.op.log_messages):
11028
        self.LogInfo("Sending log message %s", idx + 1)
11029
        feedback_fn(constants.JQT_MSGPREFIX + msg)
11030
        # Report how many test messages have been sent
11031
        self._Notify(False, constants.JQT_LOGMSG, idx + 1)
11032

    
11033
    if self.op.fail:
11034
      raise errors.OpExecError("Opcode failure was requested")
11035

    
11036
    return True
11037

    
11038

    
11039
class IAllocator(object):
11040
  """IAllocator framework.
11041

11042
  An IAllocator instance has three sets of attributes:
11043
    - cfg that is needed to query the cluster
11044
    - input data (all members of the _KEYS class attribute are required)
11045
    - four buffer attributes (in|out_data|text), that represent the
11046
      input (to the external script) in text and data structure format,
11047
      and the output from it, again in two formats
11048
    - the result variables from the script (success, info, nodes) for
11049
      easy usage
11050

11051
  """
11052
  # pylint: disable-msg=R0902
11053
  # lots of instance attributes
11054
  _ALLO_KEYS = [
11055
    "name", "mem_size", "disks", "disk_template",
11056
    "os", "tags", "nics", "vcpus", "hypervisor",
11057
    ]
11058
  _RELO_KEYS = [
11059
    "name", "relocate_from",
11060
    ]
11061
  _EVAC_KEYS = [
11062
    "evac_nodes",
11063
    ]
11064

    
11065
  def __init__(self, cfg, rpc, mode, **kwargs):
11066
    self.cfg = cfg
11067
    self.rpc = rpc
11068
    # init buffer variables
11069
    self.in_text = self.out_text = self.in_data = self.out_data = None
11070
    # init all input fields so that pylint is happy
11071
    self.mode = mode
11072
    self.mem_size = self.disks = self.disk_template = None
11073
    self.os = self.tags = self.nics = self.vcpus = None
11074
    self.hypervisor = None
11075
    self.relocate_from = None
11076
    self.name = None
11077
    self.evac_nodes = None
11078
    # computed fields
11079
    self.required_nodes = None
11080
    # init result fields
11081
    self.success = self.info = self.result = None
11082
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11083
      keyset = self._ALLO_KEYS
11084
      fn = self._AddNewInstance
11085
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11086
      keyset = self._RELO_KEYS
11087
      fn = self._AddRelocateInstance
11088
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11089
      keyset = self._EVAC_KEYS
11090
      fn = self._AddEvacuateNodes
11091
    else:
11092
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
11093
                                   " IAllocator" % self.mode)
11094
    for key in kwargs:
11095
      if key not in keyset:
11096
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
11097
                                     " IAllocator" % key)
11098
      setattr(self, key, kwargs[key])
11099

    
11100
    for key in keyset:
11101
      if key not in kwargs:
11102
        raise errors.ProgrammerError("Missing input parameter '%s' to"
11103
                                     " IAllocator" % key)
11104
    self._BuildInputData(fn)
11105

    
11106
  def _ComputeClusterData(self):
11107
    """Compute the generic allocator input data.
11108

11109
    This is the data that is independent of the actual operation.
11110

11111
    """
11112
    cfg = self.cfg
11113
    cluster_info = cfg.GetClusterInfo()
11114
    # cluster data
11115
    data = {
11116
      "version": constants.IALLOCATOR_VERSION,
11117
      "cluster_name": cfg.GetClusterName(),
11118
      "cluster_tags": list(cluster_info.GetTags()),
11119
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
11120
      # we don't have job IDs
11121
      }
11122
    ninfo = cfg.GetAllNodesInfo()
11123
    iinfo = cfg.GetAllInstancesInfo().values()
11124
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
11125

    
11126
    # node data
11127
    node_list = [n.name for n in ninfo.values() if n.vm_capable]
11128

    
11129
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
11130
      hypervisor_name = self.hypervisor
11131
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
11132
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
11133
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
11134
      hypervisor_name = cluster_info.enabled_hypervisors[0]
11135

    
11136
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
11137
                                        hypervisor_name)
11138
    node_iinfo = \
11139
      self.rpc.call_all_instances_info(node_list,
11140
                                       cluster_info.enabled_hypervisors)
11141

    
11142
    data["nodegroups"] = self._ComputeNodeGroupData(cfg)
11143

    
11144
    config_ndata = self._ComputeBasicNodeData(ninfo)
11145
    data["nodes"] = self._ComputeDynamicNodeData(ninfo, node_data, node_iinfo,
11146
                                                 i_list, config_ndata)
11147
    assert len(data["nodes"]) == len(ninfo), \
11148
        "Incomplete node data computed"
11149

    
11150
    data["instances"] = self._ComputeInstanceData(cluster_info, i_list)
11151

    
11152
    self.in_data = data
11153

    
11154
  @staticmethod
11155
  def _ComputeNodeGroupData(cfg):
11156
    """Compute node groups data.
11157

11158
    """
11159
    ng = {}
11160
    for guuid, gdata in cfg.GetAllNodeGroupsInfo().items():
11161
      ng[guuid] = {
11162
        "name": gdata.name,
11163
        "alloc_policy": gdata.alloc_policy,
11164
        }
11165
    return ng
11166

    
11167
  @staticmethod
11168
  def _ComputeBasicNodeData(node_cfg):
11169
    """Compute global node data.
11170

11171
    @rtype: dict
11172
    @returns: a dict of name: (node dict, node config)
11173

11174
    """
11175
    node_results = {}
11176
    for ninfo in node_cfg.values():
11177
      # fill in static (config-based) values
11178
      pnr = {
11179
        "tags": list(ninfo.GetTags()),
11180
        "primary_ip": ninfo.primary_ip,
11181
        "secondary_ip": ninfo.secondary_ip,
11182
        "offline": ninfo.offline,
11183
        "drained": ninfo.drained,
11184
        "master_candidate": ninfo.master_candidate,
11185
        "group": ninfo.group,
11186
        "master_capable": ninfo.master_capable,
11187
        "vm_capable": ninfo.vm_capable,
11188
        }
11189

    
11190
      node_results[ninfo.name] = pnr
11191

    
11192
    return node_results
11193

    
11194
  @staticmethod
11195
  def _ComputeDynamicNodeData(node_cfg, node_data, node_iinfo, i_list,
11196
                              node_results):
11197
    """Compute global node data.
11198

11199
    @param node_results: the basic node structures as filled from the config
11200

11201
    """
11202
    # make a copy of the current dict
11203
    node_results = dict(node_results)
11204
    for nname, nresult in node_data.items():
11205
      assert nname in node_results, "Missing basic data for node %s" % nname
11206
      ninfo = node_cfg[nname]
11207

    
11208
      if not (ninfo.offline or ninfo.drained):
11209
        nresult.Raise("Can't get data for node %s" % nname)
11210
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
11211
                                nname)
11212
        remote_info = nresult.payload
11213

    
11214
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
11215
                     'vg_size', 'vg_free', 'cpu_total']:
11216
          if attr not in remote_info:
11217
            raise errors.OpExecError("Node '%s' didn't return attribute"
11218
                                     " '%s'" % (nname, attr))
11219
          if not isinstance(remote_info[attr], int):
11220
            raise errors.OpExecError("Node '%s' returned invalid value"
11221
                                     " for '%s': %s" %
11222
                                     (nname, attr, remote_info[attr]))
11223
        # compute memory used by primary instances
11224
        i_p_mem = i_p_up_mem = 0
11225
        for iinfo, beinfo in i_list:
11226
          if iinfo.primary_node == nname:
11227
            i_p_mem += beinfo[constants.BE_MEMORY]
11228
            if iinfo.name not in node_iinfo[nname].payload:
11229
              i_used_mem = 0
11230
            else:
11231
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
11232
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
11233
            remote_info['memory_free'] -= max(0, i_mem_diff)
11234

    
11235
            if iinfo.admin_up:
11236
              i_p_up_mem += beinfo[constants.BE_MEMORY]
11237

    
11238
        # compute memory used by instances
11239
        pnr_dyn = {
11240
          "total_memory": remote_info['memory_total'],
11241
          "reserved_memory": remote_info['memory_dom0'],
11242
          "free_memory": remote_info['memory_free'],
11243
          "total_disk": remote_info['vg_size'],
11244
          "free_disk": remote_info['vg_free'],
11245
          "total_cpus": remote_info['cpu_total'],
11246
          "i_pri_memory": i_p_mem,
11247
          "i_pri_up_memory": i_p_up_mem,
11248
          }
11249
        pnr_dyn.update(node_results[nname])
11250
        node_results[nname] = pnr_dyn
11251

    
11252
    return node_results
11253

    
11254
  @staticmethod
11255
  def _ComputeInstanceData(cluster_info, i_list):
11256
    """Compute global instance data.
11257

11258
    """
11259
    instance_data = {}
11260
    for iinfo, beinfo in i_list:
11261
      nic_data = []
11262
      for nic in iinfo.nics:
11263
        filled_params = cluster_info.SimpleFillNIC(nic.nicparams)
11264
        nic_dict = {"mac": nic.mac,
11265
                    "ip": nic.ip,
11266
                    "mode": filled_params[constants.NIC_MODE],
11267
                    "link": filled_params[constants.NIC_LINK],
11268
                   }
11269
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
11270
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
11271
        nic_data.append(nic_dict)
11272
      pir = {
11273
        "tags": list(iinfo.GetTags()),
11274
        "admin_up": iinfo.admin_up,
11275
        "vcpus": beinfo[constants.BE_VCPUS],
11276
        "memory": beinfo[constants.BE_MEMORY],
11277
        "os": iinfo.os,
11278
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
11279
        "nics": nic_data,
11280
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
11281
        "disk_template": iinfo.disk_template,
11282
        "hypervisor": iinfo.hypervisor,
11283
        }
11284
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
11285
                                                 pir["disks"])
11286
      instance_data[iinfo.name] = pir
11287

    
11288
    return instance_data
11289

    
11290
  def _AddNewInstance(self):
11291
    """Add new instance data to allocator structure.
11292

11293
    This in combination with _AllocatorGetClusterData will create the
11294
    correct structure needed as input for the allocator.
11295

11296
    The checks for the completeness of the opcode must have already been
11297
    done.
11298

11299
    """
11300
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
11301

    
11302
    if self.disk_template in constants.DTS_NET_MIRROR:
11303
      self.required_nodes = 2
11304
    else:
11305
      self.required_nodes = 1
11306
    request = {
11307
      "name": self.name,
11308
      "disk_template": self.disk_template,
11309
      "tags": self.tags,
11310
      "os": self.os,
11311
      "vcpus": self.vcpus,
11312
      "memory": self.mem_size,
11313
      "disks": self.disks,
11314
      "disk_space_total": disk_space,
11315
      "nics": self.nics,
11316
      "required_nodes": self.required_nodes,
11317
      }
11318
    return request
11319

    
11320
  def _AddRelocateInstance(self):
11321
    """Add relocate instance data to allocator structure.
11322

11323
    This in combination with _IAllocatorGetClusterData will create the
11324
    correct structure needed as input for the allocator.
11325

11326
    The checks for the completeness of the opcode must have already been
11327
    done.
11328

11329
    """
11330
    instance = self.cfg.GetInstanceInfo(self.name)
11331
    if instance is None:
11332
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
11333
                                   " IAllocator" % self.name)
11334

    
11335
    if instance.disk_template not in constants.DTS_NET_MIRROR:
11336
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
11337
                                 errors.ECODE_INVAL)
11338

    
11339
    if len(instance.secondary_nodes) != 1:
11340
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
11341
                                 errors.ECODE_STATE)
11342

    
11343
    self.required_nodes = 1
11344
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
11345
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
11346

    
11347
    request = {
11348
      "name": self.name,
11349
      "disk_space_total": disk_space,
11350
      "required_nodes": self.required_nodes,
11351
      "relocate_from": self.relocate_from,
11352
      }
11353
    return request
11354

    
11355
  def _AddEvacuateNodes(self):
11356
    """Add evacuate nodes data to allocator structure.
11357

11358
    """
11359
    request = {
11360
      "evac_nodes": self.evac_nodes
11361
      }
11362
    return request
11363

    
11364
  def _BuildInputData(self, fn):
11365
    """Build input data structures.
11366

11367
    """
11368
    self._ComputeClusterData()
11369

    
11370
    request = fn()
11371
    request["type"] = self.mode
11372
    self.in_data["request"] = request
11373

    
11374
    self.in_text = serializer.Dump(self.in_data)
11375

    
11376
  def Run(self, name, validate=True, call_fn=None):
11377
    """Run an instance allocator and return the results.
11378

11379
    """
11380
    if call_fn is None:
11381
      call_fn = self.rpc.call_iallocator_runner
11382

    
11383
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
11384
    result.Raise("Failure while running the iallocator script")
11385

    
11386
    self.out_text = result.payload
11387
    if validate:
11388
      self._ValidateResult()
11389

    
11390
  def _ValidateResult(self):
11391
    """Process the allocator results.
11392

11393
    This will process and if successful save the result in
11394
    self.out_data and the other parameters.
11395

11396
    """
11397
    try:
11398
      rdict = serializer.Load(self.out_text)
11399
    except Exception, err:
11400
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
11401

    
11402
    if not isinstance(rdict, dict):
11403
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
11404

    
11405
    # TODO: remove backwards compatiblity in later versions
11406
    if "nodes" in rdict and "result" not in rdict:
11407
      rdict["result"] = rdict["nodes"]
11408
      del rdict["nodes"]
11409

    
11410
    for key in "success", "info", "result":
11411
      if key not in rdict:
11412
        raise errors.OpExecError("Can't parse iallocator results:"
11413
                                 " missing key '%s'" % key)
11414
      setattr(self, key, rdict[key])
11415

    
11416
    if not isinstance(rdict["result"], list):
11417
      raise errors.OpExecError("Can't parse iallocator results: 'result' key"
11418
                               " is not a list")
11419
    self.out_data = rdict
11420

    
11421

    
11422
class LUTestAllocator(NoHooksLU):
11423
  """Run allocator tests.
11424

11425
  This LU runs the allocator tests
11426

11427
  """
11428
  def CheckPrereq(self):
11429
    """Check prerequisites.
11430

11431
    This checks the opcode parameters depending on the director and mode test.
11432

11433
    """
11434
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11435
      for attr in ["mem_size", "disks", "disk_template",
11436
                   "os", "tags", "nics", "vcpus"]:
11437
        if not hasattr(self.op, attr):
11438
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
11439
                                     attr, errors.ECODE_INVAL)
11440
      iname = self.cfg.ExpandInstanceName(self.op.name)
11441
      if iname is not None:
11442
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
11443
                                   iname, errors.ECODE_EXISTS)
11444
      if not isinstance(self.op.nics, list):
11445
        raise errors.OpPrereqError("Invalid parameter 'nics'",
11446
                                   errors.ECODE_INVAL)
11447
      if not isinstance(self.op.disks, list):
11448
        raise errors.OpPrereqError("Invalid parameter 'disks'",
11449
                                   errors.ECODE_INVAL)
11450
      for row in self.op.disks:
11451
        if (not isinstance(row, dict) or
11452
            "size" not in row or
11453
            not isinstance(row["size"], int) or
11454
            "mode" not in row or
11455
            row["mode"] not in ['r', 'w']):
11456
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
11457
                                     " parameter", errors.ECODE_INVAL)
11458
      if self.op.hypervisor is None:
11459
        self.op.hypervisor = self.cfg.GetHypervisorType()
11460
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11461
      fname = _ExpandInstanceName(self.cfg, self.op.name)
11462
      self.op.name = fname
11463
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
11464
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11465
      if not hasattr(self.op, "evac_nodes"):
11466
        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
11467
                                   " opcode input", errors.ECODE_INVAL)
11468
    else:
11469
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
11470
                                 self.op.mode, errors.ECODE_INVAL)
11471

    
11472
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
11473
      if self.op.allocator is None:
11474
        raise errors.OpPrereqError("Missing allocator name",
11475
                                   errors.ECODE_INVAL)
11476
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
11477
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
11478
                                 self.op.direction, errors.ECODE_INVAL)
11479

    
11480
  def Exec(self, feedback_fn):
11481
    """Run the allocator test.
11482

11483
    """
11484
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
11485
      ial = IAllocator(self.cfg, self.rpc,
11486
                       mode=self.op.mode,
11487
                       name=self.op.name,
11488
                       mem_size=self.op.mem_size,
11489
                       disks=self.op.disks,
11490
                       disk_template=self.op.disk_template,
11491
                       os=self.op.os,
11492
                       tags=self.op.tags,
11493
                       nics=self.op.nics,
11494
                       vcpus=self.op.vcpus,
11495
                       hypervisor=self.op.hypervisor,
11496
                       )
11497
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
11498
      ial = IAllocator(self.cfg, self.rpc,
11499
                       mode=self.op.mode,
11500
                       name=self.op.name,
11501
                       relocate_from=list(self.relocate_from),
11502
                       )
11503
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
11504
      ial = IAllocator(self.cfg, self.rpc,
11505
                       mode=self.op.mode,
11506
                       evac_nodes=self.op.evac_nodes)
11507
    else:
11508
      raise errors.ProgrammerError("Uncatched mode %s in"
11509
                                   " LUTestAllocator.Exec", self.op.mode)
11510

    
11511
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
11512
      result = ial.in_text
11513
    else:
11514
      ial.Run(self.op.allocator, validate=False)
11515
      result = ial.out_text
11516
    return result
11517

    
11518

    
11519
#: Query type implementations
11520
_QUERY_IMPL = {
11521
  constants.QR_INSTANCE: _InstanceQuery,
11522
  constants.QR_NODE: _NodeQuery,
11523
  constants.QR_GROUP: _GroupQuery,
11524
  }
11525

    
11526

    
11527
def _GetQueryImplementation(name):
11528
  """Returns the implemtnation for a query type.
11529

11530
  @param name: Query type, must be one of L{constants.QR_OP_QUERY}
11531

11532
  """
11533
  try:
11534
    return _QUERY_IMPL[name]
11535
  except KeyError:
11536
    raise errors.OpPrereqError("Unknown query resource '%s'" % name,
11537
                               errors.ECODE_INVAL)