Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 7a735d6a

History | View | Annotate | Download (233.2 kB)

1
#
2
#
3

    
4
# Copyright (C) 2006, 2007, 2008 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=W0613,W0201
25

    
26
import os
27
import os.path
28
import sha
29
import time
30
import tempfile
31
import re
32
import platform
33
import logging
34
import copy
35
import random
36

    
37
from ganeti import ssh
38
from ganeti import utils
39
from ganeti import errors
40
from ganeti import hypervisor
41
from ganeti import locking
42
from ganeti import constants
43
from ganeti import objects
44
from ganeti import opcodes
45
from ganeti import serializer
46
from ganeti import ssconf
47

    
48

    
49
class LogicalUnit(object):
50
  """Logical Unit base class.
51

52
  Subclasses must follow these rules:
53
    - implement ExpandNames
54
    - implement CheckPrereq
55
    - implement Exec
56
    - implement BuildHooksEnv
57
    - redefine HPATH and HTYPE
58
    - optionally redefine their run requirements:
59
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
60

61
  Note that all commands require root permissions.
62

63
  """
64
  HPATH = None
65
  HTYPE = None
66
  _OP_REQP = []
67
  REQ_BGL = True
68

    
69
  def __init__(self, processor, op, context, rpc):
70
    """Constructor for LogicalUnit.
71

72
    This needs to be overriden in derived classes in order to check op
73
    validity.
74

75
    """
76
    self.proc = processor
77
    self.op = op
78
    self.cfg = context.cfg
79
    self.context = context
80
    self.rpc = rpc
81
    # Dicts used to declare locking needs to mcpu
82
    self.needed_locks = None
83
    self.acquired_locks = {}
84
    self.share_locks = dict(((i, 0) for i in locking.LEVELS))
85
    self.add_locks = {}
86
    self.remove_locks = {}
87
    # Used to force good behavior when calling helper functions
88
    self.recalculate_locks = {}
89
    self.__ssh = None
90
    # logging
91
    self.LogWarning = processor.LogWarning
92
    self.LogInfo = processor.LogInfo
93

    
94
    for attr_name in self._OP_REQP:
95
      attr_val = getattr(op, attr_name, None)
96
      if attr_val is None:
97
        raise errors.OpPrereqError("Required parameter '%s' missing" %
98
                                   attr_name)
99
    self.CheckArguments()
100

    
101
  def __GetSSH(self):
102
    """Returns the SshRunner object
103

104
    """
105
    if not self.__ssh:
106
      self.__ssh = ssh.SshRunner(self.cfg.GetClusterName())
107
    return self.__ssh
108

    
109
  ssh = property(fget=__GetSSH)
110

    
111
  def CheckArguments(self):
112
    """Check syntactic validity for the opcode arguments.
113

114
    This method is for doing a simple syntactic check and ensure
115
    validity of opcode parameters, without any cluster-related
116
    checks. While the same can be accomplished in ExpandNames and/or
117
    CheckPrereq, doing these separate is better because:
118

119
      - ExpandNames is left as as purely a lock-related function
120
      - CheckPrereq is run after we have aquired locks (and possible
121
        waited for them)
122

123
    The function is allowed to change the self.op attribute so that
124
    later methods can no longer worry about missing parameters.
125

126
    """
127
    pass
128

    
129
  def ExpandNames(self):
130
    """Expand names for this LU.
131

132
    This method is called before starting to execute the opcode, and it should
133
    update all the parameters of the opcode to their canonical form (e.g. a
134
    short node name must be fully expanded after this method has successfully
135
    completed). This way locking, hooks, logging, ecc. can work correctly.
136

137
    LUs which implement this method must also populate the self.needed_locks
138
    member, as a dict with lock levels as keys, and a list of needed lock names
139
    as values. Rules:
140

141
      - use an empty dict if you don't need any lock
142
      - if you don't need any lock at a particular level omit that level
143
      - don't put anything for the BGL level
144
      - if you want all locks at a level use locking.ALL_SET as a value
145

146
    If you need to share locks (rather than acquire them exclusively) at one
147
    level you can modify self.share_locks, setting a true value (usually 1) for
148
    that level. By default locks are not shared.
149

150
    Examples::
151

152
      # Acquire all nodes and one instance
153
      self.needed_locks = {
154
        locking.LEVEL_NODE: locking.ALL_SET,
155
        locking.LEVEL_INSTANCE: ['instance1.example.tld'],
156
      }
157
      # Acquire just two nodes
158
      self.needed_locks = {
159
        locking.LEVEL_NODE: ['node1.example.tld', 'node2.example.tld'],
160
      }
161
      # Acquire no locks
162
      self.needed_locks = {} # No, you can't leave it to the default value None
163

164
    """
165
    # The implementation of this method is mandatory only if the new LU is
166
    # concurrent, so that old LUs don't need to be changed all at the same
167
    # time.
168
    if self.REQ_BGL:
169
      self.needed_locks = {} # Exclusive LUs don't need locks.
170
    else:
171
      raise NotImplementedError
172

    
173
  def DeclareLocks(self, level):
174
    """Declare LU locking needs for a level
175

176
    While most LUs can just declare their locking needs at ExpandNames time,
177
    sometimes there's the need to calculate some locks after having acquired
178
    the ones before. This function is called just before acquiring locks at a
179
    particular level, but after acquiring the ones at lower levels, and permits
180
    such calculations. It can be used to modify self.needed_locks, and by
181
    default it does nothing.
182

183
    This function is only called if you have something already set in
184
    self.needed_locks for the level.
185

186
    @param level: Locking level which is going to be locked
187
    @type level: member of ganeti.locking.LEVELS
188

189
    """
190

    
191
  def CheckPrereq(self):
192
    """Check prerequisites for this LU.
193

194
    This method should check that the prerequisites for the execution
195
    of this LU are fulfilled. It can do internode communication, but
196
    it should be idempotent - no cluster or system changes are
197
    allowed.
198

199
    The method should raise errors.OpPrereqError in case something is
200
    not fulfilled. Its return value is ignored.
201

202
    This method should also update all the parameters of the opcode to
203
    their canonical form if it hasn't been done by ExpandNames before.
204

205
    """
206
    raise NotImplementedError
207

    
208
  def Exec(self, feedback_fn):
209
    """Execute the LU.
210

211
    This method should implement the actual work. It should raise
212
    errors.OpExecError for failures that are somewhat dealt with in
213
    code, or expected.
214

215
    """
216
    raise NotImplementedError
217

    
218
  def BuildHooksEnv(self):
219
    """Build hooks environment for this LU.
220

221
    This method should return a three-node tuple consisting of: a dict
222
    containing the environment that will be used for running the
223
    specific hook for this LU, a list of node names on which the hook
224
    should run before the execution, and a list of node names on which
225
    the hook should run after the execution.
226

227
    The keys of the dict must not have 'GANETI_' prefixed as this will
228
    be handled in the hooks runner. Also note additional keys will be
229
    added by the hooks runner. If the LU doesn't define any
230
    environment, an empty dict (and not None) should be returned.
231

232
    No nodes should be returned as an empty list (and not None).
233

234
    Note that if the HPATH for a LU class is None, this function will
235
    not be called.
236

237
    """
238
    raise NotImplementedError
239

    
240
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
241
    """Notify the LU about the results of its hooks.
242

243
    This method is called every time a hooks phase is executed, and notifies
244
    the Logical Unit about the hooks' result. The LU can then use it to alter
245
    its result based on the hooks.  By default the method does nothing and the
246
    previous result is passed back unchanged but any LU can define it if it
247
    wants to use the local cluster hook-scripts somehow.
248

249
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
250
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
251
    @param hook_results: the results of the multi-node hooks rpc call
252
    @param feedback_fn: function used send feedback back to the caller
253
    @param lu_result: the previous Exec result this LU had, or None
254
        in the PRE phase
255
    @return: the new Exec result, based on the previous result
256
        and hook results
257

258
    """
259
    return lu_result
260

    
261
  def _ExpandAndLockInstance(self):
262
    """Helper function to expand and lock an instance.
263

264
    Many LUs that work on an instance take its name in self.op.instance_name
265
    and need to expand it and then declare the expanded name for locking. This
266
    function does it, and then updates self.op.instance_name to the expanded
267
    name. It also initializes needed_locks as a dict, if this hasn't been done
268
    before.
269

270
    """
271
    if self.needed_locks is None:
272
      self.needed_locks = {}
273
    else:
274
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
275
        "_ExpandAndLockInstance called with instance-level locks set"
276
    expanded_name = self.cfg.ExpandInstanceName(self.op.instance_name)
277
    if expanded_name is None:
278
      raise errors.OpPrereqError("Instance '%s' not known" %
279
                                  self.op.instance_name)
280
    self.needed_locks[locking.LEVEL_INSTANCE] = expanded_name
281
    self.op.instance_name = expanded_name
282

    
283
  def _LockInstancesNodes(self, primary_only=False):
284
    """Helper function to declare instances' nodes for locking.
285

286
    This function should be called after locking one or more instances to lock
287
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
288
    with all primary or secondary nodes for instances already locked and
289
    present in self.needed_locks[locking.LEVEL_INSTANCE].
290

291
    It should be called from DeclareLocks, and for safety only works if
292
    self.recalculate_locks[locking.LEVEL_NODE] is set.
293

294
    In the future it may grow parameters to just lock some instance's nodes, or
295
    to just lock primaries or secondary nodes, if needed.
296

297
    If should be called in DeclareLocks in a way similar to::
298

299
      if level == locking.LEVEL_NODE:
300
        self._LockInstancesNodes()
301

302
    @type primary_only: boolean
303
    @param primary_only: only lock primary nodes of locked instances
304

305
    """
306
    assert locking.LEVEL_NODE in self.recalculate_locks, \
307
      "_LockInstancesNodes helper function called with no nodes to recalculate"
308

    
309
    # TODO: check if we're really been called with the instance locks held
310

    
311
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
312
    # future we might want to have different behaviors depending on the value
313
    # of self.recalculate_locks[locking.LEVEL_NODE]
314
    wanted_nodes = []
315
    for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
316
      instance = self.context.cfg.GetInstanceInfo(instance_name)
317
      wanted_nodes.append(instance.primary_node)
318
      if not primary_only:
319
        wanted_nodes.extend(instance.secondary_nodes)
320

    
321
    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
322
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
323
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
324
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
325

    
326
    del self.recalculate_locks[locking.LEVEL_NODE]
327

    
328

    
329
class NoHooksLU(LogicalUnit):
330
  """Simple LU which runs no hooks.
331

332
  This LU is intended as a parent for other LogicalUnits which will
333
  run no hooks, in order to reduce duplicate code.
334

335
  """
336
  HPATH = None
337
  HTYPE = None
338

    
339

    
340
def _GetWantedNodes(lu, nodes):
341
  """Returns list of checked and expanded node names.
342

343
  @type lu: L{LogicalUnit}
344
  @param lu: the logical unit on whose behalf we execute
345
  @type nodes: list
346
  @param nodes: list of node names or None for all nodes
347
  @rtype: list
348
  @return: the list of nodes, sorted
349
  @raise errors.OpProgrammerError: if the nodes parameter is wrong type
350

351
  """
352
  if not isinstance(nodes, list):
353
    raise errors.OpPrereqError("Invalid argument type 'nodes'")
354

    
355
  if not nodes:
356
    raise errors.ProgrammerError("_GetWantedNodes should only be called with a"
357
      " non-empty list of nodes whose name is to be expanded.")
358

    
359
  wanted = []
360
  for name in nodes:
361
    node = lu.cfg.ExpandNodeName(name)
362
    if node is None:
363
      raise errors.OpPrereqError("No such node name '%s'" % name)
364
    wanted.append(node)
365

    
366
  return utils.NiceSort(wanted)
367

    
368

    
369
def _GetWantedInstances(lu, instances):
370
  """Returns list of checked and expanded instance names.
371

372
  @type lu: L{LogicalUnit}
373
  @param lu: the logical unit on whose behalf we execute
374
  @type instances: list
375
  @param instances: list of instance names or None for all instances
376
  @rtype: list
377
  @return: the list of instances, sorted
378
  @raise errors.OpPrereqError: if the instances parameter is wrong type
379
  @raise errors.OpPrereqError: if any of the passed instances is not found
380

381
  """
382
  if not isinstance(instances, list):
383
    raise errors.OpPrereqError("Invalid argument type 'instances'")
384

    
385
  if instances:
386
    wanted = []
387

    
388
    for name in instances:
389
      instance = lu.cfg.ExpandInstanceName(name)
390
      if instance is None:
391
        raise errors.OpPrereqError("No such instance name '%s'" % name)
392
      wanted.append(instance)
393

    
394
  else:
395
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
396
  return wanted
397

    
398

    
399
def _CheckOutputFields(static, dynamic, selected):
400
  """Checks whether all selected fields are valid.
401

402
  @type static: L{utils.FieldSet}
403
  @param static: static fields set
404
  @type dynamic: L{utils.FieldSet}
405
  @param dynamic: dynamic fields set
406

407
  """
408
  f = utils.FieldSet()
409
  f.Extend(static)
410
  f.Extend(dynamic)
411

    
412
  delta = f.NonMatching(selected)
413
  if delta:
414
    raise errors.OpPrereqError("Unknown output fields selected: %s"
415
                               % ",".join(delta))
416

    
417

    
418
def _CheckBooleanOpField(op, name):
419
  """Validates boolean opcode parameters.
420

421
  This will ensure that an opcode parameter is either a boolean value,
422
  or None (but that it always exists).
423

424
  """
425
  val = getattr(op, name, None)
426
  if not (val is None or isinstance(val, bool)):
427
    raise errors.OpPrereqError("Invalid boolean parameter '%s' (%s)" %
428
                               (name, str(val)))
429
  setattr(op, name, val)
430

    
431

    
432
def _CheckNodeOnline(lu, node):
433
  """Ensure that a given node is online.
434

435
  @param lu: the LU on behalf of which we make the check
436
  @param node: the node to check
437
  @raise errors.OpPrereqError: if the nodes is offline
438

439
  """
440
  if lu.cfg.GetNodeInfo(node).offline:
441
    raise errors.OpPrereqError("Can't use offline node %s" % node)
442

    
443

    
444
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
445
                          memory, vcpus, nics):
446
  """Builds instance related env variables for hooks
447

448
  This builds the hook environment from individual variables.
449

450
  @type name: string
451
  @param name: the name of the instance
452
  @type primary_node: string
453
  @param primary_node: the name of the instance's primary node
454
  @type secondary_nodes: list
455
  @param secondary_nodes: list of secondary nodes as strings
456
  @type os_type: string
457
  @param os_type: the name of the instance's OS
458
  @type status: boolean
459
  @param status: the should_run status of the instance
460
  @type memory: string
461
  @param memory: the memory size of the instance
462
  @type vcpus: string
463
  @param vcpus: the count of VCPUs the instance has
464
  @type nics: list
465
  @param nics: list of tuples (ip, bridge, mac) representing
466
      the NICs the instance  has
467
  @rtype: dict
468
  @return: the hook environment for this instance
469

470
  """
471
  if status:
472
    str_status = "up"
473
  else:
474
    str_status = "down"
475
  env = {
476
    "OP_TARGET": name,
477
    "INSTANCE_NAME": name,
478
    "INSTANCE_PRIMARY": primary_node,
479
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
480
    "INSTANCE_OS_TYPE": os_type,
481
    "INSTANCE_STATUS": str_status,
482
    "INSTANCE_MEMORY": memory,
483
    "INSTANCE_VCPUS": vcpus,
484
  }
485

    
486
  if nics:
487
    nic_count = len(nics)
488
    for idx, (ip, bridge, mac) in enumerate(nics):
489
      if ip is None:
490
        ip = ""
491
      env["INSTANCE_NIC%d_IP" % idx] = ip
492
      env["INSTANCE_NIC%d_BRIDGE" % idx] = bridge
493
      env["INSTANCE_NIC%d_HWADDR" % idx] = mac
494
  else:
495
    nic_count = 0
496

    
497
  env["INSTANCE_NIC_COUNT"] = nic_count
498

    
499
  return env
500

    
501

    
502
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
503
  """Builds instance related env variables for hooks from an object.
504

505
  @type lu: L{LogicalUnit}
506
  @param lu: the logical unit on whose behalf we execute
507
  @type instance: L{objects.Instance}
508
  @param instance: the instance for which we should build the
509
      environment
510
  @type override: dict
511
  @param override: dictionary with key/values that will override
512
      our values
513
  @rtype: dict
514
  @return: the hook environment dictionary
515

516
  """
517
  bep = lu.cfg.GetClusterInfo().FillBE(instance)
518
  args = {
519
    'name': instance.name,
520
    'primary_node': instance.primary_node,
521
    'secondary_nodes': instance.secondary_nodes,
522
    'os_type': instance.os,
523
    'status': instance.admin_up,
524
    'memory': bep[constants.BE_MEMORY],
525
    'vcpus': bep[constants.BE_VCPUS],
526
    'nics': [(nic.ip, nic.bridge, nic.mac) for nic in instance.nics],
527
  }
528
  if override:
529
    args.update(override)
530
  return _BuildInstanceHookEnv(**args)
531

    
532

    
533
def _AdjustCandidatePool(lu):
534
  """Adjust the candidate pool after node operations.
535

536
  """
537
  mod_list = lu.cfg.MaintainCandidatePool()
538
  if mod_list:
539
    lu.LogInfo("Promoted nodes to master candidate role: %s",
540
               ", ".join(node.name for node in mod_list))
541
    for name in mod_list:
542
      lu.context.ReaddNode(name)
543
  mc_now, mc_max = lu.cfg.GetMasterCandidateStats()
544
  if mc_now > mc_max:
545
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
546
               (mc_now, mc_max))
547

    
548

    
549
def _CheckInstanceBridgesExist(lu, instance):
550
  """Check that the brigdes needed by an instance exist.
551

552
  """
553
  # check bridges existance
554
  brlist = [nic.bridge for nic in instance.nics]
555
  result = lu.rpc.call_bridges_exist(instance.primary_node, brlist)
556
  result.Raise()
557
  if not result.data:
558
    raise errors.OpPrereqError("One or more target bridges %s does not"
559
                               " exist on destination node '%s'" %
560
                               (brlist, instance.primary_node))
561

    
562

    
563
class LUDestroyCluster(NoHooksLU):
564
  """Logical unit for destroying the cluster.
565

566
  """
567
  _OP_REQP = []
568

    
569
  def CheckPrereq(self):
570
    """Check prerequisites.
571

572
    This checks whether the cluster is empty.
573

574
    Any errors are signalled by raising errors.OpPrereqError.
575

576
    """
577
    master = self.cfg.GetMasterNode()
578

    
579
    nodelist = self.cfg.GetNodeList()
580
    if len(nodelist) != 1 or nodelist[0] != master:
581
      raise errors.OpPrereqError("There are still %d node(s) in"
582
                                 " this cluster." % (len(nodelist) - 1))
583
    instancelist = self.cfg.GetInstanceList()
584
    if instancelist:
585
      raise errors.OpPrereqError("There are still %d instance(s) in"
586
                                 " this cluster." % len(instancelist))
587

    
588
  def Exec(self, feedback_fn):
589
    """Destroys the cluster.
590

591
    """
592
    master = self.cfg.GetMasterNode()
593
    result = self.rpc.call_node_stop_master(master, False)
594
    result.Raise()
595
    if not result.data:
596
      raise errors.OpExecError("Could not disable the master role")
597
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
598
    utils.CreateBackup(priv_key)
599
    utils.CreateBackup(pub_key)
600
    return master
601

    
602

    
603
class LUVerifyCluster(LogicalUnit):
604
  """Verifies the cluster status.
605

606
  """
607
  HPATH = "cluster-verify"
608
  HTYPE = constants.HTYPE_CLUSTER
609
  _OP_REQP = ["skip_checks"]
610
  REQ_BGL = False
611

    
612
  def ExpandNames(self):
613
    self.needed_locks = {
614
      locking.LEVEL_NODE: locking.ALL_SET,
615
      locking.LEVEL_INSTANCE: locking.ALL_SET,
616
    }
617
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
618

    
619
  def _VerifyNode(self, nodeinfo, file_list, local_cksum,
620
                  node_result, feedback_fn, master_files,
621
                  drbd_map):
622
    """Run multiple tests against a node.
623

624
    Test list:
625

626
      - compares ganeti version
627
      - checks vg existance and size > 20G
628
      - checks config file checksum
629
      - checks ssh to other nodes
630

631
    @type nodeinfo: L{objects.Node}
632
    @param nodeinfo: the node to check
633
    @param file_list: required list of files
634
    @param local_cksum: dictionary of local files and their checksums
635
    @param node_result: the results from the node
636
    @param feedback_fn: function used to accumulate results
637
    @param master_files: list of files that only masters should have
638
    @param drbd_map: the useddrbd minors for this node, in
639
        form of minor: (instance, must_exist) which correspond to instances
640
        and their running status
641

642
    """
643
    node = nodeinfo.name
644

    
645
    # main result, node_result should be a non-empty dict
646
    if not node_result or not isinstance(node_result, dict):
647
      feedback_fn("  - ERROR: unable to verify node %s." % (node,))
648
      return True
649

    
650
    # compares ganeti version
651
    local_version = constants.PROTOCOL_VERSION
652
    remote_version = node_result.get('version', None)
653
    if not (remote_version and isinstance(remote_version, (list, tuple)) and
654
            len(remote_version) == 2):
655
      feedback_fn("  - ERROR: connection to %s failed" % (node))
656
      return True
657

    
658
    if local_version != remote_version[0]:
659
      feedback_fn("  - ERROR: incompatible protocol versions: master %s,"
660
                  " node %s %s" % (local_version, node, remote_version[0]))
661
      return True
662

    
663
    # node seems compatible, we can actually try to look into its results
664

    
665
    bad = False
666

    
667
    # full package version
668
    if constants.RELEASE_VERSION != remote_version[1]:
669
      feedback_fn("  - WARNING: software version mismatch: master %s,"
670
                  " node %s %s" %
671
                  (constants.RELEASE_VERSION, node, remote_version[1]))
672

    
673
    # checks vg existence and size > 20G
674

    
675
    vglist = node_result.get(constants.NV_VGLIST, None)
676
    if not vglist:
677
      feedback_fn("  - ERROR: unable to check volume groups on node %s." %
678
                      (node,))
679
      bad = True
680
    else:
681
      vgstatus = utils.CheckVolumeGroupSize(vglist, self.cfg.GetVGName(),
682
                                            constants.MIN_VG_SIZE)
683
      if vgstatus:
684
        feedback_fn("  - ERROR: %s on node %s" % (vgstatus, node))
685
        bad = True
686

    
687
    # checks config file checksum
688

    
689
    remote_cksum = node_result.get(constants.NV_FILELIST, None)
690
    if not isinstance(remote_cksum, dict):
691
      bad = True
692
      feedback_fn("  - ERROR: node hasn't returned file checksum data")
693
    else:
694
      for file_name in file_list:
695
        node_is_mc = nodeinfo.master_candidate
696
        must_have_file = file_name not in master_files
697
        if file_name not in remote_cksum:
698
          if node_is_mc or must_have_file:
699
            bad = True
700
            feedback_fn("  - ERROR: file '%s' missing" % file_name)
701
        elif remote_cksum[file_name] != local_cksum[file_name]:
702
          if node_is_mc or must_have_file:
703
            bad = True
704
            feedback_fn("  - ERROR: file '%s' has wrong checksum" % file_name)
705
          else:
706
            # not candidate and this is not a must-have file
707
            bad = True
708
            feedback_fn("  - ERROR: non master-candidate has old/wrong file"
709
                        " '%s'" % file_name)
710
        else:
711
          # all good, except non-master/non-must have combination
712
          if not node_is_mc and not must_have_file:
713
            feedback_fn("  - ERROR: file '%s' should not exist on non master"
714
                        " candidates" % file_name)
715

    
716
    # checks ssh to any
717

    
718
    if constants.NV_NODELIST not in node_result:
719
      bad = True
720
      feedback_fn("  - ERROR: node hasn't returned node ssh connectivity data")
721
    else:
722
      if node_result[constants.NV_NODELIST]:
723
        bad = True
724
        for node in node_result[constants.NV_NODELIST]:
725
          feedback_fn("  - ERROR: ssh communication with node '%s': %s" %
726
                          (node, node_result[constants.NV_NODELIST][node]))
727

    
728
    if constants.NV_NODENETTEST not in node_result:
729
      bad = True
730
      feedback_fn("  - ERROR: node hasn't returned node tcp connectivity data")
731
    else:
732
      if node_result[constants.NV_NODENETTEST]:
733
        bad = True
734
        nlist = utils.NiceSort(node_result[constants.NV_NODENETTEST].keys())
735
        for node in nlist:
736
          feedback_fn("  - ERROR: tcp communication with node '%s': %s" %
737
                          (node, node_result[constants.NV_NODENETTEST][node]))
738

    
739
    hyp_result = node_result.get(constants.NV_HYPERVISOR, None)
740
    if isinstance(hyp_result, dict):
741
      for hv_name, hv_result in hyp_result.iteritems():
742
        if hv_result is not None:
743
          feedback_fn("  - ERROR: hypervisor %s verify failure: '%s'" %
744
                      (hv_name, hv_result))
745

    
746
    # check used drbd list
747
    used_minors = node_result.get(constants.NV_DRBDLIST, [])
748
    for minor, (iname, must_exist) in drbd_map.items():
749
      if minor not in used_minors and must_exist:
750
        feedback_fn("  - ERROR: drbd minor %d of instance %s is not active" %
751
                    (minor, iname))
752
        bad = True
753
    for minor in used_minors:
754
      if minor not in drbd_map:
755
        feedback_fn("  - ERROR: unallocated drbd minor %d is in use" % minor)
756
        bad = True
757

    
758
    return bad
759

    
760
  def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
761
                      node_instance, feedback_fn, n_offline):
762
    """Verify an instance.
763

764
    This function checks to see if the required block devices are
765
    available on the instance's node.
766

767
    """
768
    bad = False
769

    
770
    node_current = instanceconfig.primary_node
771

    
772
    node_vol_should = {}
773
    instanceconfig.MapLVsByNode(node_vol_should)
774

    
775
    for node in node_vol_should:
776
      if node in n_offline:
777
        # ignore missing volumes on offline nodes
778
        continue
779
      for volume in node_vol_should[node]:
780
        if node not in node_vol_is or volume not in node_vol_is[node]:
781
          feedback_fn("  - ERROR: volume %s missing on node %s" %
782
                          (volume, node))
783
          bad = True
784

    
785
    if instanceconfig.admin_up:
786
      if ((node_current not in node_instance or
787
          not instance in node_instance[node_current]) and
788
          node_current not in n_offline):
789
        feedback_fn("  - ERROR: instance %s not running on node %s" %
790
                        (instance, node_current))
791
        bad = True
792

    
793
    for node in node_instance:
794
      if (not node == node_current):
795
        if instance in node_instance[node]:
796
          feedback_fn("  - ERROR: instance %s should not run on node %s" %
797
                          (instance, node))
798
          bad = True
799

    
800
    return bad
801

    
802
  def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is, feedback_fn):
803
    """Verify if there are any unknown volumes in the cluster.
804

805
    The .os, .swap and backup volumes are ignored. All other volumes are
806
    reported as unknown.
807

808
    """
809
    bad = False
810

    
811
    for node in node_vol_is:
812
      for volume in node_vol_is[node]:
813
        if node not in node_vol_should or volume not in node_vol_should[node]:
814
          feedback_fn("  - ERROR: volume %s on node %s should not exist" %
815
                      (volume, node))
816
          bad = True
817
    return bad
818

    
819
  def _VerifyOrphanInstances(self, instancelist, node_instance, feedback_fn):
820
    """Verify the list of running instances.
821

822
    This checks what instances are running but unknown to the cluster.
823

824
    """
825
    bad = False
826
    for node in node_instance:
827
      for runninginstance in node_instance[node]:
828
        if runninginstance not in instancelist:
829
          feedback_fn("  - ERROR: instance %s on node %s should not exist" %
830
                          (runninginstance, node))
831
          bad = True
832
    return bad
833

    
834
  def _VerifyNPlusOneMemory(self, node_info, instance_cfg, feedback_fn):
835
    """Verify N+1 Memory Resilience.
836

837
    Check that if one single node dies we can still start all the instances it
838
    was primary for.
839

840
    """
841
    bad = False
842

    
843
    for node, nodeinfo in node_info.iteritems():
844
      # This code checks that every node which is now listed as secondary has
845
      # enough memory to host all instances it is supposed to should a single
846
      # other node in the cluster fail.
847
      # FIXME: not ready for failover to an arbitrary node
848
      # FIXME: does not support file-backed instances
849
      # WARNING: we currently take into account down instances as well as up
850
      # ones, considering that even if they're down someone might want to start
851
      # them even in the event of a node failure.
852
      for prinode, instances in nodeinfo['sinst-by-pnode'].iteritems():
853
        needed_mem = 0
854
        for instance in instances:
855
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
856
          if bep[constants.BE_AUTO_BALANCE]:
857
            needed_mem += bep[constants.BE_MEMORY]
858
        if nodeinfo['mfree'] < needed_mem:
859
          feedback_fn("  - ERROR: not enough memory on node %s to accomodate"
860
                      " failovers should node %s fail" % (node, prinode))
861
          bad = True
862
    return bad
863

    
864
  def CheckPrereq(self):
865
    """Check prerequisites.
866

867
    Transform the list of checks we're going to skip into a set and check that
868
    all its members are valid.
869

870
    """
871
    self.skip_set = frozenset(self.op.skip_checks)
872
    if not constants.VERIFY_OPTIONAL_CHECKS.issuperset(self.skip_set):
873
      raise errors.OpPrereqError("Invalid checks to be skipped specified")
874

    
875
  def BuildHooksEnv(self):
876
    """Build hooks env.
877

878
    Cluster-Verify hooks just rone in the post phase and their failure makes
879
    the output be logged in the verify output and the verification to fail.
880

881
    """
882
    all_nodes = self.cfg.GetNodeList()
883
    # TODO: populate the environment with useful information for verify hooks
884
    env = {}
885
    return env, [], all_nodes
886

    
887
  def Exec(self, feedback_fn):
888
    """Verify integrity of cluster, performing various test on nodes.
889

890
    """
891
    bad = False
892
    feedback_fn("* Verifying global settings")
893
    for msg in self.cfg.VerifyConfig():
894
      feedback_fn("  - ERROR: %s" % msg)
895

    
896
    vg_name = self.cfg.GetVGName()
897
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
898
    nodelist = utils.NiceSort(self.cfg.GetNodeList())
899
    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
900
    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
901
    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
902
                        for iname in instancelist)
903
    i_non_redundant = [] # Non redundant instances
904
    i_non_a_balanced = [] # Non auto-balanced instances
905
    n_offline = [] # List of offline nodes
906
    node_volume = {}
907
    node_instance = {}
908
    node_info = {}
909
    instance_cfg = {}
910

    
911
    # FIXME: verify OS list
912
    # do local checksums
913
    master_files = [constants.CLUSTER_CONF_FILE]
914

    
915
    file_names = ssconf.SimpleStore().GetFileList()
916
    file_names.append(constants.SSL_CERT_FILE)
917
    file_names.append(constants.RAPI_CERT_FILE)
918
    file_names.extend(master_files)
919

    
920
    local_checksums = utils.FingerprintFiles(file_names)
921

    
922
    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
923
    node_verify_param = {
924
      constants.NV_FILELIST: file_names,
925
      constants.NV_NODELIST: [node.name for node in nodeinfo
926
                              if not node.offline],
927
      constants.NV_HYPERVISOR: hypervisors,
928
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
929
                                  node.secondary_ip) for node in nodeinfo
930
                                 if not node.offline],
931
      constants.NV_LVLIST: vg_name,
932
      constants.NV_INSTANCELIST: hypervisors,
933
      constants.NV_VGLIST: None,
934
      constants.NV_VERSION: None,
935
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
936
      constants.NV_DRBDLIST: None,
937
      }
938
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
939
                                           self.cfg.GetClusterName())
940

    
941
    cluster = self.cfg.GetClusterInfo()
942
    master_node = self.cfg.GetMasterNode()
943
    all_drbd_map = self.cfg.ComputeDRBDMap()
944

    
945
    for node_i in nodeinfo:
946
      node = node_i.name
947
      nresult = all_nvinfo[node].data
948

    
949
      if node_i.offline:
950
        feedback_fn("* Skipping offline node %s" % (node,))
951
        n_offline.append(node)
952
        continue
953

    
954
      if node == master_node:
955
        ntype = "master"
956
      elif node_i.master_candidate:
957
        ntype = "master candidate"
958
      else:
959
        ntype = "regular"
960
      feedback_fn("* Verifying node %s (%s)" % (node, ntype))
961

    
962
      if all_nvinfo[node].failed or not isinstance(nresult, dict):
963
        feedback_fn("  - ERROR: connection to %s failed" % (node,))
964
        bad = True
965
        continue
966

    
967
      node_drbd = {}
968
      for minor, instance in all_drbd_map[node].items():
969
        instance = instanceinfo[instance]
970
        node_drbd[minor] = (instance.name, instance.admin_up)
971
      result = self._VerifyNode(node_i, file_names, local_checksums,
972
                                nresult, feedback_fn, master_files,
973
                                node_drbd)
974
      bad = bad or result
975

    
976
      lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
977
      if isinstance(lvdata, basestring):
978
        feedback_fn("  - ERROR: LVM problem on node %s: %s" %
979
                    (node, lvdata.encode('string_escape')))
980
        bad = True
981
        node_volume[node] = {}
982
      elif not isinstance(lvdata, dict):
983
        feedback_fn("  - ERROR: connection to %s failed (lvlist)" % (node,))
984
        bad = True
985
        continue
986
      else:
987
        node_volume[node] = lvdata
988

    
989
      # node_instance
990
      idata = nresult.get(constants.NV_INSTANCELIST, None)
991
      if not isinstance(idata, list):
992
        feedback_fn("  - ERROR: connection to %s failed (instancelist)" %
993
                    (node,))
994
        bad = True
995
        continue
996

    
997
      node_instance[node] = idata
998

    
999
      # node_info
1000
      nodeinfo = nresult.get(constants.NV_HVINFO, None)
1001
      if not isinstance(nodeinfo, dict):
1002
        feedback_fn("  - ERROR: connection to %s failed (hvinfo)" % (node,))
1003
        bad = True
1004
        continue
1005

    
1006
      try:
1007
        node_info[node] = {
1008
          "mfree": int(nodeinfo['memory_free']),
1009
          "dfree": int(nresult[constants.NV_VGLIST][vg_name]),
1010
          "pinst": [],
1011
          "sinst": [],
1012
          # dictionary holding all instances this node is secondary for,
1013
          # grouped by their primary node. Each key is a cluster node, and each
1014
          # value is a list of instances which have the key as primary and the
1015
          # current node as secondary.  this is handy to calculate N+1 memory
1016
          # availability if you can only failover from a primary to its
1017
          # secondary.
1018
          "sinst-by-pnode": {},
1019
        }
1020
      except ValueError:
1021
        feedback_fn("  - ERROR: invalid value returned from node %s" % (node,))
1022
        bad = True
1023
        continue
1024

    
1025
    node_vol_should = {}
1026

    
1027
    for instance in instancelist:
1028
      feedback_fn("* Verifying instance %s" % instance)
1029
      inst_config = instanceinfo[instance]
1030
      result =  self._VerifyInstance(instance, inst_config, node_volume,
1031
                                     node_instance, feedback_fn, n_offline)
1032
      bad = bad or result
1033
      inst_nodes_offline = []
1034

    
1035
      inst_config.MapLVsByNode(node_vol_should)
1036

    
1037
      instance_cfg[instance] = inst_config
1038

    
1039
      pnode = inst_config.primary_node
1040
      if pnode in node_info:
1041
        node_info[pnode]['pinst'].append(instance)
1042
      elif pnode not in n_offline:
1043
        feedback_fn("  - ERROR: instance %s, connection to primary node"
1044
                    " %s failed" % (instance, pnode))
1045
        bad = True
1046

    
1047
      if pnode in n_offline:
1048
        inst_nodes_offline.append(pnode)
1049

    
1050
      # If the instance is non-redundant we cannot survive losing its primary
1051
      # node, so we are not N+1 compliant. On the other hand we have no disk
1052
      # templates with more than one secondary so that situation is not well
1053
      # supported either.
1054
      # FIXME: does not support file-backed instances
1055
      if len(inst_config.secondary_nodes) == 0:
1056
        i_non_redundant.append(instance)
1057
      elif len(inst_config.secondary_nodes) > 1:
1058
        feedback_fn("  - WARNING: multiple secondaries for instance %s"
1059
                    % instance)
1060

    
1061
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
1062
        i_non_a_balanced.append(instance)
1063

    
1064
      for snode in inst_config.secondary_nodes:
1065
        if snode in node_info:
1066
          node_info[snode]['sinst'].append(instance)
1067
          if pnode not in node_info[snode]['sinst-by-pnode']:
1068
            node_info[snode]['sinst-by-pnode'][pnode] = []
1069
          node_info[snode]['sinst-by-pnode'][pnode].append(instance)
1070
        elif snode not in n_offline:
1071
          feedback_fn("  - ERROR: instance %s, connection to secondary node"
1072
                      " %s failed" % (instance, snode))
1073
          bad = True
1074
        if snode in n_offline:
1075
          inst_nodes_offline.append(snode)
1076

    
1077
      if inst_nodes_offline:
1078
        # warn that the instance lives on offline nodes, and set bad=True
1079
        feedback_fn("  - ERROR: instance lives on offline node(s) %s" %
1080
                    ", ".join(inst_nodes_offline))
1081
        bad = True
1082

    
1083
    feedback_fn("* Verifying orphan volumes")
1084
    result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
1085
                                       feedback_fn)
1086
    bad = bad or result
1087

    
1088
    feedback_fn("* Verifying remaining instances")
1089
    result = self._VerifyOrphanInstances(instancelist, node_instance,
1090
                                         feedback_fn)
1091
    bad = bad or result
1092

    
1093
    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
1094
      feedback_fn("* Verifying N+1 Memory redundancy")
1095
      result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
1096
      bad = bad or result
1097

    
1098
    feedback_fn("* Other Notes")
1099
    if i_non_redundant:
1100
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
1101
                  % len(i_non_redundant))
1102

    
1103
    if i_non_a_balanced:
1104
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
1105
                  % len(i_non_a_balanced))
1106

    
1107
    if n_offline:
1108
      feedback_fn("  - NOTICE: %d offline node(s) found." % len(n_offline))
1109

    
1110
    return not bad
1111

    
1112
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
1113
    """Analize the post-hooks' result
1114

1115
    This method analyses the hook result, handles it, and sends some
1116
    nicely-formatted feedback back to the user.
1117

1118
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
1119
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
1120
    @param hooks_results: the results of the multi-node hooks rpc call
1121
    @param feedback_fn: function used send feedback back to the caller
1122
    @param lu_result: previous Exec result
1123
    @return: the new Exec result, based on the previous result
1124
        and hook results
1125

1126
    """
1127
    # We only really run POST phase hooks, and are only interested in
1128
    # their results
1129
    if phase == constants.HOOKS_PHASE_POST:
1130
      # Used to change hooks' output to proper indentation
1131
      indent_re = re.compile('^', re.M)
1132
      feedback_fn("* Hooks Results")
1133
      if not hooks_results:
1134
        feedback_fn("  - ERROR: general communication failure")
1135
        lu_result = 1
1136
      else:
1137
        for node_name in hooks_results:
1138
          show_node_header = True
1139
          res = hooks_results[node_name]
1140
          if res.failed or res.data is False or not isinstance(res.data, list):
1141
            if res.offline:
1142
              # no need to warn or set fail return value
1143
              continue
1144
            feedback_fn("    Communication failure in hooks execution")
1145
            lu_result = 1
1146
            continue
1147
          for script, hkr, output in res.data:
1148
            if hkr == constants.HKR_FAIL:
1149
              # The node header is only shown once, if there are
1150
              # failing hooks on that node
1151
              if show_node_header:
1152
                feedback_fn("  Node %s:" % node_name)
1153
                show_node_header = False
1154
              feedback_fn("    ERROR: Script %s failed, output:" % script)
1155
              output = indent_re.sub('      ', output)
1156
              feedback_fn("%s" % output)
1157
              lu_result = 1
1158

    
1159
      return lu_result
1160

    
1161

    
1162
class LUVerifyDisks(NoHooksLU):
1163
  """Verifies the cluster disks status.
1164

1165
  """
1166
  _OP_REQP = []
1167
  REQ_BGL = False
1168

    
1169
  def ExpandNames(self):
1170
    self.needed_locks = {
1171
      locking.LEVEL_NODE: locking.ALL_SET,
1172
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1173
    }
1174
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
1175

    
1176
  def CheckPrereq(self):
1177
    """Check prerequisites.
1178

1179
    This has no prerequisites.
1180

1181
    """
1182
    pass
1183

    
1184
  def Exec(self, feedback_fn):
1185
    """Verify integrity of cluster disks.
1186

1187
    """
1188
    result = res_nodes, res_nlvm, res_instances, res_missing = [], {}, [], {}
1189

    
1190
    vg_name = self.cfg.GetVGName()
1191
    nodes = utils.NiceSort(self.cfg.GetNodeList())
1192
    instances = [self.cfg.GetInstanceInfo(name)
1193
                 for name in self.cfg.GetInstanceList()]
1194

    
1195
    nv_dict = {}
1196
    for inst in instances:
1197
      inst_lvs = {}
1198
      if (not inst.admin_up or
1199
          inst.disk_template not in constants.DTS_NET_MIRROR):
1200
        continue
1201
      inst.MapLVsByNode(inst_lvs)
1202
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
1203
      for node, vol_list in inst_lvs.iteritems():
1204
        for vol in vol_list:
1205
          nv_dict[(node, vol)] = inst
1206

    
1207
    if not nv_dict:
1208
      return result
1209

    
1210
    node_lvs = self.rpc.call_volume_list(nodes, vg_name)
1211

    
1212
    to_act = set()
1213
    for node in nodes:
1214
      # node_volume
1215
      lvs = node_lvs[node]
1216
      if lvs.failed:
1217
        if not lvs.offline:
1218
          self.LogWarning("Connection to node %s failed: %s" %
1219
                          (node, lvs.data))
1220
        continue
1221
      lvs = lvs.data
1222
      if isinstance(lvs, basestring):
1223
        logging.warning("Error enumerating LVs on node %s: %s", node, lvs)
1224
        res_nlvm[node] = lvs
1225
      elif not isinstance(lvs, dict):
1226
        logging.warning("Connection to node %s failed or invalid data"
1227
                        " returned", node)
1228
        res_nodes.append(node)
1229
        continue
1230

    
1231
      for lv_name, (_, lv_inactive, lv_online) in lvs.iteritems():
1232
        inst = nv_dict.pop((node, lv_name), None)
1233
        if (not lv_online and inst is not None
1234
            and inst.name not in res_instances):
1235
          res_instances.append(inst.name)
1236

    
1237
    # any leftover items in nv_dict are missing LVs, let's arrange the
1238
    # data better
1239
    for key, inst in nv_dict.iteritems():
1240
      if inst.name not in res_missing:
1241
        res_missing[inst.name] = []
1242
      res_missing[inst.name].append(key)
1243

    
1244
    return result
1245

    
1246

    
1247
class LURenameCluster(LogicalUnit):
1248
  """Rename the cluster.
1249

1250
  """
1251
  HPATH = "cluster-rename"
1252
  HTYPE = constants.HTYPE_CLUSTER
1253
  _OP_REQP = ["name"]
1254

    
1255
  def BuildHooksEnv(self):
1256
    """Build hooks env.
1257

1258
    """
1259
    env = {
1260
      "OP_TARGET": self.cfg.GetClusterName(),
1261
      "NEW_NAME": self.op.name,
1262
      }
1263
    mn = self.cfg.GetMasterNode()
1264
    return env, [mn], [mn]
1265

    
1266
  def CheckPrereq(self):
1267
    """Verify that the passed name is a valid one.
1268

1269
    """
1270
    hostname = utils.HostInfo(self.op.name)
1271

    
1272
    new_name = hostname.name
1273
    self.ip = new_ip = hostname.ip
1274
    old_name = self.cfg.GetClusterName()
1275
    old_ip = self.cfg.GetMasterIP()
1276
    if new_name == old_name and new_ip == old_ip:
1277
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
1278
                                 " cluster has changed")
1279
    if new_ip != old_ip:
1280
      if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
1281
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
1282
                                   " reachable on the network. Aborting." %
1283
                                   new_ip)
1284

    
1285
    self.op.name = new_name
1286

    
1287
  def Exec(self, feedback_fn):
1288
    """Rename the cluster.
1289

1290
    """
1291
    clustername = self.op.name
1292
    ip = self.ip
1293

    
1294
    # shutdown the master IP
1295
    master = self.cfg.GetMasterNode()
1296
    result = self.rpc.call_node_stop_master(master, False)
1297
    if result.failed or not result.data:
1298
      raise errors.OpExecError("Could not disable the master role")
1299

    
1300
    try:
1301
      cluster = self.cfg.GetClusterInfo()
1302
      cluster.cluster_name = clustername
1303
      cluster.master_ip = ip
1304
      self.cfg.Update(cluster)
1305

    
1306
      # update the known hosts file
1307
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
1308
      node_list = self.cfg.GetNodeList()
1309
      try:
1310
        node_list.remove(master)
1311
      except ValueError:
1312
        pass
1313
      result = self.rpc.call_upload_file(node_list,
1314
                                         constants.SSH_KNOWN_HOSTS_FILE)
1315
      for to_node, to_result in result.iteritems():
1316
        if to_result.failed or not to_result.data:
1317
          logging.error("Copy of file %s to node %s failed",
1318
                        constants.SSH_KNOWN_HOSTS_FILE, to_node)
1319

    
1320
    finally:
1321
      result = self.rpc.call_node_start_master(master, False)
1322
      if result.failed or not result.data:
1323
        self.LogWarning("Could not re-enable the master role on"
1324
                        " the master, please restart manually.")
1325

    
1326

    
1327
def _RecursiveCheckIfLVMBased(disk):
1328
  """Check if the given disk or its children are lvm-based.
1329

1330
  @type disk: L{objects.Disk}
1331
  @param disk: the disk to check
1332
  @rtype: booleean
1333
  @return: boolean indicating whether a LD_LV dev_type was found or not
1334

1335
  """
1336
  if disk.children:
1337
    for chdisk in disk.children:
1338
      if _RecursiveCheckIfLVMBased(chdisk):
1339
        return True
1340
  return disk.dev_type == constants.LD_LV
1341

    
1342

    
1343
class LUSetClusterParams(LogicalUnit):
1344
  """Change the parameters of the cluster.
1345

1346
  """
1347
  HPATH = "cluster-modify"
1348
  HTYPE = constants.HTYPE_CLUSTER
1349
  _OP_REQP = []
1350
  REQ_BGL = False
1351

    
1352
  def CheckParameters(self):
1353
    """Check parameters
1354

1355
    """
1356
    if not hasattr(self.op, "candidate_pool_size"):
1357
      self.op.candidate_pool_size = None
1358
    if self.op.candidate_pool_size is not None:
1359
      try:
1360
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1361
      except ValueError, err:
1362
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1363
                                   str(err))
1364
      if self.op.candidate_pool_size < 1:
1365
        raise errors.OpPrereqError("At least one master candidate needed")
1366

    
1367
  def ExpandNames(self):
1368
    # FIXME: in the future maybe other cluster params won't require checking on
1369
    # all nodes to be modified.
1370
    self.needed_locks = {
1371
      locking.LEVEL_NODE: locking.ALL_SET,
1372
    }
1373
    self.share_locks[locking.LEVEL_NODE] = 1
1374

    
1375
  def BuildHooksEnv(self):
1376
    """Build hooks env.
1377

1378
    """
1379
    env = {
1380
      "OP_TARGET": self.cfg.GetClusterName(),
1381
      "NEW_VG_NAME": self.op.vg_name,
1382
      }
1383
    mn = self.cfg.GetMasterNode()
1384
    return env, [mn], [mn]
1385

    
1386
  def CheckPrereq(self):
1387
    """Check prerequisites.
1388

1389
    This checks whether the given params don't conflict and
1390
    if the given volume group is valid.
1391

1392
    """
1393
    # FIXME: This only works because there is only one parameter that can be
1394
    # changed or removed.
1395
    if self.op.vg_name is not None and not self.op.vg_name:
1396
      instances = self.cfg.GetAllInstancesInfo().values()
1397
      for inst in instances:
1398
        for disk in inst.disks:
1399
          if _RecursiveCheckIfLVMBased(disk):
1400
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1401
                                       " lvm-based instances exist")
1402

    
1403
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1404

    
1405
    # if vg_name not None, checks given volume group on all nodes
1406
    if self.op.vg_name:
1407
      vglist = self.rpc.call_vg_list(node_list)
1408
      for node in node_list:
1409
        if vglist[node].failed:
1410
          # ignoring down node
1411
          self.LogWarning("Node %s unreachable/error, ignoring" % node)
1412
          continue
1413
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].data,
1414
                                              self.op.vg_name,
1415
                                              constants.MIN_VG_SIZE)
1416
        if vgstatus:
1417
          raise errors.OpPrereqError("Error on node '%s': %s" %
1418
                                     (node, vgstatus))
1419

    
1420
    self.cluster = cluster = self.cfg.GetClusterInfo()
1421
    # validate beparams changes
1422
    if self.op.beparams:
1423
      utils.CheckBEParams(self.op.beparams)
1424
      self.new_beparams = cluster.FillDict(
1425
        cluster.beparams[constants.BEGR_DEFAULT], self.op.beparams)
1426

    
1427
    # hypervisor list/parameters
1428
    self.new_hvparams = cluster.FillDict(cluster.hvparams, {})
1429
    if self.op.hvparams:
1430
      if not isinstance(self.op.hvparams, dict):
1431
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1432
      for hv_name, hv_dict in self.op.hvparams.items():
1433
        if hv_name not in self.new_hvparams:
1434
          self.new_hvparams[hv_name] = hv_dict
1435
        else:
1436
          self.new_hvparams[hv_name].update(hv_dict)
1437

    
1438
    if self.op.enabled_hypervisors is not None:
1439
      self.hv_list = self.op.enabled_hypervisors
1440
    else:
1441
      self.hv_list = cluster.enabled_hypervisors
1442

    
1443
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1444
      # either the enabled list has changed, or the parameters have, validate
1445
      for hv_name, hv_params in self.new_hvparams.items():
1446
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1447
            (self.op.enabled_hypervisors and
1448
             hv_name in self.op.enabled_hypervisors)):
1449
          # either this is a new hypervisor, or its parameters have changed
1450
          hv_class = hypervisor.GetHypervisor(hv_name)
1451
          hv_class.CheckParameterSyntax(hv_params)
1452
          _CheckHVParams(self, node_list, hv_name, hv_params)
1453

    
1454
  def Exec(self, feedback_fn):
1455
    """Change the parameters of the cluster.
1456

1457
    """
1458
    if self.op.vg_name is not None:
1459
      if self.op.vg_name != self.cfg.GetVGName():
1460
        self.cfg.SetVGName(self.op.vg_name)
1461
      else:
1462
        feedback_fn("Cluster LVM configuration already in desired"
1463
                    " state, not changing")
1464
    if self.op.hvparams:
1465
      self.cluster.hvparams = self.new_hvparams
1466
    if self.op.enabled_hypervisors is not None:
1467
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1468
    if self.op.beparams:
1469
      self.cluster.beparams[constants.BEGR_DEFAULT] = self.new_beparams
1470
    if self.op.candidate_pool_size is not None:
1471
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1472

    
1473
    self.cfg.Update(self.cluster)
1474

    
1475
    # we want to update nodes after the cluster so that if any errors
1476
    # happen, we have recorded and saved the cluster info
1477
    if self.op.candidate_pool_size is not None:
1478
      _AdjustCandidatePool(self)
1479

    
1480

    
1481
class LURedistributeConfig(NoHooksLU):
1482
  """Force the redistribution of cluster configuration.
1483

1484
  This is a very simple LU.
1485

1486
  """
1487
  _OP_REQP = []
1488
  REQ_BGL = False
1489

    
1490
  def ExpandNames(self):
1491
    self.needed_locks = {
1492
      locking.LEVEL_NODE: locking.ALL_SET,
1493
    }
1494
    self.share_locks[locking.LEVEL_NODE] = 1
1495

    
1496
  def CheckPrereq(self):
1497
    """Check prerequisites.
1498

1499
    """
1500

    
1501
  def Exec(self, feedback_fn):
1502
    """Redistribute the configuration.
1503

1504
    """
1505
    self.cfg.Update(self.cfg.GetClusterInfo())
1506

    
1507

    
1508
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1509
  """Sleep and poll for an instance's disk to sync.
1510

1511
  """
1512
  if not instance.disks:
1513
    return True
1514

    
1515
  if not oneshot:
1516
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1517

    
1518
  node = instance.primary_node
1519

    
1520
  for dev in instance.disks:
1521
    lu.cfg.SetDiskID(dev, node)
1522

    
1523
  retries = 0
1524
  while True:
1525
    max_time = 0
1526
    done = True
1527
    cumul_degraded = False
1528
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1529
    if rstats.failed or not rstats.data:
1530
      lu.LogWarning("Can't get any data from node %s", node)
1531
      retries += 1
1532
      if retries >= 10:
1533
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1534
                                 " aborting." % node)
1535
      time.sleep(6)
1536
      continue
1537
    rstats = rstats.data
1538
    retries = 0
1539
    for i, mstat in enumerate(rstats):
1540
      if mstat is None:
1541
        lu.LogWarning("Can't compute data for node %s/%s",
1542
                           node, instance.disks[i].iv_name)
1543
        continue
1544
      # we ignore the ldisk parameter
1545
      perc_done, est_time, is_degraded, _ = mstat
1546
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1547
      if perc_done is not None:
1548
        done = False
1549
        if est_time is not None:
1550
          rem_time = "%d estimated seconds remaining" % est_time
1551
          max_time = est_time
1552
        else:
1553
          rem_time = "no time estimate"
1554
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1555
                        (instance.disks[i].iv_name, perc_done, rem_time))
1556
    if done or oneshot:
1557
      break
1558

    
1559
    time.sleep(min(60, max_time))
1560

    
1561
  if done:
1562
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1563
  return not cumul_degraded
1564

    
1565

    
1566
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1567
  """Check that mirrors are not degraded.
1568

1569
  The ldisk parameter, if True, will change the test from the
1570
  is_degraded attribute (which represents overall non-ok status for
1571
  the device(s)) to the ldisk (representing the local storage status).
1572

1573
  """
1574
  lu.cfg.SetDiskID(dev, node)
1575
  if ldisk:
1576
    idx = 6
1577
  else:
1578
    idx = 5
1579

    
1580
  result = True
1581
  if on_primary or dev.AssembleOnSecondary():
1582
    rstats = lu.rpc.call_blockdev_find(node, dev)
1583
    if rstats.failed or not rstats.data:
1584
      logging.warning("Node %s: disk degraded, not found or node down", node)
1585
      result = False
1586
    else:
1587
      result = result and (not rstats.data[idx])
1588
  if dev.children:
1589
    for child in dev.children:
1590
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1591

    
1592
  return result
1593

    
1594

    
1595
class LUDiagnoseOS(NoHooksLU):
1596
  """Logical unit for OS diagnose/query.
1597

1598
  """
1599
  _OP_REQP = ["output_fields", "names"]
1600
  REQ_BGL = False
1601
  _FIELDS_STATIC = utils.FieldSet()
1602
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1603

    
1604
  def ExpandNames(self):
1605
    if self.op.names:
1606
      raise errors.OpPrereqError("Selective OS query not supported")
1607

    
1608
    _CheckOutputFields(static=self._FIELDS_STATIC,
1609
                       dynamic=self._FIELDS_DYNAMIC,
1610
                       selected=self.op.output_fields)
1611

    
1612
    # Lock all nodes, in shared mode
1613
    self.needed_locks = {}
1614
    self.share_locks[locking.LEVEL_NODE] = 1
1615
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1616

    
1617
  def CheckPrereq(self):
1618
    """Check prerequisites.
1619

1620
    """
1621

    
1622
  @staticmethod
1623
  def _DiagnoseByOS(node_list, rlist):
1624
    """Remaps a per-node return list into an a per-os per-node dictionary
1625

1626
    @param node_list: a list with the names of all nodes
1627
    @param rlist: a map with node names as keys and OS objects as values
1628

1629
    @rtype: dict
1630
    @returns: a dictionary with osnames as keys and as value another map, with
1631
        nodes as keys and list of OS objects as values, eg::
1632

1633
          {"debian-etch": {"node1": [<object>,...],
1634
                           "node2": [<object>,]}
1635
          }
1636

1637
    """
1638
    all_os = {}
1639
    for node_name, nr in rlist.iteritems():
1640
      if nr.failed or not nr.data:
1641
        continue
1642
      for os_obj in nr.data:
1643
        if os_obj.name not in all_os:
1644
          # build a list of nodes for this os containing empty lists
1645
          # for each node in node_list
1646
          all_os[os_obj.name] = {}
1647
          for nname in node_list:
1648
            all_os[os_obj.name][nname] = []
1649
        all_os[os_obj.name][node_name].append(os_obj)
1650
    return all_os
1651

    
1652
  def Exec(self, feedback_fn):
1653
    """Compute the list of OSes.
1654

1655
    """
1656
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1657
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()
1658
                   if node in node_list]
1659
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1660
    if node_data == False:
1661
      raise errors.OpExecError("Can't gather the list of OSes")
1662
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1663
    output = []
1664
    for os_name, os_data in pol.iteritems():
1665
      row = []
1666
      for field in self.op.output_fields:
1667
        if field == "name":
1668
          val = os_name
1669
        elif field == "valid":
1670
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1671
        elif field == "node_status":
1672
          val = {}
1673
          for node_name, nos_list in os_data.iteritems():
1674
            val[node_name] = [(v.status, v.path) for v in nos_list]
1675
        else:
1676
          raise errors.ParameterError(field)
1677
        row.append(val)
1678
      output.append(row)
1679

    
1680
    return output
1681

    
1682

    
1683
class LURemoveNode(LogicalUnit):
1684
  """Logical unit for removing a node.
1685

1686
  """
1687
  HPATH = "node-remove"
1688
  HTYPE = constants.HTYPE_NODE
1689
  _OP_REQP = ["node_name"]
1690

    
1691
  def BuildHooksEnv(self):
1692
    """Build hooks env.
1693

1694
    This doesn't run on the target node in the pre phase as a failed
1695
    node would then be impossible to remove.
1696

1697
    """
1698
    env = {
1699
      "OP_TARGET": self.op.node_name,
1700
      "NODE_NAME": self.op.node_name,
1701
      }
1702
    all_nodes = self.cfg.GetNodeList()
1703
    all_nodes.remove(self.op.node_name)
1704
    return env, all_nodes, all_nodes
1705

    
1706
  def CheckPrereq(self):
1707
    """Check prerequisites.
1708

1709
    This checks:
1710
     - the node exists in the configuration
1711
     - it does not have primary or secondary instances
1712
     - it's not the master
1713

1714
    Any errors are signalled by raising errors.OpPrereqError.
1715

1716
    """
1717
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1718
    if node is None:
1719
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1720

    
1721
    instance_list = self.cfg.GetInstanceList()
1722

    
1723
    masternode = self.cfg.GetMasterNode()
1724
    if node.name == masternode:
1725
      raise errors.OpPrereqError("Node is the master node,"
1726
                                 " you need to failover first.")
1727

    
1728
    for instance_name in instance_list:
1729
      instance = self.cfg.GetInstanceInfo(instance_name)
1730
      if node.name in instance.all_nodes:
1731
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1732
                                   " please remove first." % instance_name)
1733
    self.op.node_name = node.name
1734
    self.node = node
1735

    
1736
  def Exec(self, feedback_fn):
1737
    """Removes the node from the cluster.
1738

1739
    """
1740
    node = self.node
1741
    logging.info("Stopping the node daemon and removing configs from node %s",
1742
                 node.name)
1743

    
1744
    self.context.RemoveNode(node.name)
1745

    
1746
    self.rpc.call_node_leave_cluster(node.name)
1747

    
1748
    # Promote nodes to master candidate as needed
1749
    _AdjustCandidatePool(self)
1750

    
1751

    
1752
class LUQueryNodes(NoHooksLU):
1753
  """Logical unit for querying nodes.
1754

1755
  """
1756
  _OP_REQP = ["output_fields", "names"]
1757
  REQ_BGL = False
1758
  _FIELDS_DYNAMIC = utils.FieldSet(
1759
    "dtotal", "dfree",
1760
    "mtotal", "mnode", "mfree",
1761
    "bootid",
1762
    "ctotal",
1763
    )
1764

    
1765
  _FIELDS_STATIC = utils.FieldSet(
1766
    "name", "pinst_cnt", "sinst_cnt",
1767
    "pinst_list", "sinst_list",
1768
    "pip", "sip", "tags",
1769
    "serial_no",
1770
    "master_candidate",
1771
    "master",
1772
    "offline",
1773
    )
1774

    
1775
  def ExpandNames(self):
1776
    _CheckOutputFields(static=self._FIELDS_STATIC,
1777
                       dynamic=self._FIELDS_DYNAMIC,
1778
                       selected=self.op.output_fields)
1779

    
1780
    self.needed_locks = {}
1781
    self.share_locks[locking.LEVEL_NODE] = 1
1782

    
1783
    if self.op.names:
1784
      self.wanted = _GetWantedNodes(self, self.op.names)
1785
    else:
1786
      self.wanted = locking.ALL_SET
1787

    
1788
    self.do_locking = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
1789
    if self.do_locking:
1790
      # if we don't request only static fields, we need to lock the nodes
1791
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
1792

    
1793

    
1794
  def CheckPrereq(self):
1795
    """Check prerequisites.
1796

1797
    """
1798
    # The validation of the node list is done in the _GetWantedNodes,
1799
    # if non empty, and if empty, there's no validation to do
1800
    pass
1801

    
1802
  def Exec(self, feedback_fn):
1803
    """Computes the list of nodes and their attributes.
1804

1805
    """
1806
    all_info = self.cfg.GetAllNodesInfo()
1807
    if self.do_locking:
1808
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
1809
    elif self.wanted != locking.ALL_SET:
1810
      nodenames = self.wanted
1811
      missing = set(nodenames).difference(all_info.keys())
1812
      if missing:
1813
        raise errors.OpExecError(
1814
          "Some nodes were removed before retrieving their data: %s" % missing)
1815
    else:
1816
      nodenames = all_info.keys()
1817

    
1818
    nodenames = utils.NiceSort(nodenames)
1819
    nodelist = [all_info[name] for name in nodenames]
1820

    
1821
    # begin data gathering
1822

    
1823
    if self.do_locking:
1824
      live_data = {}
1825
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
1826
                                          self.cfg.GetHypervisorType())
1827
      for name in nodenames:
1828
        nodeinfo = node_data[name]
1829
        if not nodeinfo.failed and nodeinfo.data:
1830
          nodeinfo = nodeinfo.data
1831
          fn = utils.TryConvert
1832
          live_data[name] = {
1833
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
1834
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
1835
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
1836
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
1837
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
1838
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
1839
            "bootid": nodeinfo.get('bootid', None),
1840
            }
1841
        else:
1842
          live_data[name] = {}
1843
    else:
1844
      live_data = dict.fromkeys(nodenames, {})
1845

    
1846
    node_to_primary = dict([(name, set()) for name in nodenames])
1847
    node_to_secondary = dict([(name, set()) for name in nodenames])
1848

    
1849
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
1850
                             "sinst_cnt", "sinst_list"))
1851
    if inst_fields & frozenset(self.op.output_fields):
1852
      instancelist = self.cfg.GetInstanceList()
1853

    
1854
      for instance_name in instancelist:
1855
        inst = self.cfg.GetInstanceInfo(instance_name)
1856
        if inst.primary_node in node_to_primary:
1857
          node_to_primary[inst.primary_node].add(inst.name)
1858
        for secnode in inst.secondary_nodes:
1859
          if secnode in node_to_secondary:
1860
            node_to_secondary[secnode].add(inst.name)
1861

    
1862
    master_node = self.cfg.GetMasterNode()
1863

    
1864
    # end data gathering
1865

    
1866
    output = []
1867
    for node in nodelist:
1868
      node_output = []
1869
      for field in self.op.output_fields:
1870
        if field == "name":
1871
          val = node.name
1872
        elif field == "pinst_list":
1873
          val = list(node_to_primary[node.name])
1874
        elif field == "sinst_list":
1875
          val = list(node_to_secondary[node.name])
1876
        elif field == "pinst_cnt":
1877
          val = len(node_to_primary[node.name])
1878
        elif field == "sinst_cnt":
1879
          val = len(node_to_secondary[node.name])
1880
        elif field == "pip":
1881
          val = node.primary_ip
1882
        elif field == "sip":
1883
          val = node.secondary_ip
1884
        elif field == "tags":
1885
          val = list(node.GetTags())
1886
        elif field == "serial_no":
1887
          val = node.serial_no
1888
        elif field == "master_candidate":
1889
          val = node.master_candidate
1890
        elif field == "master":
1891
          val = node.name == master_node
1892
        elif field == "offline":
1893
          val = node.offline
1894
        elif self._FIELDS_DYNAMIC.Matches(field):
1895
          val = live_data[node.name].get(field, None)
1896
        else:
1897
          raise errors.ParameterError(field)
1898
        node_output.append(val)
1899
      output.append(node_output)
1900

    
1901
    return output
1902

    
1903

    
1904
class LUQueryNodeVolumes(NoHooksLU):
1905
  """Logical unit for getting volumes on node(s).
1906

1907
  """
1908
  _OP_REQP = ["nodes", "output_fields"]
1909
  REQ_BGL = False
1910
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
1911
  _FIELDS_STATIC = utils.FieldSet("node")
1912

    
1913
  def ExpandNames(self):
1914
    _CheckOutputFields(static=self._FIELDS_STATIC,
1915
                       dynamic=self._FIELDS_DYNAMIC,
1916
                       selected=self.op.output_fields)
1917

    
1918
    self.needed_locks = {}
1919
    self.share_locks[locking.LEVEL_NODE] = 1
1920
    if not self.op.nodes:
1921
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1922
    else:
1923
      self.needed_locks[locking.LEVEL_NODE] = \
1924
        _GetWantedNodes(self, self.op.nodes)
1925

    
1926
  def CheckPrereq(self):
1927
    """Check prerequisites.
1928

1929
    This checks that the fields required are valid output fields.
1930

1931
    """
1932
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
1933

    
1934
  def Exec(self, feedback_fn):
1935
    """Computes the list of nodes and their attributes.
1936

1937
    """
1938
    nodenames = self.nodes
1939
    volumes = self.rpc.call_node_volumes(nodenames)
1940

    
1941
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
1942
             in self.cfg.GetInstanceList()]
1943

    
1944
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
1945

    
1946
    output = []
1947
    for node in nodenames:
1948
      if node not in volumes or volumes[node].failed or not volumes[node].data:
1949
        continue
1950

    
1951
      node_vols = volumes[node].data[:]
1952
      node_vols.sort(key=lambda vol: vol['dev'])
1953

    
1954
      for vol in node_vols:
1955
        node_output = []
1956
        for field in self.op.output_fields:
1957
          if field == "node":
1958
            val = node
1959
          elif field == "phys":
1960
            val = vol['dev']
1961
          elif field == "vg":
1962
            val = vol['vg']
1963
          elif field == "name":
1964
            val = vol['name']
1965
          elif field == "size":
1966
            val = int(float(vol['size']))
1967
          elif field == "instance":
1968
            for inst in ilist:
1969
              if node not in lv_by_node[inst]:
1970
                continue
1971
              if vol['name'] in lv_by_node[inst][node]:
1972
                val = inst.name
1973
                break
1974
            else:
1975
              val = '-'
1976
          else:
1977
            raise errors.ParameterError(field)
1978
          node_output.append(str(val))
1979

    
1980
        output.append(node_output)
1981

    
1982
    return output
1983

    
1984

    
1985
class LUAddNode(LogicalUnit):
1986
  """Logical unit for adding node to the cluster.
1987

1988
  """
1989
  HPATH = "node-add"
1990
  HTYPE = constants.HTYPE_NODE
1991
  _OP_REQP = ["node_name"]
1992

    
1993
  def BuildHooksEnv(self):
1994
    """Build hooks env.
1995

1996
    This will run on all nodes before, and on all nodes + the new node after.
1997

1998
    """
1999
    env = {
2000
      "OP_TARGET": self.op.node_name,
2001
      "NODE_NAME": self.op.node_name,
2002
      "NODE_PIP": self.op.primary_ip,
2003
      "NODE_SIP": self.op.secondary_ip,
2004
      }
2005
    nodes_0 = self.cfg.GetNodeList()
2006
    nodes_1 = nodes_0 + [self.op.node_name, ]
2007
    return env, nodes_0, nodes_1
2008

    
2009
  def CheckPrereq(self):
2010
    """Check prerequisites.
2011

2012
    This checks:
2013
     - the new node is not already in the config
2014
     - it is resolvable
2015
     - its parameters (single/dual homed) matches the cluster
2016

2017
    Any errors are signalled by raising errors.OpPrereqError.
2018

2019
    """
2020
    node_name = self.op.node_name
2021
    cfg = self.cfg
2022

    
2023
    dns_data = utils.HostInfo(node_name)
2024

    
2025
    node = dns_data.name
2026
    primary_ip = self.op.primary_ip = dns_data.ip
2027
    secondary_ip = getattr(self.op, "secondary_ip", None)
2028
    if secondary_ip is None:
2029
      secondary_ip = primary_ip
2030
    if not utils.IsValidIP(secondary_ip):
2031
      raise errors.OpPrereqError("Invalid secondary IP given")
2032
    self.op.secondary_ip = secondary_ip
2033

    
2034
    node_list = cfg.GetNodeList()
2035
    if not self.op.readd and node in node_list:
2036
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2037
                                 node)
2038
    elif self.op.readd and node not in node_list:
2039
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2040

    
2041
    for existing_node_name in node_list:
2042
      existing_node = cfg.GetNodeInfo(existing_node_name)
2043

    
2044
      if self.op.readd and node == existing_node_name:
2045
        if (existing_node.primary_ip != primary_ip or
2046
            existing_node.secondary_ip != secondary_ip):
2047
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2048
                                     " address configuration as before")
2049
        continue
2050

    
2051
      if (existing_node.primary_ip == primary_ip or
2052
          existing_node.secondary_ip == primary_ip or
2053
          existing_node.primary_ip == secondary_ip or
2054
          existing_node.secondary_ip == secondary_ip):
2055
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2056
                                   " existing node %s" % existing_node.name)
2057

    
2058
    # check that the type of the node (single versus dual homed) is the
2059
    # same as for the master
2060
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2061
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2062
    newbie_singlehomed = secondary_ip == primary_ip
2063
    if master_singlehomed != newbie_singlehomed:
2064
      if master_singlehomed:
2065
        raise errors.OpPrereqError("The master has no private ip but the"
2066
                                   " new node has one")
2067
      else:
2068
        raise errors.OpPrereqError("The master has a private ip but the"
2069
                                   " new node doesn't have one")
2070

    
2071
    # checks reachablity
2072
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2073
      raise errors.OpPrereqError("Node not reachable by ping")
2074

    
2075
    if not newbie_singlehomed:
2076
      # check reachability from my secondary ip to newbie's secondary ip
2077
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2078
                           source=myself.secondary_ip):
2079
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2080
                                   " based ping to noded port")
2081

    
2082
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2083
    mc_now, _ = self.cfg.GetMasterCandidateStats()
2084
    master_candidate = mc_now < cp_size
2085

    
2086
    self.new_node = objects.Node(name=node,
2087
                                 primary_ip=primary_ip,
2088
                                 secondary_ip=secondary_ip,
2089
                                 master_candidate=master_candidate,
2090
                                 offline=False)
2091

    
2092
  def Exec(self, feedback_fn):
2093
    """Adds the new node to the cluster.
2094

2095
    """
2096
    new_node = self.new_node
2097
    node = new_node.name
2098

    
2099
    # check connectivity
2100
    result = self.rpc.call_version([node])[node]
2101
    result.Raise()
2102
    if result.data:
2103
      if constants.PROTOCOL_VERSION == result.data:
2104
        logging.info("Communication to node %s fine, sw version %s match",
2105
                     node, result.data)
2106
      else:
2107
        raise errors.OpExecError("Version mismatch master version %s,"
2108
                                 " node version %s" %
2109
                                 (constants.PROTOCOL_VERSION, result.data))
2110
    else:
2111
      raise errors.OpExecError("Cannot get version from the new node")
2112

    
2113
    # setup ssh on node
2114
    logging.info("Copy ssh key to node %s", node)
2115
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2116
    keyarray = []
2117
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2118
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2119
                priv_key, pub_key]
2120

    
2121
    for i in keyfiles:
2122
      f = open(i, 'r')
2123
      try:
2124
        keyarray.append(f.read())
2125
      finally:
2126
        f.close()
2127

    
2128
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2129
                                    keyarray[2],
2130
                                    keyarray[3], keyarray[4], keyarray[5])
2131

    
2132
    if result.failed or not result.data:
2133
      raise errors.OpExecError("Cannot transfer ssh keys to the new node")
2134

    
2135
    # Add node to our /etc/hosts, and add key to known_hosts
2136
    utils.AddHostToEtcHosts(new_node.name)
2137

    
2138
    if new_node.secondary_ip != new_node.primary_ip:
2139
      result = self.rpc.call_node_has_ip_address(new_node.name,
2140
                                                 new_node.secondary_ip)
2141
      if result.failed or not result.data:
2142
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2143
                                 " you gave (%s). Please fix and re-run this"
2144
                                 " command." % new_node.secondary_ip)
2145

    
2146
    node_verify_list = [self.cfg.GetMasterNode()]
2147
    node_verify_param = {
2148
      'nodelist': [node],
2149
      # TODO: do a node-net-test as well?
2150
    }
2151

    
2152
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2153
                                       self.cfg.GetClusterName())
2154
    for verifier in node_verify_list:
2155
      if result[verifier].failed or not result[verifier].data:
2156
        raise errors.OpExecError("Cannot communicate with %s's node daemon"
2157
                                 " for remote verification" % verifier)
2158
      if result[verifier].data['nodelist']:
2159
        for failed in result[verifier].data['nodelist']:
2160
          feedback_fn("ssh/hostname verification failed %s -> %s" %
2161
                      (verifier, result[verifier]['nodelist'][failed]))
2162
        raise errors.OpExecError("ssh/hostname verification failed.")
2163

    
2164
    # Distribute updated /etc/hosts and known_hosts to all nodes,
2165
    # including the node just added
2166
    myself = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
2167
    dist_nodes = self.cfg.GetNodeList()
2168
    if not self.op.readd:
2169
      dist_nodes.append(node)
2170
    if myself.name in dist_nodes:
2171
      dist_nodes.remove(myself.name)
2172

    
2173
    logging.debug("Copying hosts and known_hosts to all nodes")
2174
    for fname in (constants.ETC_HOSTS, constants.SSH_KNOWN_HOSTS_FILE):
2175
      result = self.rpc.call_upload_file(dist_nodes, fname)
2176
      for to_node, to_result in result.iteritems():
2177
        if to_result.failed or not to_result.data:
2178
          logging.error("Copy of file %s to node %s failed", fname, to_node)
2179

    
2180
    to_copy = []
2181
    enabled_hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
2182
    if constants.HTS_USE_VNC.intersection(enabled_hypervisors):
2183
      to_copy.append(constants.VNC_PASSWORD_FILE)
2184

    
2185
    for fname in to_copy:
2186
      result = self.rpc.call_upload_file([node], fname)
2187
      if result[node].failed or not result[node]:
2188
        logging.error("Could not copy file %s to node %s", fname, node)
2189

    
2190
    if self.op.readd:
2191
      self.context.ReaddNode(new_node)
2192
    else:
2193
      self.context.AddNode(new_node)
2194

    
2195

    
2196
class LUSetNodeParams(LogicalUnit):
2197
  """Modifies the parameters of a node.
2198

2199
  """
2200
  HPATH = "node-modify"
2201
  HTYPE = constants.HTYPE_NODE
2202
  _OP_REQP = ["node_name"]
2203
  REQ_BGL = False
2204

    
2205
  def CheckArguments(self):
2206
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2207
    if node_name is None:
2208
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2209
    self.op.node_name = node_name
2210
    _CheckBooleanOpField(self.op, 'master_candidate')
2211
    _CheckBooleanOpField(self.op, 'offline')
2212
    if self.op.master_candidate is None and self.op.offline is None:
2213
      raise errors.OpPrereqError("Please pass at least one modification")
2214
    if self.op.offline == True and self.op.master_candidate == True:
2215
      raise errors.OpPrereqError("Can't set the node into offline and"
2216
                                 " master_candidate at the same time")
2217

    
2218
  def ExpandNames(self):
2219
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2220

    
2221
  def BuildHooksEnv(self):
2222
    """Build hooks env.
2223

2224
    This runs on the master node.
2225

2226
    """
2227
    env = {
2228
      "OP_TARGET": self.op.node_name,
2229
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2230
      "OFFLINE": str(self.op.offline),
2231
      }
2232
    nl = [self.cfg.GetMasterNode(),
2233
          self.op.node_name]
2234
    return env, nl, nl
2235

    
2236
  def CheckPrereq(self):
2237
    """Check prerequisites.
2238

2239
    This only checks the instance list against the existing names.
2240

2241
    """
2242
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2243

    
2244
    if ((self.op.master_candidate == False or self.op.offline == True)
2245
        and node.master_candidate):
2246
      # we will demote the node from master_candidate
2247
      if self.op.node_name == self.cfg.GetMasterNode():
2248
        raise errors.OpPrereqError("The master node has to be a"
2249
                                   " master candidate and online")
2250
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2251
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2252
      if num_candidates <= cp_size:
2253
        msg = ("Not enough master candidates (desired"
2254
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2255
        if self.op.force:
2256
          self.LogWarning(msg)
2257
        else:
2258
          raise errors.OpPrereqError(msg)
2259

    
2260
    if (self.op.master_candidate == True and node.offline and
2261
        not self.op.offline == False):
2262
      raise errors.OpPrereqError("Can't set an offline node to"
2263
                                 " master_candidate")
2264

    
2265
    return
2266

    
2267
  def Exec(self, feedback_fn):
2268
    """Modifies a node.
2269

2270
    """
2271
    node = self.node
2272

    
2273
    result = []
2274

    
2275
    if self.op.offline is not None:
2276
      node.offline = self.op.offline
2277
      result.append(("offline", str(self.op.offline)))
2278
      if self.op.offline == True and node.master_candidate:
2279
        node.master_candidate = False
2280
        result.append(("master_candidate", "auto-demotion due to offline"))
2281

    
2282
    if self.op.master_candidate is not None:
2283
      node.master_candidate = self.op.master_candidate
2284
      result.append(("master_candidate", str(self.op.master_candidate)))
2285
      if self.op.master_candidate == False:
2286
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2287
        if (rrc.failed or not isinstance(rrc.data, (tuple, list))
2288
            or len(rrc.data) != 2):
2289
          self.LogWarning("Node rpc error: %s" % rrc.error)
2290
        elif not rrc.data[0]:
2291
          self.LogWarning("Node failed to demote itself: %s" % rrc.data[1])
2292

    
2293
    # this will trigger configuration file update, if needed
2294
    self.cfg.Update(node)
2295
    # this will trigger job queue propagation or cleanup
2296
    if self.op.node_name != self.cfg.GetMasterNode():
2297
      self.context.ReaddNode(node)
2298

    
2299
    return result
2300

    
2301

    
2302
class LUQueryClusterInfo(NoHooksLU):
2303
  """Query cluster configuration.
2304

2305
  """
2306
  _OP_REQP = []
2307
  REQ_BGL = False
2308

    
2309
  def ExpandNames(self):
2310
    self.needed_locks = {}
2311

    
2312
  def CheckPrereq(self):
2313
    """No prerequsites needed for this LU.
2314

2315
    """
2316
    pass
2317

    
2318
  def Exec(self, feedback_fn):
2319
    """Return cluster config.
2320

2321
    """
2322
    cluster = self.cfg.GetClusterInfo()
2323
    result = {
2324
      "software_version": constants.RELEASE_VERSION,
2325
      "protocol_version": constants.PROTOCOL_VERSION,
2326
      "config_version": constants.CONFIG_VERSION,
2327
      "os_api_version": constants.OS_API_VERSION,
2328
      "export_version": constants.EXPORT_VERSION,
2329
      "architecture": (platform.architecture()[0], platform.machine()),
2330
      "name": cluster.cluster_name,
2331
      "master": cluster.master_node,
2332
      "default_hypervisor": cluster.default_hypervisor,
2333
      "enabled_hypervisors": cluster.enabled_hypervisors,
2334
      "hvparams": dict([(hypervisor, cluster.hvparams[hypervisor])
2335
                        for hypervisor in cluster.enabled_hypervisors]),
2336
      "beparams": cluster.beparams,
2337
      "candidate_pool_size": cluster.candidate_pool_size,
2338
      }
2339

    
2340
    return result
2341

    
2342

    
2343
class LUQueryConfigValues(NoHooksLU):
2344
  """Return configuration values.
2345

2346
  """
2347
  _OP_REQP = []
2348
  REQ_BGL = False
2349
  _FIELDS_DYNAMIC = utils.FieldSet()
2350
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2351

    
2352
  def ExpandNames(self):
2353
    self.needed_locks = {}
2354

    
2355
    _CheckOutputFields(static=self._FIELDS_STATIC,
2356
                       dynamic=self._FIELDS_DYNAMIC,
2357
                       selected=self.op.output_fields)
2358

    
2359
  def CheckPrereq(self):
2360
    """No prerequisites.
2361

2362
    """
2363
    pass
2364

    
2365
  def Exec(self, feedback_fn):
2366
    """Dump a representation of the cluster config to the standard output.
2367

2368
    """
2369
    values = []
2370
    for field in self.op.output_fields:
2371
      if field == "cluster_name":
2372
        entry = self.cfg.GetClusterName()
2373
      elif field == "master_node":
2374
        entry = self.cfg.GetMasterNode()
2375
      elif field == "drain_flag":
2376
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2377
      else:
2378
        raise errors.ParameterError(field)
2379
      values.append(entry)
2380
    return values
2381

    
2382

    
2383
class LUActivateInstanceDisks(NoHooksLU):
2384
  """Bring up an instance's disks.
2385

2386
  """
2387
  _OP_REQP = ["instance_name"]
2388
  REQ_BGL = False
2389

    
2390
  def ExpandNames(self):
2391
    self._ExpandAndLockInstance()
2392
    self.needed_locks[locking.LEVEL_NODE] = []
2393
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2394

    
2395
  def DeclareLocks(self, level):
2396
    if level == locking.LEVEL_NODE:
2397
      self._LockInstancesNodes()
2398

    
2399
  def CheckPrereq(self):
2400
    """Check prerequisites.
2401

2402
    This checks that the instance is in the cluster.
2403

2404
    """
2405
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2406
    assert self.instance is not None, \
2407
      "Cannot retrieve locked instance %s" % self.op.instance_name
2408
    _CheckNodeOnline(self, self.instance.primary_node)
2409

    
2410
  def Exec(self, feedback_fn):
2411
    """Activate the disks.
2412

2413
    """
2414
    disks_ok, disks_info = _AssembleInstanceDisks(self, self.instance)
2415
    if not disks_ok:
2416
      raise errors.OpExecError("Cannot activate block devices")
2417

    
2418
    return disks_info
2419

    
2420

    
2421
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False):
2422
  """Prepare the block devices for an instance.
2423

2424
  This sets up the block devices on all nodes.
2425

2426
  @type lu: L{LogicalUnit}
2427
  @param lu: the logical unit on whose behalf we execute
2428
  @type instance: L{objects.Instance}
2429
  @param instance: the instance for whose disks we assemble
2430
  @type ignore_secondaries: boolean
2431
  @param ignore_secondaries: if true, errors on secondary nodes
2432
      won't result in an error return from the function
2433
  @return: False if the operation failed, otherwise a list of
2434
      (host, instance_visible_name, node_visible_name)
2435
      with the mapping from node devices to instance devices
2436

2437
  """
2438
  device_info = []
2439
  disks_ok = True
2440
  iname = instance.name
2441
  # With the two passes mechanism we try to reduce the window of
2442
  # opportunity for the race condition of switching DRBD to primary
2443
  # before handshaking occured, but we do not eliminate it
2444

    
2445
  # The proper fix would be to wait (with some limits) until the
2446
  # connection has been made and drbd transitions from WFConnection
2447
  # into any other network-connected state (Connected, SyncTarget,
2448
  # SyncSource, etc.)
2449

    
2450
  # 1st pass, assemble on all nodes in secondary mode
2451
  for inst_disk in instance.disks:
2452
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2453
      lu.cfg.SetDiskID(node_disk, node)
2454
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2455
      if result.failed or not result:
2456
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2457
                           " (is_primary=False, pass=1)",
2458
                           inst_disk.iv_name, node)
2459
        if not ignore_secondaries:
2460
          disks_ok = False
2461

    
2462
  # FIXME: race condition on drbd migration to primary
2463

    
2464
  # 2nd pass, do only the primary node
2465
  for inst_disk in instance.disks:
2466
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2467
      if node != instance.primary_node:
2468
        continue
2469
      lu.cfg.SetDiskID(node_disk, node)
2470
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2471
      if result.failed or not result:
2472
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2473
                           " (is_primary=True, pass=2)",
2474
                           inst_disk.iv_name, node)
2475
        disks_ok = False
2476
    device_info.append((instance.primary_node, inst_disk.iv_name, result.data))
2477

    
2478
  # leave the disks configured for the primary node
2479
  # this is a workaround that would be fixed better by
2480
  # improving the logical/physical id handling
2481
  for disk in instance.disks:
2482
    lu.cfg.SetDiskID(disk, instance.primary_node)
2483

    
2484
  return disks_ok, device_info
2485

    
2486

    
2487
def _StartInstanceDisks(lu, instance, force):
2488
  """Start the disks of an instance.
2489

2490
  """
2491
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2492
                                           ignore_secondaries=force)
2493
  if not disks_ok:
2494
    _ShutdownInstanceDisks(lu, instance)
2495
    if force is not None and not force:
2496
      lu.proc.LogWarning("", hint="If the message above refers to a"
2497
                         " secondary node,"
2498
                         " you can retry the operation using '--force'.")
2499
    raise errors.OpExecError("Disk consistency error")
2500

    
2501

    
2502
class LUDeactivateInstanceDisks(NoHooksLU):
2503
  """Shutdown an instance's disks.
2504

2505
  """
2506
  _OP_REQP = ["instance_name"]
2507
  REQ_BGL = False
2508

    
2509
  def ExpandNames(self):
2510
    self._ExpandAndLockInstance()
2511
    self.needed_locks[locking.LEVEL_NODE] = []
2512
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2513

    
2514
  def DeclareLocks(self, level):
2515
    if level == locking.LEVEL_NODE:
2516
      self._LockInstancesNodes()
2517

    
2518
  def CheckPrereq(self):
2519
    """Check prerequisites.
2520

2521
    This checks that the instance is in the cluster.
2522

2523
    """
2524
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2525
    assert self.instance is not None, \
2526
      "Cannot retrieve locked instance %s" % self.op.instance_name
2527

    
2528
  def Exec(self, feedback_fn):
2529
    """Deactivate the disks
2530

2531
    """
2532
    instance = self.instance
2533
    _SafeShutdownInstanceDisks(self, instance)
2534

    
2535

    
2536
def _SafeShutdownInstanceDisks(lu, instance):
2537
  """Shutdown block devices of an instance.
2538

2539
  This function checks if an instance is running, before calling
2540
  _ShutdownInstanceDisks.
2541

2542
  """
2543
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2544
                                      [instance.hypervisor])
2545
  ins_l = ins_l[instance.primary_node]
2546
  if ins_l.failed or not isinstance(ins_l.data, list):
2547
    raise errors.OpExecError("Can't contact node '%s'" %
2548
                             instance.primary_node)
2549

    
2550
  if instance.name in ins_l.data:
2551
    raise errors.OpExecError("Instance is running, can't shutdown"
2552
                             " block devices.")
2553

    
2554
  _ShutdownInstanceDisks(lu, instance)
2555

    
2556

    
2557
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2558
  """Shutdown block devices of an instance.
2559

2560
  This does the shutdown on all nodes of the instance.
2561

2562
  If the ignore_primary is false, errors on the primary node are
2563
  ignored.
2564

2565
  """
2566
  result = True
2567
  for disk in instance.disks:
2568
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2569
      lu.cfg.SetDiskID(top_disk, node)
2570
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2571
      if result.failed or not result.data:
2572
        logging.error("Could not shutdown block device %s on node %s",
2573
                      disk.iv_name, node)
2574
        if not ignore_primary or node != instance.primary_node:
2575
          result = False
2576
  return result
2577

    
2578

    
2579
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2580
  """Checks if a node has enough free memory.
2581

2582
  This function check if a given node has the needed amount of free
2583
  memory. In case the node has less memory or we cannot get the
2584
  information from the node, this function raise an OpPrereqError
2585
  exception.
2586

2587
  @type lu: C{LogicalUnit}
2588
  @param lu: a logical unit from which we get configuration data
2589
  @type node: C{str}
2590
  @param node: the node to check
2591
  @type reason: C{str}
2592
  @param reason: string to use in the error message
2593
  @type requested: C{int}
2594
  @param requested: the amount of memory in MiB to check for
2595
  @type hypervisor_name: C{str}
2596
  @param hypervisor_name: the hypervisor to ask for memory stats
2597
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2598
      we cannot check the node
2599

2600
  """
2601
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2602
  nodeinfo[node].Raise()
2603
  free_mem = nodeinfo[node].data.get('memory_free')
2604
  if not isinstance(free_mem, int):
2605
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2606
                             " was '%s'" % (node, free_mem))
2607
  if requested > free_mem:
2608
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2609
                             " needed %s MiB, available %s MiB" %
2610
                             (node, reason, requested, free_mem))
2611

    
2612

    
2613
class LUStartupInstance(LogicalUnit):
2614
  """Starts an instance.
2615

2616
  """
2617
  HPATH = "instance-start"
2618
  HTYPE = constants.HTYPE_INSTANCE
2619
  _OP_REQP = ["instance_name", "force"]
2620
  REQ_BGL = False
2621

    
2622
  def ExpandNames(self):
2623
    self._ExpandAndLockInstance()
2624

    
2625
  def BuildHooksEnv(self):
2626
    """Build hooks env.
2627

2628
    This runs on master, primary and secondary nodes of the instance.
2629

2630
    """
2631
    env = {
2632
      "FORCE": self.op.force,
2633
      }
2634
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2635
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2636
    return env, nl, nl
2637

    
2638
  def CheckPrereq(self):
2639
    """Check prerequisites.
2640

2641
    This checks that the instance is in the cluster.
2642

2643
    """
2644
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2645
    assert self.instance is not None, \
2646
      "Cannot retrieve locked instance %s" % self.op.instance_name
2647

    
2648
    _CheckNodeOnline(self, instance.primary_node)
2649

    
2650
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2651
    # check bridges existance
2652
    _CheckInstanceBridgesExist(self, instance)
2653

    
2654
    _CheckNodeFreeMemory(self, instance.primary_node,
2655
                         "starting instance %s" % instance.name,
2656
                         bep[constants.BE_MEMORY], instance.hypervisor)
2657

    
2658
  def Exec(self, feedback_fn):
2659
    """Start the instance.
2660

2661
    """
2662
    instance = self.instance
2663
    force = self.op.force
2664
    extra_args = getattr(self.op, "extra_args", "")
2665

    
2666
    self.cfg.MarkInstanceUp(instance.name)
2667

    
2668
    node_current = instance.primary_node
2669

    
2670
    _StartInstanceDisks(self, instance, force)
2671

    
2672
    result = self.rpc.call_instance_start(node_current, instance, extra_args)
2673
    msg = result.RemoteFailMsg()
2674
    if msg:
2675
      _ShutdownInstanceDisks(self, instance)
2676
      raise errors.OpExecError("Could not start instance: %s" % msg)
2677

    
2678

    
2679
class LURebootInstance(LogicalUnit):
2680
  """Reboot an instance.
2681

2682
  """
2683
  HPATH = "instance-reboot"
2684
  HTYPE = constants.HTYPE_INSTANCE
2685
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2686
  REQ_BGL = False
2687

    
2688
  def ExpandNames(self):
2689
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2690
                                   constants.INSTANCE_REBOOT_HARD,
2691
                                   constants.INSTANCE_REBOOT_FULL]:
2692
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2693
                                  (constants.INSTANCE_REBOOT_SOFT,
2694
                                   constants.INSTANCE_REBOOT_HARD,
2695
                                   constants.INSTANCE_REBOOT_FULL))
2696
    self._ExpandAndLockInstance()
2697

    
2698
  def BuildHooksEnv(self):
2699
    """Build hooks env.
2700

2701
    This runs on master, primary and secondary nodes of the instance.
2702

2703
    """
2704
    env = {
2705
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2706
      }
2707
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2708
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2709
    return env, nl, nl
2710

    
2711
  def CheckPrereq(self):
2712
    """Check prerequisites.
2713

2714
    This checks that the instance is in the cluster.
2715

2716
    """
2717
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2718
    assert self.instance is not None, \
2719
      "Cannot retrieve locked instance %s" % self.op.instance_name
2720

    
2721
    _CheckNodeOnline(self, instance.primary_node)
2722

    
2723
    # check bridges existance
2724
    _CheckInstanceBridgesExist(self, instance)
2725

    
2726
  def Exec(self, feedback_fn):
2727
    """Reboot the instance.
2728

2729
    """
2730
    instance = self.instance
2731
    ignore_secondaries = self.op.ignore_secondaries
2732
    reboot_type = self.op.reboot_type
2733
    extra_args = getattr(self.op, "extra_args", "")
2734

    
2735
    node_current = instance.primary_node
2736

    
2737
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
2738
                       constants.INSTANCE_REBOOT_HARD]:
2739
      result = self.rpc.call_instance_reboot(node_current, instance,
2740
                                             reboot_type, extra_args)
2741
      if result.failed or not result.data:
2742
        raise errors.OpExecError("Could not reboot instance")
2743
    else:
2744
      if not self.rpc.call_instance_shutdown(node_current, instance):
2745
        raise errors.OpExecError("could not shutdown instance for full reboot")
2746
      _ShutdownInstanceDisks(self, instance)
2747
      _StartInstanceDisks(self, instance, ignore_secondaries)
2748
      result = self.rpc.call_instance_start(node_current, instance, extra_args)
2749
      msg = result.RemoteFailMsg()
2750
      if msg:
2751
        _ShutdownInstanceDisks(self, instance)
2752
        raise errors.OpExecError("Could not start instance for"
2753
                                 " full reboot: %s" % msg)
2754

    
2755
    self.cfg.MarkInstanceUp(instance.name)
2756

    
2757

    
2758
class LUShutdownInstance(LogicalUnit):
2759
  """Shutdown an instance.
2760

2761
  """
2762
  HPATH = "instance-stop"
2763
  HTYPE = constants.HTYPE_INSTANCE
2764
  _OP_REQP = ["instance_name"]
2765
  REQ_BGL = False
2766

    
2767
  def ExpandNames(self):
2768
    self._ExpandAndLockInstance()
2769

    
2770
  def BuildHooksEnv(self):
2771
    """Build hooks env.
2772

2773
    This runs on master, primary and secondary nodes of the instance.
2774

2775
    """
2776
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2777
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2778
    return env, nl, nl
2779

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

2783
    This checks that the instance is in the cluster.
2784

2785
    """
2786
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2787
    assert self.instance is not None, \
2788
      "Cannot retrieve locked instance %s" % self.op.instance_name
2789
    _CheckNodeOnline(self, self.instance.primary_node)
2790

    
2791
  def Exec(self, feedback_fn):
2792
    """Shutdown the instance.
2793

2794
    """
2795
    instance = self.instance
2796
    node_current = instance.primary_node
2797
    self.cfg.MarkInstanceDown(instance.name)
2798
    result = self.rpc.call_instance_shutdown(node_current, instance)
2799
    if result.failed or not result.data:
2800
      self.proc.LogWarning("Could not shutdown instance")
2801

    
2802
    _ShutdownInstanceDisks(self, instance)
2803

    
2804

    
2805
class LUReinstallInstance(LogicalUnit):
2806
  """Reinstall an instance.
2807

2808
  """
2809
  HPATH = "instance-reinstall"
2810
  HTYPE = constants.HTYPE_INSTANCE
2811
  _OP_REQP = ["instance_name"]
2812
  REQ_BGL = False
2813

    
2814
  def ExpandNames(self):
2815
    self._ExpandAndLockInstance()
2816

    
2817
  def BuildHooksEnv(self):
2818
    """Build hooks env.
2819

2820
    This runs on master, primary and secondary nodes of the instance.
2821

2822
    """
2823
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2824
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2825
    return env, nl, nl
2826

    
2827
  def CheckPrereq(self):
2828
    """Check prerequisites.
2829

2830
    This checks that the instance is in the cluster and is not running.
2831

2832
    """
2833
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2834
    assert instance is not None, \
2835
      "Cannot retrieve locked instance %s" % self.op.instance_name
2836
    _CheckNodeOnline(self, instance.primary_node)
2837

    
2838
    if instance.disk_template == constants.DT_DISKLESS:
2839
      raise errors.OpPrereqError("Instance '%s' has no disks" %
2840
                                 self.op.instance_name)
2841
    if instance.admin_up:
2842
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
2843
                                 self.op.instance_name)
2844
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2845
                                              instance.name,
2846
                                              instance.hypervisor)
2847
    if remote_info.failed or remote_info.data:
2848
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
2849
                                 (self.op.instance_name,
2850
                                  instance.primary_node))
2851

    
2852
    self.op.os_type = getattr(self.op, "os_type", None)
2853
    if self.op.os_type is not None:
2854
      # OS verification
2855
      pnode = self.cfg.GetNodeInfo(
2856
        self.cfg.ExpandNodeName(instance.primary_node))
2857
      if pnode is None:
2858
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
2859
                                   self.op.pnode)
2860
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
2861
      result.Raise()
2862
      if not isinstance(result.data, objects.OS):
2863
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
2864
                                   " primary node"  % self.op.os_type)
2865

    
2866
    self.instance = instance
2867

    
2868
  def Exec(self, feedback_fn):
2869
    """Reinstall the instance.
2870

2871
    """
2872
    inst = self.instance
2873

    
2874
    if self.op.os_type is not None:
2875
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
2876
      inst.os = self.op.os_type
2877
      self.cfg.Update(inst)
2878

    
2879
    _StartInstanceDisks(self, inst, None)
2880
    try:
2881
      feedback_fn("Running the instance OS create scripts...")
2882
      result = self.rpc.call_instance_os_add(inst.primary_node, inst)
2883
      msg = result.RemoteFailMsg()
2884
      if msg:
2885
        raise errors.OpExecError("Could not install OS for instance %s"
2886
                                 " on node %s: %s" %
2887
                                 (inst.name, inst.primary_node, msg))
2888
    finally:
2889
      _ShutdownInstanceDisks(self, inst)
2890

    
2891

    
2892
class LURenameInstance(LogicalUnit):
2893
  """Rename an instance.
2894

2895
  """
2896
  HPATH = "instance-rename"
2897
  HTYPE = constants.HTYPE_INSTANCE
2898
  _OP_REQP = ["instance_name", "new_name"]
2899

    
2900
  def BuildHooksEnv(self):
2901
    """Build hooks env.
2902

2903
    This runs on master, primary and secondary nodes of the instance.
2904

2905
    """
2906
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2907
    env["INSTANCE_NEW_NAME"] = self.op.new_name
2908
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2909
    return env, nl, nl
2910

    
2911
  def CheckPrereq(self):
2912
    """Check prerequisites.
2913

2914
    This checks that the instance is in the cluster and is not running.
2915

2916
    """
2917
    instance = self.cfg.GetInstanceInfo(
2918
      self.cfg.ExpandInstanceName(self.op.instance_name))
2919
    if instance is None:
2920
      raise errors.OpPrereqError("Instance '%s' not known" %
2921
                                 self.op.instance_name)
2922
    _CheckNodeOnline(self, instance.primary_node)
2923

    
2924
    if instance.admin_up:
2925
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
2926
                                 self.op.instance_name)
2927
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2928
                                              instance.name,
2929
                                              instance.hypervisor)
2930
    remote_info.Raise()
2931
    if remote_info.data:
2932
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
2933
                                 (self.op.instance_name,
2934
                                  instance.primary_node))
2935
    self.instance = instance
2936

    
2937
    # new name verification
2938
    name_info = utils.HostInfo(self.op.new_name)
2939

    
2940
    self.op.new_name = new_name = name_info.name
2941
    instance_list = self.cfg.GetInstanceList()
2942
    if new_name in instance_list:
2943
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
2944
                                 new_name)
2945

    
2946
    if not getattr(self.op, "ignore_ip", False):
2947
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
2948
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
2949
                                   (name_info.ip, new_name))
2950

    
2951

    
2952
  def Exec(self, feedback_fn):
2953
    """Reinstall the instance.
2954

2955
    """
2956
    inst = self.instance
2957
    old_name = inst.name
2958

    
2959
    if inst.disk_template == constants.DT_FILE:
2960
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
2961

    
2962
    self.cfg.RenameInstance(inst.name, self.op.new_name)
2963
    # Change the instance lock. This is definitely safe while we hold the BGL
2964
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
2965
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
2966

    
2967
    # re-read the instance from the configuration after rename
2968
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
2969

    
2970
    if inst.disk_template == constants.DT_FILE:
2971
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
2972
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
2973
                                                     old_file_storage_dir,
2974
                                                     new_file_storage_dir)
2975
      result.Raise()
2976
      if not result.data:
2977
        raise errors.OpExecError("Could not connect to node '%s' to rename"
2978
                                 " directory '%s' to '%s' (but the instance"
2979
                                 " has been renamed in Ganeti)" % (
2980
                                 inst.primary_node, old_file_storage_dir,
2981
                                 new_file_storage_dir))
2982

    
2983
      if not result.data[0]:
2984
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
2985
                                 " (but the instance has been renamed in"
2986
                                 " Ganeti)" % (old_file_storage_dir,
2987
                                               new_file_storage_dir))
2988

    
2989
    _StartInstanceDisks(self, inst, None)
2990
    try:
2991
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
2992
                                                 old_name)
2993
      msg = result.RemoteFailMsg()
2994
      if msg:
2995
        msg = ("Could not run OS rename script for instance %s on node %s"
2996
               " (but the instance has been renamed in Ganeti): %s" %
2997
               (inst.name, inst.primary_node, msg))
2998
        self.proc.LogWarning(msg)
2999
    finally:
3000
      _ShutdownInstanceDisks(self, inst)
3001

    
3002

    
3003
class LURemoveInstance(LogicalUnit):
3004
  """Remove an instance.
3005

3006
  """
3007
  HPATH = "instance-remove"
3008
  HTYPE = constants.HTYPE_INSTANCE
3009
  _OP_REQP = ["instance_name", "ignore_failures"]
3010
  REQ_BGL = False
3011

    
3012
  def ExpandNames(self):
3013
    self._ExpandAndLockInstance()
3014
    self.needed_locks[locking.LEVEL_NODE] = []
3015
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3016

    
3017
  def DeclareLocks(self, level):
3018
    if level == locking.LEVEL_NODE:
3019
      self._LockInstancesNodes()
3020

    
3021
  def BuildHooksEnv(self):
3022
    """Build hooks env.
3023

3024
    This runs on master, primary and secondary nodes of the instance.
3025

3026
    """
3027
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3028
    nl = [self.cfg.GetMasterNode()]
3029
    return env, nl, nl
3030

    
3031
  def CheckPrereq(self):
3032
    """Check prerequisites.
3033

3034
    This checks that the instance is in the cluster.
3035

3036
    """
3037
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3038
    assert self.instance is not None, \
3039
      "Cannot retrieve locked instance %s" % self.op.instance_name
3040

    
3041
  def Exec(self, feedback_fn):
3042
    """Remove the instance.
3043

3044
    """
3045
    instance = self.instance
3046
    logging.info("Shutting down instance %s on node %s",
3047
                 instance.name, instance.primary_node)
3048

    
3049
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3050
    if result.failed or not result.data:
3051
      if self.op.ignore_failures:
3052
        feedback_fn("Warning: can't shutdown instance")
3053
      else:
3054
        raise errors.OpExecError("Could not shutdown instance %s on node %s" %
3055
                                 (instance.name, instance.primary_node))
3056

    
3057
    logging.info("Removing block devices for instance %s", instance.name)
3058

    
3059
    if not _RemoveDisks(self, instance):
3060
      if self.op.ignore_failures:
3061
        feedback_fn("Warning: can't remove instance's disks")
3062
      else:
3063
        raise errors.OpExecError("Can't remove instance's disks")
3064

    
3065
    logging.info("Removing instance %s out of cluster config", instance.name)
3066

    
3067
    self.cfg.RemoveInstance(instance.name)
3068
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3069

    
3070

    
3071
class LUQueryInstances(NoHooksLU):
3072
  """Logical unit for querying instances.
3073

3074
  """
3075
  _OP_REQP = ["output_fields", "names"]
3076
  REQ_BGL = False
3077
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3078
                                    "admin_state", "admin_ram",
3079
                                    "disk_template", "ip", "mac", "bridge",
3080
                                    "sda_size", "sdb_size", "vcpus", "tags",
3081
                                    "network_port", "beparams",
3082
                                    "(disk).(size)/([0-9]+)",
3083
                                    "(disk).(sizes)",
3084
                                    "(nic).(mac|ip|bridge)/([0-9]+)",
3085
                                    "(nic).(macs|ips|bridges)",
3086
                                    "(disk|nic).(count)",
3087
                                    "serial_no", "hypervisor", "hvparams",] +
3088
                                  ["hv/%s" % name
3089
                                   for name in constants.HVS_PARAMETERS] +
3090
                                  ["be/%s" % name
3091
                                   for name in constants.BES_PARAMETERS])
3092
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3093

    
3094

    
3095
  def ExpandNames(self):
3096
    _CheckOutputFields(static=self._FIELDS_STATIC,
3097
                       dynamic=self._FIELDS_DYNAMIC,
3098
                       selected=self.op.output_fields)
3099

    
3100
    self.needed_locks = {}
3101
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3102
    self.share_locks[locking.LEVEL_NODE] = 1
3103

    
3104
    if self.op.names:
3105
      self.wanted = _GetWantedInstances(self, self.op.names)
3106
    else:
3107
      self.wanted = locking.ALL_SET
3108

    
3109
    self.do_locking = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3110
    if self.do_locking:
3111
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3112
      self.needed_locks[locking.LEVEL_NODE] = []
3113
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3114

    
3115
  def DeclareLocks(self, level):
3116
    if level == locking.LEVEL_NODE and self.do_locking:
3117
      self._LockInstancesNodes()
3118

    
3119
  def CheckPrereq(self):
3120
    """Check prerequisites.
3121

3122
    """
3123
    pass
3124

    
3125
  def Exec(self, feedback_fn):
3126
    """Computes the list of nodes and their attributes.
3127

3128
    """
3129
    all_info = self.cfg.GetAllInstancesInfo()
3130
    if self.wanted == locking.ALL_SET:
3131
      # caller didn't specify instance names, so ordering is not important
3132
      if self.do_locking:
3133
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3134
      else:
3135
        instance_names = all_info.keys()
3136
      instance_names = utils.NiceSort(instance_names)
3137
    else:
3138
      # caller did specify names, so we must keep the ordering
3139
      if self.do_locking:
3140
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3141
      else:
3142
        tgt_set = all_info.keys()
3143
      missing = set(self.wanted).difference(tgt_set)
3144
      if missing:
3145
        raise errors.OpExecError("Some instances were removed before"
3146
                                 " retrieving their data: %s" % missing)
3147
      instance_names = self.wanted
3148

    
3149
    instance_list = [all_info[iname] for iname in instance_names]
3150

    
3151
    # begin data gathering
3152

    
3153
    nodes = frozenset([inst.primary_node for inst in instance_list])
3154
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3155

    
3156
    bad_nodes = []
3157
    off_nodes = []
3158
    if self.do_locking:
3159
      live_data = {}
3160
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3161
      for name in nodes:
3162
        result = node_data[name]
3163
        if result.offline:
3164
          # offline nodes will be in both lists
3165
          off_nodes.append(name)
3166
        if result.failed:
3167
          bad_nodes.append(name)
3168
        else:
3169
          if result.data:
3170
            live_data.update(result.data)
3171
            # else no instance is alive
3172
    else:
3173
      live_data = dict([(name, {}) for name in instance_names])
3174

    
3175
    # end data gathering
3176

    
3177
    HVPREFIX = "hv/"
3178
    BEPREFIX = "be/"
3179
    output = []
3180
    for instance in instance_list:
3181
      iout = []
3182
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3183
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3184
      for field in self.op.output_fields:
3185
        st_match = self._FIELDS_STATIC.Matches(field)
3186
        if field == "name":
3187
          val = instance.name
3188
        elif field == "os":
3189
          val = instance.os
3190
        elif field == "pnode":
3191
          val = instance.primary_node
3192
        elif field == "snodes":
3193
          val = list(instance.secondary_nodes)
3194
        elif field == "admin_state":
3195
          val = instance.admin_up
3196
        elif field == "oper_state":
3197
          if instance.primary_node in bad_nodes:
3198
            val = None
3199
          else:
3200
            val = bool(live_data.get(instance.name))
3201
        elif field == "status":
3202
          if instance.primary_node in off_nodes:
3203
            val = "ERROR_nodeoffline"
3204
          elif instance.primary_node in bad_nodes:
3205
            val = "ERROR_nodedown"
3206
          else:
3207
            running = bool(live_data.get(instance.name))
3208
            if running:
3209
              if instance.admin_up:
3210
                val = "running"
3211
              else:
3212
                val = "ERROR_up"
3213
            else:
3214
              if instance.admin_up:
3215
                val = "ERROR_down"
3216
              else:
3217
                val = "ADMIN_down"
3218
        elif field == "oper_ram":
3219
          if instance.primary_node in bad_nodes:
3220
            val = None
3221
          elif instance.name in live_data:
3222
            val = live_data[instance.name].get("memory", "?")
3223
          else:
3224
            val = "-"
3225
        elif field == "disk_template":
3226
          val = instance.disk_template
3227
        elif field == "ip":
3228
          val = instance.nics[0].ip
3229
        elif field == "bridge":
3230
          val = instance.nics[0].bridge
3231
        elif field == "mac":
3232
          val = instance.nics[0].mac
3233
        elif field == "sda_size" or field == "sdb_size":
3234
          idx = ord(field[2]) - ord('a')
3235
          try:
3236
            val = instance.FindDisk(idx).size
3237
          except errors.OpPrereqError:
3238
            val = None
3239
        elif field == "tags":
3240
          val = list(instance.GetTags())
3241
        elif field == "serial_no":
3242
          val = instance.serial_no
3243
        elif field == "network_port":
3244
          val = instance.network_port
3245
        elif field == "hypervisor":
3246
          val = instance.hypervisor
3247
        elif field == "hvparams":
3248
          val = i_hv
3249
        elif (field.startswith(HVPREFIX) and
3250
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3251
          val = i_hv.get(field[len(HVPREFIX):], None)
3252
        elif field == "beparams":
3253
          val = i_be
3254
        elif (field.startswith(BEPREFIX) and
3255
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3256
          val = i_be.get(field[len(BEPREFIX):], None)
3257
        elif st_match and st_match.groups():
3258
          # matches a variable list
3259
          st_groups = st_match.groups()
3260
          if st_groups and st_groups[0] == "disk":
3261
            if st_groups[1] == "count":
3262
              val = len(instance.disks)
3263
            elif st_groups[1] == "sizes":
3264
              val = [disk.size for disk in instance.disks]
3265
            elif st_groups[1] == "size":
3266
              try:
3267
                val = instance.FindDisk(st_groups[2]).size
3268
              except errors.OpPrereqError:
3269
                val = None
3270
            else:
3271
              assert False, "Unhandled disk parameter"
3272
          elif st_groups[0] == "nic":
3273
            if st_groups[1] == "count":
3274
              val = len(instance.nics)
3275
            elif st_groups[1] == "macs":
3276
              val = [nic.mac for nic in instance.nics]
3277
            elif st_groups[1] == "ips":
3278
              val = [nic.ip for nic in instance.nics]
3279
            elif st_groups[1] == "bridges":
3280
              val = [nic.bridge for nic in instance.nics]
3281
            else:
3282
              # index-based item
3283
              nic_idx = int(st_groups[2])
3284
              if nic_idx >= len(instance.nics):
3285
                val = None
3286
              else:
3287
                if st_groups[1] == "mac":
3288
                  val = instance.nics[nic_idx].mac
3289
                elif st_groups[1] == "ip":
3290
                  val = instance.nics[nic_idx].ip
3291
                elif st_groups[1] == "bridge":
3292
                  val = instance.nics[nic_idx].bridge
3293
                else:
3294
                  assert False, "Unhandled NIC parameter"
3295
          else:
3296
            assert False, "Unhandled variable parameter"
3297
        else:
3298
          raise errors.ParameterError(field)
3299
        iout.append(val)
3300
      output.append(iout)
3301

    
3302
    return output
3303

    
3304

    
3305
class LUFailoverInstance(LogicalUnit):
3306
  """Failover an instance.
3307

3308
  """
3309
  HPATH = "instance-failover"
3310
  HTYPE = constants.HTYPE_INSTANCE
3311
  _OP_REQP = ["instance_name", "ignore_consistency"]
3312
  REQ_BGL = False
3313

    
3314
  def ExpandNames(self):
3315
    self._ExpandAndLockInstance()
3316
    self.needed_locks[locking.LEVEL_NODE] = []
3317
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3318

    
3319
  def DeclareLocks(self, level):
3320
    if level == locking.LEVEL_NODE:
3321
      self._LockInstancesNodes()
3322

    
3323
  def BuildHooksEnv(self):
3324
    """Build hooks env.
3325

3326
    This runs on master, primary and secondary nodes of the instance.
3327

3328
    """
3329
    env = {
3330
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3331
      }
3332
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3333
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3334
    return env, nl, nl
3335

    
3336
  def CheckPrereq(self):
3337
    """Check prerequisites.
3338

3339
    This checks that the instance is in the cluster.
3340

3341
    """
3342
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3343
    assert self.instance is not None, \
3344
      "Cannot retrieve locked instance %s" % self.op.instance_name
3345

    
3346
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3347
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3348
      raise errors.OpPrereqError("Instance's disk layout is not"
3349
                                 " network mirrored, cannot failover.")
3350

    
3351
    secondary_nodes = instance.secondary_nodes
3352
    if not secondary_nodes:
3353
      raise errors.ProgrammerError("no secondary node but using "
3354
                                   "a mirrored disk template")
3355

    
3356
    target_node = secondary_nodes[0]
3357
    _CheckNodeOnline(self, target_node)
3358
    # check memory requirements on the secondary node
3359
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3360
                         instance.name, bep[constants.BE_MEMORY],
3361
                         instance.hypervisor)
3362

    
3363
    # check bridge existance
3364
    brlist = [nic.bridge for nic in instance.nics]
3365
    result = self.rpc.call_bridges_exist(target_node, brlist)
3366
    result.Raise()
3367
    if not result.data:
3368
      raise errors.OpPrereqError("One or more target bridges %s does not"
3369
                                 " exist on destination node '%s'" %
3370
                                 (brlist, target_node))
3371

    
3372
  def Exec(self, feedback_fn):
3373
    """Failover an instance.
3374

3375
    The failover is done by shutting it down on its present node and
3376
    starting it on the secondary.
3377

3378
    """
3379
    instance = self.instance
3380

    
3381
    source_node = instance.primary_node
3382
    target_node = instance.secondary_nodes[0]
3383

    
3384
    feedback_fn("* checking disk consistency between source and target")
3385
    for dev in instance.disks:
3386
      # for drbd, these are drbd over lvm
3387
      if not _CheckDiskConsistency(self, dev, target_node, False):
3388
        if instance.admin_up and not self.op.ignore_consistency:
3389
          raise errors.OpExecError("Disk %s is degraded on target node,"
3390
                                   " aborting failover." % dev.iv_name)
3391

    
3392
    feedback_fn("* shutting down instance on source node")
3393
    logging.info("Shutting down instance %s on node %s",
3394
                 instance.name, source_node)
3395

    
3396
    result = self.rpc.call_instance_shutdown(source_node, instance)
3397
    if result.failed or not result.data:
3398
      if self.op.ignore_consistency:
3399
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3400
                             " Proceeding"
3401
                             " anyway. Please make sure node %s is down",
3402
                             instance.name, source_node, source_node)
3403
      else:
3404
        raise errors.OpExecError("Could not shutdown instance %s on node %s" %
3405
                                 (instance.name, source_node))
3406

    
3407
    feedback_fn("* deactivating the instance's disks on source node")
3408
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3409
      raise errors.OpExecError("Can't shut down the instance's disks.")
3410

    
3411
    instance.primary_node = target_node
3412
    # distribute new instance config to the other nodes
3413
    self.cfg.Update(instance)
3414

    
3415
    # Only start the instance if it's marked as up
3416
    if instance.admin_up:
3417
      feedback_fn("* activating the instance's disks on target node")
3418
      logging.info("Starting instance %s on node %s",
3419
                   instance.name, target_node)
3420

    
3421
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3422
                                               ignore_secondaries=True)
3423
      if not disks_ok:
3424
        _ShutdownInstanceDisks(self, instance)
3425
        raise errors.OpExecError("Can't activate the instance's disks")
3426

    
3427
      feedback_fn("* starting the instance on the target node")
3428
      result = self.rpc.call_instance_start(target_node, instance, None)
3429
      msg = result.RemoteFailMsg()
3430
      if msg:
3431
        _ShutdownInstanceDisks(self, instance)
3432
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3433
                                 (instance.name, target_node, msg))
3434

    
3435

    
3436
class LUMigrateInstance(LogicalUnit):
3437
  """Migrate an instance.
3438

3439
  This is migration without shutting down, compared to the failover,
3440
  which is done with shutdown.
3441

3442
  """
3443
  HPATH = "instance-migrate"
3444
  HTYPE = constants.HTYPE_INSTANCE
3445
  _OP_REQP = ["instance_name", "live", "cleanup"]
3446

    
3447
  REQ_BGL = False
3448

    
3449
  def ExpandNames(self):
3450
    self._ExpandAndLockInstance()
3451
    self.needed_locks[locking.LEVEL_NODE] = []
3452
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3453

    
3454
  def DeclareLocks(self, level):
3455
    if level == locking.LEVEL_NODE:
3456
      self._LockInstancesNodes()
3457

    
3458
  def BuildHooksEnv(self):
3459
    """Build hooks env.
3460

3461
    This runs on master, primary and secondary nodes of the instance.
3462

3463
    """
3464
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3465
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3466
    return env, nl, nl
3467

    
3468
  def CheckPrereq(self):
3469
    """Check prerequisites.
3470

3471
    This checks that the instance is in the cluster.
3472

3473
    """
3474
    instance = self.cfg.GetInstanceInfo(
3475
      self.cfg.ExpandInstanceName(self.op.instance_name))
3476
    if instance is None:
3477
      raise errors.OpPrereqError("Instance '%s' not known" %
3478
                                 self.op.instance_name)
3479

    
3480
    if instance.disk_template != constants.DT_DRBD8:
3481
      raise errors.OpPrereqError("Instance's disk layout is not"
3482
                                 " drbd8, cannot migrate.")
3483

    
3484
    secondary_nodes = instance.secondary_nodes
3485
    if not secondary_nodes:
3486
      raise errors.ProgrammerError("no secondary node but using "
3487
                                   "drbd8 disk template")
3488

    
3489
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3490

    
3491
    target_node = secondary_nodes[0]
3492
    # check memory requirements on the secondary node
3493
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3494
                         instance.name, i_be[constants.BE_MEMORY],
3495
                         instance.hypervisor)
3496

    
3497
    # check bridge existance
3498
    brlist = [nic.bridge for nic in instance.nics]
3499
    result = self.rpc.call_bridges_exist(target_node, brlist)
3500
    if result.failed or not result.data:
3501
      raise errors.OpPrereqError("One or more target bridges %s does not"
3502
                                 " exist on destination node '%s'" %
3503
                                 (brlist, target_node))
3504

    
3505
    if not self.op.cleanup:
3506
      result = self.rpc.call_instance_migratable(instance.primary_node,
3507
                                                 instance)
3508
      msg = result.RemoteFailMsg()
3509
      if msg:
3510
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3511
                                   msg)
3512

    
3513
    self.instance = instance
3514

    
3515
  def _WaitUntilSync(self):
3516
    """Poll with custom rpc for disk sync.
3517

3518
    This uses our own step-based rpc call.
3519

3520
    """
3521
    self.feedback_fn("* wait until resync is done")
3522
    all_done = False
3523
    while not all_done:
3524
      all_done = True
3525
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3526
                                            self.nodes_ip,
3527
                                            self.instance.disks)
3528
      min_percent = 100
3529
      for node, nres in result.items():
3530
        msg = nres.RemoteFailMsg()
3531
        if msg:
3532
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3533
                                   (node, msg))
3534
        node_done, node_percent = nres.data[1]
3535
        all_done = all_done and node_done
3536
        if node_percent is not None:
3537
          min_percent = min(min_percent, node_percent)
3538
      if not all_done:
3539
        if min_percent < 100:
3540
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3541
        time.sleep(2)
3542

    
3543
  def _EnsureSecondary(self, node):
3544
    """Demote a node to secondary.
3545

3546
    """
3547
    self.feedback_fn("* switching node %s to secondary mode" % node)
3548

    
3549
    for dev in self.instance.disks:
3550
      self.cfg.SetDiskID(dev, node)
3551

    
3552
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3553
                                          self.instance.disks)
3554
    msg = result.RemoteFailMsg()
3555
    if msg:
3556
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3557
                               " error %s" % (node, msg))
3558

    
3559
  def _GoStandalone(self):
3560
    """Disconnect from the network.
3561

3562
    """
3563
    self.feedback_fn("* changing into standalone mode")
3564
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3565
                                               self.instance.disks)
3566
    for node, nres in result.items():
3567
      msg = nres.RemoteFailMsg()
3568
      if msg:
3569
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3570
                                 " error %s" % (node, msg))
3571

    
3572
  def _GoReconnect(self, multimaster):
3573
    """Reconnect to the network.
3574

3575
    """
3576
    if multimaster:
3577
      msg = "dual-master"
3578
    else:
3579
      msg = "single-master"
3580
    self.feedback_fn("* changing disks into %s mode" % msg)
3581
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3582
                                           self.instance.disks,
3583
                                           self.instance.name, multimaster)
3584
    for node, nres in result.items():
3585
      msg = nres.RemoteFailMsg()
3586
      if msg:
3587
        raise errors.OpExecError("Cannot change disks config on node %s,"
3588
                                 " error: %s" % (node, msg))
3589

    
3590
  def _ExecCleanup(self):
3591
    """Try to cleanup after a failed migration.
3592

3593
    The cleanup is done by:
3594
      - check that the instance is running only on one node
3595
        (and update the config if needed)
3596
      - change disks on its secondary node to secondary
3597
      - wait until disks are fully synchronized
3598
      - disconnect from the network
3599
      - change disks into single-master mode
3600
      - wait again until disks are fully synchronized
3601

3602
    """
3603
    instance = self.instance
3604
    target_node = self.target_node
3605
    source_node = self.source_node
3606

    
3607
    # check running on only one node
3608
    self.feedback_fn("* checking where the instance actually runs"
3609
                     " (if this hangs, the hypervisor might be in"
3610
                     " a bad state)")
3611
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3612
    for node, result in ins_l.items():
3613
      result.Raise()
3614
      if not isinstance(result.data, list):
3615
        raise errors.OpExecError("Can't contact node '%s'" % node)
3616

    
3617
    runningon_source = instance.name in ins_l[source_node].data
3618
    runningon_target = instance.name in ins_l[target_node].data
3619

    
3620
    if runningon_source and runningon_target:
3621
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3622
                               " or the hypervisor is confused. You will have"
3623
                               " to ensure manually that it runs only on one"
3624
                               " and restart this operation.")
3625

    
3626
    if not (runningon_source or runningon_target):
3627
      raise errors.OpExecError("Instance does not seem to be running at all."
3628
                               " In this case, it's safer to repair by"
3629
                               " running 'gnt-instance stop' to ensure disk"
3630
                               " shutdown, and then restarting it.")
3631

    
3632
    if runningon_target:
3633
      # the migration has actually succeeded, we need to update the config
3634
      self.feedback_fn("* instance running on secondary node (%s),"
3635
                       " updating config" % target_node)
3636
      instance.primary_node = target_node
3637
      self.cfg.Update(instance)
3638
      demoted_node = source_node
3639
    else:
3640
      self.feedback_fn("* instance confirmed to be running on its"
3641
                       " primary node (%s)" % source_node)
3642
      demoted_node = target_node
3643

    
3644
    self._EnsureSecondary(demoted_node)
3645
    try:
3646
      self._WaitUntilSync()
3647
    except errors.OpExecError:
3648
      # we ignore here errors, since if the device is standalone, it
3649
      # won't be able to sync
3650
      pass
3651
    self._GoStandalone()
3652
    self._GoReconnect(False)
3653
    self._WaitUntilSync()
3654

    
3655
    self.feedback_fn("* done")
3656

    
3657
  def _RevertDiskStatus(self):
3658
    """Try to revert the disk status after a failed migration.
3659

3660
    """
3661
    target_node = self.target_node
3662
    try:
3663
      self._EnsureSecondary(target_node)
3664
      self._GoStandalone()
3665
      self._GoReconnect(False)
3666
      self._WaitUntilSync()
3667
    except errors.OpExecError, err:
3668
      self.LogWarning("Migration failed and I can't reconnect the"
3669
                      " drives: error '%s'\n"
3670
                      "Please look and recover the instance status" %
3671
                      str(err))
3672

    
3673
  def _AbortMigration(self):
3674
    """Call the hypervisor code to abort a started migration.
3675

3676
    """
3677
    instance = self.instance
3678
    target_node = self.target_node
3679
    migration_info = self.migration_info
3680

    
3681
    abort_result = self.rpc.call_finalize_migration(target_node,
3682
                                                    instance,
3683
                                                    migration_info,
3684
                                                    False)
3685
    abort_msg = abort_result.RemoteFailMsg()
3686
    if abort_msg:
3687
      logging.error("Aborting migration failed on target node %s: %s" %
3688
                    (target_node, abort_msg))
3689
      # Don't raise an exception here, as we stil have to try to revert the
3690
      # disk status, even if this step failed.
3691

    
3692
  def _ExecMigration(self):
3693
    """Migrate an instance.
3694

3695
    The migrate is done by:
3696
      - change the disks into dual-master mode
3697
      - wait until disks are fully synchronized again
3698
      - migrate the instance
3699
      - change disks on the new secondary node (the old primary) to secondary
3700
      - wait until disks are fully synchronized
3701
      - change disks into single-master mode
3702

3703
    """
3704
    instance = self.instance
3705
    target_node = self.target_node
3706
    source_node = self.source_node
3707

    
3708
    self.feedback_fn("* checking disk consistency between source and target")
3709
    for dev in instance.disks:
3710
      if not _CheckDiskConsistency(self, dev, target_node, False):
3711
        raise errors.OpExecError("Disk %s is degraded or not fully"
3712
                                 " synchronized on target node,"
3713
                                 " aborting migrate." % dev.iv_name)
3714

    
3715
    # First get the migration information from the remote node
3716
    result = self.rpc.call_migration_info(source_node, instance)
3717
    msg = result.RemoteFailMsg()
3718
    if msg:
3719
      log_err = ("Failed fetching source migration information from %s: %s" %
3720
                  (source_node, msg))
3721
      logging.error(log_err)
3722
      raise errors.OpExecError(log_err)
3723

    
3724
    self.migration_info = migration_info = result.data[1]
3725

    
3726
    # Then switch the disks to master/master mode
3727
    self._EnsureSecondary(target_node)
3728
    self._GoStandalone()
3729
    self._GoReconnect(True)
3730
    self._WaitUntilSync()
3731

    
3732
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
3733
    result = self.rpc.call_accept_instance(target_node,
3734
                                           instance,
3735
                                           migration_info,
3736
                                           self.nodes_ip[target_node])
3737

    
3738
    msg = result.RemoteFailMsg()
3739
    if msg:
3740
      logging.error("Instance pre-migration failed, trying to revert"
3741
                    " disk status: %s", msg)
3742
      self._AbortMigration()
3743
      self._RevertDiskStatus()
3744
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
3745
                               (instance.name, msg))
3746

    
3747
    self.feedback_fn("* migrating instance to %s" % target_node)
3748
    time.sleep(10)
3749
    result = self.rpc.call_instance_migrate(source_node, instance,
3750
                                            self.nodes_ip[target_node],
3751
                                            self.op.live)
3752
    msg = result.RemoteFailMsg()
3753
    if msg:
3754
      logging.error("Instance migration failed, trying to revert"
3755
                    " disk status: %s", msg)
3756
      self._AbortMigration()
3757
      self._RevertDiskStatus()
3758
      raise errors.OpExecError("Could not migrate instance %s: %s" %
3759
                               (instance.name, msg))
3760
    time.sleep(10)
3761

    
3762
    instance.primary_node = target_node
3763
    # distribute new instance config to the other nodes
3764
    self.cfg.Update(instance)
3765

    
3766
    result = self.rpc.call_finalize_migration(target_node,
3767
                                              instance,
3768
                                              migration_info,
3769
                                              True)
3770
    msg = result.RemoteFailMsg()
3771
    if msg:
3772
      logging.error("Instance migration succeeded, but finalization failed:"
3773
                    " %s" % msg)
3774
      raise errors.OpExecError("Could not finalize instance migration: %s" %
3775
                               msg)
3776

    
3777
    self._EnsureSecondary(source_node)
3778
    self._WaitUntilSync()
3779
    self._GoStandalone()
3780
    self._GoReconnect(False)
3781
    self._WaitUntilSync()
3782

    
3783
    self.feedback_fn("* done")
3784

    
3785
  def Exec(self, feedback_fn):
3786
    """Perform the migration.
3787

3788
    """
3789
    self.feedback_fn = feedback_fn
3790

    
3791
    self.source_node = self.instance.primary_node
3792
    self.target_node = self.instance.secondary_nodes[0]
3793
    self.all_nodes = [self.source_node, self.target_node]
3794
    self.nodes_ip = {
3795
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
3796
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
3797
      }
3798
    if self.op.cleanup:
3799
      return self._ExecCleanup()
3800
    else:
3801
      return self._ExecMigration()
3802

    
3803

    
3804
def _CreateBlockDev(lu, node, instance, device, force_create,
3805
                    info, force_open):
3806
  """Create a tree of block devices on a given node.
3807

3808
  If this device type has to be created on secondaries, create it and
3809
  all its children.
3810

3811
  If not, just recurse to children keeping the same 'force' value.
3812

3813
  @param lu: the lu on whose behalf we execute
3814
  @param node: the node on which to create the device
3815
  @type instance: L{objects.Instance}
3816
  @param instance: the instance which owns the device
3817
  @type device: L{objects.Disk}
3818
  @param device: the device to create
3819
  @type force_create: boolean
3820
  @param force_create: whether to force creation of this device; this
3821
      will be change to True whenever we find a device which has
3822
      CreateOnSecondary() attribute
3823
  @param info: the extra 'metadata' we should attach to the device
3824
      (this will be represented as a LVM tag)
3825
  @type force_open: boolean
3826
  @param force_open: this parameter will be passes to the
3827
      L{backend.CreateBlockDevice} function where it specifies
3828
      whether we run on primary or not, and it affects both
3829
      the child assembly and the device own Open() execution
3830

3831
  """
3832
  if device.CreateOnSecondary():
3833
    force_create = True
3834

    
3835
  if device.children:
3836
    for child in device.children:
3837
      _CreateBlockDev(lu, node, instance, child, force_create,
3838
                      info, force_open)
3839

    
3840
  if not force_create:
3841
    return
3842

    
3843
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
3844

    
3845

    
3846
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
3847
  """Create a single block device on a given node.
3848

3849
  This will not recurse over children of the device, so they must be
3850
  created in advance.
3851

3852
  @param lu: the lu on whose behalf we execute
3853
  @param node: the node on which to create the device
3854
  @type instance: L{objects.Instance}
3855
  @param instance: the instance which owns the device
3856
  @type device: L{objects.Disk}
3857
  @param device: the device to create
3858
  @param info: the extra 'metadata' we should attach to the device
3859
      (this will be represented as a LVM tag)
3860
  @type force_open: boolean
3861
  @param force_open: this parameter will be passes to the
3862
      L{backend.CreateBlockDevice} function where it specifies
3863
      whether we run on primary or not, and it affects both
3864
      the child assembly and the device own Open() execution
3865

3866
  """
3867
  lu.cfg.SetDiskID(device, node)
3868
  result = lu.rpc.call_blockdev_create(node, device, device.size,
3869
                                       instance.name, force_open, info)
3870
  msg = result.RemoteFailMsg()
3871
  if msg:
3872
    raise errors.OpExecError("Can't create block device %s on"
3873
                             " node %s for instance %s: %s" %
3874
                             (device, node, instance.name, msg))
3875
  if device.physical_id is None:
3876
    device.physical_id = result.data[1]
3877

    
3878

    
3879
def _GenerateUniqueNames(lu, exts):
3880
  """Generate a suitable LV name.
3881

3882
  This will generate a logical volume name for the given instance.
3883

3884
  """
3885
  results = []
3886
  for val in exts:
3887
    new_id = lu.cfg.GenerateUniqueID()
3888
    results.append("%s%s" % (new_id, val))
3889
  return results
3890

    
3891

    
3892
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
3893
                         p_minor, s_minor):
3894
  """Generate a drbd8 device complete with its children.
3895

3896
  """
3897
  port = lu.cfg.AllocatePort()
3898
  vgname = lu.cfg.GetVGName()
3899
  shared_secret = lu.cfg.GenerateDRBDSecret()
3900
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
3901
                          logical_id=(vgname, names[0]))
3902
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
3903
                          logical_id=(vgname, names[1]))
3904
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
3905
                          logical_id=(primary, secondary, port,
3906
                                      p_minor, s_minor,
3907
                                      shared_secret),
3908
                          children=[dev_data, dev_meta],
3909
                          iv_name=iv_name)
3910
  return drbd_dev
3911

    
3912

    
3913
def _GenerateDiskTemplate(lu, template_name,
3914
                          instance_name, primary_node,
3915
                          secondary_nodes, disk_info,
3916
                          file_storage_dir, file_driver,
3917
                          base_index):
3918
  """Generate the entire disk layout for a given template type.
3919

3920
  """
3921
  #TODO: compute space requirements
3922

    
3923
  vgname = lu.cfg.GetVGName()
3924
  disk_count = len(disk_info)
3925
  disks = []
3926
  if template_name == constants.DT_DISKLESS:
3927
    pass
3928
  elif template_name == constants.DT_PLAIN:
3929
    if len(secondary_nodes) != 0:
3930
      raise errors.ProgrammerError("Wrong template configuration")
3931

    
3932
    names = _GenerateUniqueNames(lu, [".disk%d" % i
3933
                                      for i in range(disk_count)])
3934
    for idx, disk in enumerate(disk_info):
3935
      disk_index = idx + base_index
3936
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
3937
                              logical_id=(vgname, names[idx]),
3938
                              iv_name="disk/%d" % disk_index,
3939
                              mode=disk["mode"])
3940
      disks.append(disk_dev)
3941
  elif template_name == constants.DT_DRBD8:
3942
    if len(secondary_nodes) != 1:
3943
      raise errors.ProgrammerError("Wrong template configuration")
3944
    remote_node = secondary_nodes[0]
3945
    minors = lu.cfg.AllocateDRBDMinor(
3946
      [primary_node, remote_node] * len(disk_info), instance_name)
3947

    
3948
    names = []
3949
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % i
3950
                                               for i in range(disk_count)]):
3951
      names.append(lv_prefix + "_data")
3952
      names.append(lv_prefix + "_meta")
3953
    for idx, disk in enumerate(disk_info):
3954
      disk_index = idx + base_index
3955
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
3956
                                      disk["size"], names[idx*2:idx*2+2],
3957
                                      "disk/%d" % disk_index,
3958
                                      minors[idx*2], minors[idx*2+1])
3959
      disk_dev.mode = disk["mode"]
3960
      disks.append(disk_dev)
3961
  elif template_name == constants.DT_FILE:
3962
    if len(secondary_nodes) != 0:
3963
      raise errors.ProgrammerError("Wrong template configuration")
3964

    
3965
    for idx, disk in enumerate(disk_info):
3966
      disk_index = idx + base_index
3967
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
3968
                              iv_name="disk/%d" % disk_index,
3969
                              logical_id=(file_driver,
3970
                                          "%s/disk%d" % (file_storage_dir,
3971
                                                         idx)),
3972
                              mode=disk["mode"])
3973
      disks.append(disk_dev)
3974
  else:
3975
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
3976
  return disks
3977

    
3978

    
3979
def _GetInstanceInfoText(instance):
3980
  """Compute that text that should be added to the disk's metadata.
3981

3982
  """
3983
  return "originstname+%s" % instance.name
3984

    
3985

    
3986
def _CreateDisks(lu, instance):
3987
  """Create all disks for an instance.
3988

3989
  This abstracts away some work from AddInstance.
3990

3991
  @type lu: L{LogicalUnit}
3992
  @param lu: the logical unit on whose behalf we execute
3993
  @type instance: L{objects.Instance}
3994
  @param instance: the instance whose disks we should create
3995
  @rtype: boolean
3996
  @return: the success of the creation
3997

3998
  """
3999
  info = _GetInstanceInfoText(instance)
4000
  pnode = instance.primary_node
4001

    
4002
  if instance.disk_template == constants.DT_FILE:
4003
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4004
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
4005

    
4006
    if result.failed or not result.data:
4007
      raise errors.OpExecError("Could not connect to node '%s'" % pnode)
4008

    
4009
    if not result.data[0]:
4010
      raise errors.OpExecError("Failed to create directory '%s'" %
4011
                               file_storage_dir)
4012

    
4013
  # Note: this needs to be kept in sync with adding of disks in
4014
  # LUSetInstanceParams
4015
  for device in instance.disks:
4016
    logging.info("Creating volume %s for instance %s",
4017
                 device.iv_name, instance.name)
4018
    #HARDCODE
4019
    for node in instance.all_nodes:
4020
      f_create = node == pnode
4021
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
4022

    
4023

    
4024
def _RemoveDisks(lu, instance):
4025
  """Remove all disks for an instance.
4026

4027
  This abstracts away some work from `AddInstance()` and
4028
  `RemoveInstance()`. Note that in case some of the devices couldn't
4029
  be removed, the removal will continue with the other ones (compare
4030
  with `_CreateDisks()`).
4031

4032
  @type lu: L{LogicalUnit}
4033
  @param lu: the logical unit on whose behalf we execute
4034
  @type instance: L{objects.Instance}
4035
  @param instance: the instance whose disks we should remove
4036
  @rtype: boolean
4037
  @return: the success of the removal
4038

4039
  """
4040
  logging.info("Removing block devices for instance %s", instance.name)
4041

    
4042
  result = True
4043
  for device in instance.disks:
4044
    for node, disk in device.ComputeNodeTree(instance.primary_node):
4045
      lu.cfg.SetDiskID(disk, node)
4046
      result = lu.rpc.call_blockdev_remove(node, disk)
4047
      if result.failed or not result.data:
4048
        lu.proc.LogWarning("Could not remove block device %s on node %s,"
4049
                           " continuing anyway", device.iv_name, node)
4050
        result = False
4051

    
4052
  if instance.disk_template == constants.DT_FILE:
4053
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4054
    result = lu.rpc.call_file_storage_dir_remove(instance.primary_node,
4055
                                                 file_storage_dir)
4056
    if result.failed or not result.data:
4057
      logging.error("Could not remove directory '%s'", file_storage_dir)
4058
      result = False
4059

    
4060
  return result
4061

    
4062

    
4063
def _ComputeDiskSize(disk_template, disks):
4064
  """Compute disk size requirements in the volume group
4065

4066
  """
4067
  # Required free disk space as a function of disk and swap space
4068
  req_size_dict = {
4069
    constants.DT_DISKLESS: None,
4070
    constants.DT_PLAIN: sum(d["size"] for d in disks),
4071
    # 128 MB are added for drbd metadata for each disk
4072
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
4073
    constants.DT_FILE: None,
4074
  }
4075

    
4076
  if disk_template not in req_size_dict:
4077
    raise errors.ProgrammerError("Disk template '%s' size requirement"
4078
                                 " is unknown" %  disk_template)
4079

    
4080
  return req_size_dict[disk_template]
4081

    
4082

    
4083
def _CheckHVParams(lu, nodenames, hvname, hvparams):
4084
  """Hypervisor parameter validation.
4085

4086
  This function abstract the hypervisor parameter validation to be
4087
  used in both instance create and instance modify.
4088

4089
  @type lu: L{LogicalUnit}
4090
  @param lu: the logical unit for which we check
4091
  @type nodenames: list
4092
  @param nodenames: the list of nodes on which we should check
4093
  @type hvname: string
4094
  @param hvname: the name of the hypervisor we should use
4095
  @type hvparams: dict
4096
  @param hvparams: the parameters which we need to check
4097
  @raise errors.OpPrereqError: if the parameters are not valid
4098

4099
  """
4100
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
4101
                                                  hvname,
4102
                                                  hvparams)
4103
  for node in nodenames:
4104
    info = hvinfo[node]
4105
    if info.offline:
4106
      continue
4107
    info.Raise()
4108
    if not info.data or not isinstance(info.data, (tuple, list)):
4109
      raise errors.OpPrereqError("Cannot get current information"
4110
                                 " from node '%s' (%s)" % (node, info.data))
4111
    if not info.data[0]:
4112
      raise errors.OpPrereqError("Hypervisor parameter validation failed:"
4113
                                 " %s" % info.data[1])
4114

    
4115

    
4116
class LUCreateInstance(LogicalUnit):
4117
  """Create an instance.
4118

4119
  """
4120
  HPATH = "instance-add"
4121
  HTYPE = constants.HTYPE_INSTANCE
4122
  _OP_REQP = ["instance_name", "disks", "disk_template",
4123
              "mode", "start",
4124
              "wait_for_sync", "ip_check", "nics",
4125
              "hvparams", "beparams"]
4126
  REQ_BGL = False
4127

    
4128
  def _ExpandNode(self, node):
4129
    """Expands and checks one node name.
4130

4131
    """
4132
    node_full = self.cfg.ExpandNodeName(node)
4133
    if node_full is None:
4134
      raise errors.OpPrereqError("Unknown node %s" % node)
4135
    return node_full
4136

    
4137
  def ExpandNames(self):
4138
    """ExpandNames for CreateInstance.
4139

4140
    Figure out the right locks for instance creation.
4141

4142
    """
4143
    self.needed_locks = {}
4144

    
4145
    # set optional parameters to none if they don't exist
4146
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
4147
      if not hasattr(self.op, attr):
4148
        setattr(self.op, attr, None)
4149

    
4150
    # cheap checks, mostly valid constants given
4151

    
4152
    # verify creation mode
4153
    if self.op.mode not in (constants.INSTANCE_CREATE,
4154
                            constants.INSTANCE_IMPORT):
4155
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
4156
                                 self.op.mode)
4157

    
4158
    # disk template and mirror node verification
4159
    if self.op.disk_template not in constants.DISK_TEMPLATES:
4160
      raise errors.OpPrereqError("Invalid disk template name")
4161

    
4162
    if self.op.hypervisor is None:
4163
      self.op.hypervisor = self.cfg.GetHypervisorType()
4164

    
4165
    cluster = self.cfg.GetClusterInfo()
4166
    enabled_hvs = cluster.enabled_hypervisors
4167
    if self.op.hypervisor not in enabled_hvs:
4168
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
4169
                                 " cluster (%s)" % (self.op.hypervisor,
4170
                                  ",".join(enabled_hvs)))
4171

    
4172
    # check hypervisor parameter syntax (locally)
4173

    
4174
    filled_hvp = cluster.FillDict(cluster.hvparams[self.op.hypervisor],
4175
                                  self.op.hvparams)
4176
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
4177
    hv_type.CheckParameterSyntax(filled_hvp)
4178

    
4179
    # fill and remember the beparams dict
4180
    utils.CheckBEParams(self.op.beparams)
4181
    self.be_full = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
4182
                                    self.op.beparams)
4183

    
4184
    #### instance parameters check
4185

    
4186
    # instance name verification
4187
    hostname1 = utils.HostInfo(self.op.instance_name)
4188
    self.op.instance_name = instance_name = hostname1.name
4189

    
4190
    # this is just a preventive check, but someone might still add this
4191
    # instance in the meantime, and creation will fail at lock-add time
4192
    if instance_name in self.cfg.GetInstanceList():
4193
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4194
                                 instance_name)
4195

    
4196
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
4197

    
4198
    # NIC buildup
4199
    self.nics = []
4200
    for nic in self.op.nics:
4201
      # ip validity checks
4202
      ip = nic.get("ip", None)
4203
      if ip is None or ip.lower() == "none":
4204
        nic_ip = None
4205
      elif ip.lower() == constants.VALUE_AUTO:
4206
        nic_ip = hostname1.ip
4207
      else:
4208
        if not utils.IsValidIP(ip):
4209
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
4210
                                     " like a valid IP" % ip)
4211
        nic_ip = ip
4212

    
4213
      # MAC address verification
4214
      mac = nic.get("mac", constants.VALUE_AUTO)
4215
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4216
        if not utils.IsValidMac(mac.lower()):
4217
          raise errors.OpPrereqError("Invalid MAC address specified: %s" %
4218
                                     mac)
4219
      # bridge verification
4220
      bridge = nic.get("bridge", None)
4221
      if bridge is None:
4222
        bridge = self.cfg.GetDefBridge()
4223
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
4224

    
4225
    # disk checks/pre-build
4226
    self.disks = []
4227
    for disk in self.op.disks:
4228
      mode = disk.get("mode", constants.DISK_RDWR)
4229
      if mode not in constants.DISK_ACCESS_SET:
4230
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
4231
                                   mode)
4232
      size = disk.get("size", None)
4233
      if size is None:
4234
        raise errors.OpPrereqError("Missing disk size")
4235
      try:
4236
        size = int(size)
4237
      except ValueError:
4238
        raise errors.OpPrereqError("Invalid disk size '%s'" % size)
4239
      self.disks.append({"size": size, "mode": mode})
4240

    
4241
    # used in CheckPrereq for ip ping check
4242
    self.check_ip = hostname1.ip
4243

    
4244
    # file storage checks
4245
    if (self.op.file_driver and
4246
        not self.op.file_driver in constants.FILE_DRIVER):
4247
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
4248
                                 self.op.file_driver)
4249

    
4250
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
4251
      raise errors.OpPrereqError("File storage directory path not absolute")
4252

    
4253
    ### Node/iallocator related checks
4254
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
4255
      raise errors.OpPrereqError("One and only one of iallocator and primary"
4256
                                 " node must be given")
4257

    
4258
    if self.op.iallocator:
4259
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4260
    else:
4261
      self.op.pnode = self._ExpandNode(self.op.pnode)
4262
      nodelist = [self.op.pnode]
4263
      if self.op.snode is not None:
4264
        self.op.snode = self._ExpandNode(self.op.snode)
4265
        nodelist.append(self.op.snode)
4266
      self.needed_locks[locking.LEVEL_NODE] = nodelist
4267

    
4268
    # in case of import lock the source node too
4269
    if self.op.mode == constants.INSTANCE_IMPORT:
4270
      src_node = getattr(self.op, "src_node", None)
4271
      src_path = getattr(self.op, "src_path", None)
4272

    
4273
      if src_path is None:
4274
        self.op.src_path = src_path = self.op.instance_name
4275

    
4276
      if src_node is None:
4277
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4278
        self.op.src_node = None
4279
        if os.path.isabs(src_path):
4280
          raise errors.OpPrereqError("Importing an instance from an absolute"
4281
                                     " path requires a source node option.")
4282
      else:
4283
        self.op.src_node = src_node = self._ExpandNode(src_node)
4284
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4285
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
4286
        if not os.path.isabs(src_path):
4287
          self.op.src_path = src_path = \
4288
            os.path.join(constants.EXPORT_DIR, src_path)
4289

    
4290
    else: # INSTANCE_CREATE
4291
      if getattr(self.op, "os_type", None) is None:
4292
        raise errors.OpPrereqError("No guest OS specified")
4293

    
4294
  def _RunAllocator(self):
4295
    """Run the allocator based on input opcode.
4296

4297
    """
4298
    nics = [n.ToDict() for n in self.nics]
4299
    ial = IAllocator(self,
4300
                     mode=constants.IALLOCATOR_MODE_ALLOC,
4301
                     name=self.op.instance_name,
4302
                     disk_template=self.op.disk_template,
4303
                     tags=[],
4304
                     os=self.op.os_type,
4305
                     vcpus=self.be_full[constants.BE_VCPUS],
4306
                     mem_size=self.be_full[constants.BE_MEMORY],
4307
                     disks=self.disks,
4308
                     nics=nics,
4309
                     hypervisor=self.op.hypervisor,
4310
                     )
4311

    
4312
    ial.Run(self.op.iallocator)
4313

    
4314
    if not ial.success:
4315
      raise errors.OpPrereqError("Can't compute nodes using"
4316
                                 " iallocator '%s': %s" % (self.op.iallocator,
4317
                                                           ial.info))
4318
    if len(ial.nodes) != ial.required_nodes:
4319
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4320
                                 " of nodes (%s), required %s" %
4321
                                 (self.op.iallocator, len(ial.nodes),
4322
                                  ial.required_nodes))
4323
    self.op.pnode = ial.nodes[0]
4324
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
4325
                 self.op.instance_name, self.op.iallocator,
4326
                 ", ".join(ial.nodes))
4327
    if ial.required_nodes == 2:
4328
      self.op.snode = ial.nodes[1]
4329

    
4330
  def BuildHooksEnv(self):
4331
    """Build hooks env.
4332

4333
    This runs on master, primary and secondary nodes of the instance.
4334

4335
    """
4336
    env = {
4337
      "INSTANCE_DISK_TEMPLATE": self.op.disk_template,
4338
      "INSTANCE_DISK_SIZE": ",".join(str(d["size"]) for d in self.disks),
4339
      "INSTANCE_ADD_MODE": self.op.mode,
4340
      }
4341
    if self.op.mode == constants.INSTANCE_IMPORT:
4342
      env["INSTANCE_SRC_NODE"] = self.op.src_node
4343
      env["INSTANCE_SRC_PATH"] = self.op.src_path
4344
      env["INSTANCE_SRC_IMAGES"] = self.src_images
4345

    
4346
    env.update(_BuildInstanceHookEnv(name=self.op.instance_name,
4347
      primary_node=self.op.pnode,
4348
      secondary_nodes=self.secondaries,
4349
      status=self.instance_status,
4350
      os_type=self.op.os_type,
4351
      memory=self.be_full[constants.BE_MEMORY],
4352
      vcpus=self.be_full[constants.BE_VCPUS],
4353
      nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
4354
    ))
4355

    
4356
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
4357
          self.secondaries)
4358
    return env, nl, nl
4359

    
4360

    
4361
  def CheckPrereq(self):
4362
    """Check prerequisites.
4363

4364
    """
4365
    if (not self.cfg.GetVGName() and
4366
        self.op.disk_template not in constants.DTS_NOT_LVM):
4367
      raise errors.OpPrereqError("Cluster does not support lvm-based"
4368
                                 " instances")
4369

    
4370

    
4371
    if self.op.mode == constants.INSTANCE_IMPORT:
4372
      src_node = self.op.src_node
4373
      src_path = self.op.src_path
4374

    
4375
      if src_node is None:
4376
        exp_list = self.rpc.call_export_list(
4377
          self.acquired_locks[locking.LEVEL_NODE])
4378
        found = False
4379
        for node in exp_list:
4380
          if not exp_list[node].failed and src_path in exp_list[node].data:
4381
            found = True
4382
            self.op.src_node = src_node = node
4383
            self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
4384
                                                       src_path)
4385
            break
4386
        if not found:
4387
          raise errors.OpPrereqError("No export found for relative path %s" %
4388
                                      src_path)
4389

    
4390
      _CheckNodeOnline(self, src_node)
4391
      result = self.rpc.call_export_info(src_node, src_path)
4392
      result.Raise()
4393
      if not result.data:
4394
        raise errors.OpPrereqError("No export found in dir %s" % src_path)
4395

    
4396
      export_info = result.data
4397
      if not export_info.has_section(constants.INISECT_EXP):
4398
        raise errors.ProgrammerError("Corrupted export config")
4399

    
4400
      ei_version = export_info.get(constants.INISECT_EXP, 'version')
4401
      if (int(ei_version) != constants.EXPORT_VERSION):
4402
        raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
4403
                                   (ei_version, constants.EXPORT_VERSION))
4404

    
4405
      # Check that the new instance doesn't have less disks than the export
4406
      instance_disks = len(self.disks)
4407
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
4408
      if instance_disks < export_disks:
4409
        raise errors.OpPrereqError("Not enough disks to import."
4410
                                   " (instance: %d, export: %d)" %
4411
                                   (instance_disks, export_disks))
4412

    
4413
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
4414
      disk_images = []
4415
      for idx in range(export_disks):
4416
        option = 'disk%d_dump' % idx
4417
        if export_info.has_option(constants.INISECT_INS, option):
4418
          # FIXME: are the old os-es, disk sizes, etc. useful?
4419
          export_name = export_info.get(constants.INISECT_INS, option)
4420
          image = os.path.join(src_path, export_name)
4421
          disk_images.append(image)
4422
        else:
4423
          disk_images.append(False)
4424

    
4425
      self.src_images = disk_images
4426

    
4427
      old_name = export_info.get(constants.INISECT_INS, 'name')
4428
      # FIXME: int() here could throw a ValueError on broken exports
4429
      exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
4430
      if self.op.instance_name == old_name:
4431
        for idx, nic in enumerate(self.nics):
4432
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
4433
            nic_mac_ini = 'nic%d_mac' % idx
4434
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
4435

    
4436
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
4437
    if self.op.start and not self.op.ip_check:
4438
      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
4439
                                 " adding an instance in start mode")
4440

    
4441
    if self.op.ip_check:
4442
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
4443
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4444
                                   (self.check_ip, self.op.instance_name))
4445

    
4446
    #### allocator run
4447

    
4448
    if self.op.iallocator is not None:
4449
      self._RunAllocator()
4450

    
4451
    #### node related checks
4452

    
4453
    # check primary node
4454
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
4455
    assert self.pnode is not None, \
4456
      "Cannot retrieve locked node %s" % self.op.pnode
4457
    if pnode.offline:
4458
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
4459
                                 pnode.name)
4460

    
4461
    self.secondaries = []
4462

    
4463
    # mirror node verification
4464
    if self.op.disk_template in constants.DTS_NET_MIRROR:
4465
      if self.op.snode is None:
4466
        raise errors.OpPrereqError("The networked disk templates need"
4467
                                   " a mirror node")
4468
      if self.op.snode == pnode.name:
4469
        raise errors.OpPrereqError("The secondary node cannot be"
4470
                                   " the primary node.")
4471
      self.secondaries.append(self.op.snode)
4472
      _CheckNodeOnline(self, self.op.snode)
4473

    
4474
    nodenames = [pnode.name] + self.secondaries
4475

    
4476
    req_size = _ComputeDiskSize(self.op.disk_template,
4477
                                self.disks)
4478

    
4479
    # Check lv size requirements
4480
    if req_size is not None:
4481
      nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
4482
                                         self.op.hypervisor)
4483
      for node in nodenames:
4484
        info = nodeinfo[node]
4485
        info.Raise()
4486
        info = info.data
4487
        if not info:
4488
          raise errors.OpPrereqError("Cannot get current information"
4489
                                     " from node '%s'" % node)
4490
        vg_free = info.get('vg_free', None)
4491
        if not isinstance(vg_free, int):
4492
          raise errors.OpPrereqError("Can't compute free disk space on"
4493
                                     " node %s" % node)
4494
        if req_size > info['vg_free']:
4495
          raise errors.OpPrereqError("Not enough disk space on target node %s."
4496
                                     " %d MB available, %d MB required" %
4497
                                     (node, info['vg_free'], req_size))
4498

    
4499
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
4500

    
4501
    # os verification
4502
    result = self.rpc.call_os_get(pnode.name, self.op.os_type)
4503
    result.Raise()
4504
    if not isinstance(result.data, objects.OS):
4505
      raise errors.OpPrereqError("OS '%s' not in supported os list for"
4506
                                 " primary node"  % self.op.os_type)
4507

    
4508
    # bridge check on primary node
4509
    bridges = [n.bridge for n in self.nics]
4510
    result = self.rpc.call_bridges_exist(self.pnode.name, bridges)
4511
    result.Raise()
4512
    if not result.data:
4513
      raise errors.OpPrereqError("One of the target bridges '%s' does not"
4514
                                 " exist on destination node '%s'" %
4515
                                 (",".join(bridges), pnode.name))
4516

    
4517
    # memory check on primary node
4518
    if self.op.start:
4519
      _CheckNodeFreeMemory(self, self.pnode.name,
4520
                           "creating instance %s" % self.op.instance_name,
4521
                           self.be_full[constants.BE_MEMORY],
4522
                           self.op.hypervisor)
4523

    
4524
    self.instance_status = self.op.start
4525

    
4526
  def Exec(self, feedback_fn):
4527
    """Create and add the instance to the cluster.
4528

4529
    """
4530
    instance = self.op.instance_name
4531
    pnode_name = self.pnode.name
4532

    
4533
    for nic in self.nics:
4534
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4535
        nic.mac = self.cfg.GenerateMAC()
4536

    
4537
    ht_kind = self.op.hypervisor
4538
    if ht_kind in constants.HTS_REQ_PORT:
4539
      network_port = self.cfg.AllocatePort()
4540
    else:
4541
      network_port = None
4542

    
4543
    ##if self.op.vnc_bind_address is None:
4544
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
4545

    
4546
    # this is needed because os.path.join does not accept None arguments
4547
    if self.op.file_storage_dir is None:
4548
      string_file_storage_dir = ""
4549
    else:
4550
      string_file_storage_dir = self.op.file_storage_dir
4551

    
4552
    # build the full file storage dir path
4553
    file_storage_dir = os.path.normpath(os.path.join(
4554
                                        self.cfg.GetFileStorageDir(),
4555
                                        string_file_storage_dir, instance))
4556

    
4557

    
4558
    disks = _GenerateDiskTemplate(self,
4559
                                  self.op.disk_template,
4560
                                  instance, pnode_name,
4561
                                  self.secondaries,
4562
                                  self.disks,
4563
                                  file_storage_dir,
4564
                                  self.op.file_driver,
4565
                                  0)
4566

    
4567
    iobj = objects.Instance(name=instance, os=self.op.os_type,
4568
                            primary_node=pnode_name,
4569
                            nics=self.nics, disks=disks,
4570
                            disk_template=self.op.disk_template,
4571
                            admin_up=self.instance_status,
4572
                            network_port=network_port,
4573
                            beparams=self.op.beparams,
4574
                            hvparams=self.op.hvparams,
4575
                            hypervisor=self.op.hypervisor,
4576
                            )
4577

    
4578
    feedback_fn("* creating instance disks...")
4579
    try:
4580
      _CreateDisks(self, iobj)
4581
    except errors.OpExecError:
4582
      self.LogWarning("Device creation failed, reverting...")
4583
      try:
4584
        _RemoveDisks(self, iobj)
4585
      finally:
4586
        self.cfg.ReleaseDRBDMinors(instance)
4587
        raise
4588

    
4589
    feedback_fn("adding instance %s to cluster config" % instance)
4590

    
4591
    self.cfg.AddInstance(iobj)
4592
    # Declare that we don't want to remove the instance lock anymore, as we've
4593
    # added the instance to the config
4594
    del self.remove_locks[locking.LEVEL_INSTANCE]
4595
    # Unlock all the nodes
4596
    if self.op.mode == constants.INSTANCE_IMPORT:
4597
      nodes_keep = [self.op.src_node]
4598
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
4599
                       if node != self.op.src_node]
4600
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
4601
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
4602
    else:
4603
      self.context.glm.release(locking.LEVEL_NODE)
4604
      del self.acquired_locks[locking.LEVEL_NODE]
4605

    
4606
    if self.op.wait_for_sync:
4607
      disk_abort = not _WaitForSync(self, iobj)
4608
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
4609
      # make sure the disks are not degraded (still sync-ing is ok)
4610
      time.sleep(15)
4611
      feedback_fn("* checking mirrors status")
4612
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
4613
    else:
4614
      disk_abort = False
4615

    
4616
    if disk_abort:
4617
      _RemoveDisks(self, iobj)
4618
      self.cfg.RemoveInstance(iobj.name)
4619
      # Make sure the instance lock gets removed
4620
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
4621
      raise errors.OpExecError("There are some degraded disks for"
4622
                               " this instance")
4623

    
4624
    feedback_fn("creating os for instance %s on node %s" %
4625
                (instance, pnode_name))
4626

    
4627
    if iobj.disk_template != constants.DT_DISKLESS:
4628
      if self.op.mode == constants.INSTANCE_CREATE:
4629
        feedback_fn("* running the instance OS create scripts...")
4630
        result = self.rpc.call_instance_os_add(pnode_name, iobj)
4631
        msg = result.RemoteFailMsg()
4632
        if msg:
4633
          raise errors.OpExecError("Could not add os for instance %s"
4634
                                   " on node %s: %s" %
4635
                                   (instance, pnode_name, msg))
4636

    
4637
      elif self.op.mode == constants.INSTANCE_IMPORT:
4638
        feedback_fn("* running the instance OS import scripts...")
4639
        src_node = self.op.src_node
4640
        src_images = self.src_images
4641
        cluster_name = self.cfg.GetClusterName()
4642
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
4643
                                                         src_node, src_images,
4644
                                                         cluster_name)
4645
        import_result.Raise()
4646
        for idx, result in enumerate(import_result.data):
4647
          if not result:
4648
            self.LogWarning("Could not import the image %s for instance"
4649
                            " %s, disk %d, on node %s" %
4650
                            (src_images[idx], instance, idx, pnode_name))
4651
      else:
4652
        # also checked in the prereq part
4653
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
4654
                                     % self.op.mode)
4655

    
4656
    if self.op.start:
4657
      logging.info("Starting instance %s on node %s", instance, pnode_name)
4658
      feedback_fn("* starting instance...")
4659
      result = self.rpc.call_instance_start(pnode_name, iobj, None)
4660
      msg = result.RemoteFailMsg()
4661
      if msg:
4662
        raise errors.OpExecError("Could not start instance: %s" % msg)
4663

    
4664

    
4665
class LUConnectConsole(NoHooksLU):
4666
  """Connect to an instance's console.
4667

4668
  This is somewhat special in that it returns the command line that
4669
  you need to run on the master node in order to connect to the
4670
  console.
4671

4672
  """
4673
  _OP_REQP = ["instance_name"]
4674
  REQ_BGL = False
4675

    
4676
  def ExpandNames(self):
4677
    self._ExpandAndLockInstance()
4678

    
4679
  def CheckPrereq(self):
4680
    """Check prerequisites.
4681

4682
    This checks that the instance is in the cluster.
4683

4684
    """
4685
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4686
    assert self.instance is not None, \
4687
      "Cannot retrieve locked instance %s" % self.op.instance_name
4688
    _CheckNodeOnline(self, self.instance.primary_node)
4689

    
4690
  def Exec(self, feedback_fn):
4691
    """Connect to the console of an instance
4692

4693
    """
4694
    instance = self.instance
4695
    node = instance.primary_node
4696

    
4697
    node_insts = self.rpc.call_instance_list([node],
4698
                                             [instance.hypervisor])[node]
4699
    node_insts.Raise()
4700

    
4701
    if instance.name not in node_insts.data:
4702
      raise errors.OpExecError("Instance %s is not running." % instance.name)
4703

    
4704
    logging.debug("Connecting to console of %s on %s", instance.name, node)
4705

    
4706
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
4707
    cluster = self.cfg.GetClusterInfo()
4708
    # beparams and hvparams are passed separately, to avoid editing the
4709
    # instance and then saving the defaults in the instance itself.
4710
    hvparams = cluster.FillHV(instance)
4711
    beparams = cluster.FillBE(instance)
4712
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
4713

    
4714
    # build ssh cmdline
4715
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
4716

    
4717

    
4718
class LUReplaceDisks(LogicalUnit):
4719
  """Replace the disks of an instance.
4720

4721
  """
4722
  HPATH = "mirrors-replace"
4723
  HTYPE = constants.HTYPE_INSTANCE
4724
  _OP_REQP = ["instance_name", "mode", "disks"]
4725
  REQ_BGL = False
4726

    
4727
  def CheckArguments(self):
4728
    if not hasattr(self.op, "remote_node"):
4729
      self.op.remote_node = None
4730
    if not hasattr(self.op, "iallocator"):
4731
      self.op.iallocator = None
4732

    
4733
    # check for valid parameter combination
4734
    cnt = [self.op.remote_node, self.op.iallocator].count(None)
4735
    if self.op.mode == constants.REPLACE_DISK_CHG:
4736
      if cnt == 2:
4737
        raise errors.OpPrereqError("When changing the secondary either an"
4738
                                   " iallocator script must be used or the"
4739
                                   " new node given")
4740
      elif cnt == 0:
4741
        raise errors.OpPrereqError("Give either the iallocator or the new"
4742
                                   " secondary, not both")
4743
    else: # not replacing the secondary
4744
      if cnt != 2:
4745
        raise errors.OpPrereqError("The iallocator and new node options can"
4746
                                   " be used only when changing the"
4747
                                   " secondary node")
4748

    
4749
  def ExpandNames(self):
4750
    self._ExpandAndLockInstance()
4751

    
4752
    if self.op.iallocator is not None:
4753
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4754
    elif self.op.remote_node is not None:
4755
      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
4756
      if remote_node is None:
4757
        raise errors.OpPrereqError("Node '%s' not known" %
4758
                                   self.op.remote_node)
4759
      self.op.remote_node = remote_node
4760
      # Warning: do not remove the locking of the new secondary here
4761
      # unless DRBD8.AddChildren is changed to work in parallel;
4762
      # currently it doesn't since parallel invocations of
4763
      # FindUnusedMinor will conflict
4764
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
4765
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
4766
    else:
4767
      self.needed_locks[locking.LEVEL_NODE] = []
4768
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4769

    
4770
  def DeclareLocks(self, level):
4771
    # If we're not already locking all nodes in the set we have to declare the
4772
    # instance's primary/secondary nodes.
4773
    if (level == locking.LEVEL_NODE and
4774
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
4775
      self._LockInstancesNodes()
4776

    
4777
  def _RunAllocator(self):
4778
    """Compute a new secondary node using an IAllocator.
4779

4780
    """
4781
    ial = IAllocator(self,
4782
                     mode=constants.IALLOCATOR_MODE_RELOC,
4783
                     name=self.op.instance_name,
4784
                     relocate_from=[self.sec_node])
4785

    
4786
    ial.Run(self.op.iallocator)
4787

    
4788
    if not ial.success:
4789
      raise errors.OpPrereqError("Can't compute nodes using"
4790
                                 " iallocator '%s': %s" % (self.op.iallocator,
4791
                                                           ial.info))
4792
    if len(ial.nodes) != ial.required_nodes:
4793
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4794
                                 " of nodes (%s), required %s" %
4795
                                 (len(ial.nodes), ial.required_nodes))
4796
    self.op.remote_node = ial.nodes[0]
4797
    self.LogInfo("Selected new secondary for the instance: %s",
4798
                 self.op.remote_node)
4799

    
4800
  def BuildHooksEnv(self):
4801
    """Build hooks env.
4802

4803
    This runs on the master, the primary and all the secondaries.
4804

4805
    """
4806
    env = {
4807
      "MODE": self.op.mode,
4808
      "NEW_SECONDARY": self.op.remote_node,
4809
      "OLD_SECONDARY": self.instance.secondary_nodes[0],
4810
      }
4811
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
4812
    nl = [
4813
      self.cfg.GetMasterNode(),
4814
      self.instance.primary_node,
4815
      ]
4816
    if self.op.remote_node is not None:
4817
      nl.append(self.op.remote_node)
4818
    return env, nl, nl
4819

    
4820
  def CheckPrereq(self):
4821
    """Check prerequisites.
4822

4823
    This checks that the instance is in the cluster.
4824

4825
    """
4826
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4827
    assert instance is not None, \
4828
      "Cannot retrieve locked instance %s" % self.op.instance_name
4829
    self.instance = instance
4830

    
4831
    if instance.disk_template != constants.DT_DRBD8:
4832
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
4833
                                 " instances")
4834

    
4835
    if len(instance.secondary_nodes) != 1:
4836
      raise errors.OpPrereqError("The instance has a strange layout,"
4837
                                 " expected one secondary but found %d" %
4838
                                 len(instance.secondary_nodes))
4839

    
4840
    self.sec_node = instance.secondary_nodes[0]
4841

    
4842
    if self.op.iallocator is not None:
4843
      self._RunAllocator()
4844

    
4845
    remote_node = self.op.remote_node
4846
    if remote_node is not None:
4847
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
4848
      assert self.remote_node_info is not None, \
4849
        "Cannot retrieve locked node %s" % remote_node
4850
    else:
4851
      self.remote_node_info = None
4852
    if remote_node == instance.primary_node:
4853
      raise errors.OpPrereqError("The specified node is the primary node of"
4854
                                 " the instance.")
4855
    elif remote_node == self.sec_node:
4856
      raise errors.OpPrereqError("The specified node is already the"
4857
                                 " secondary node of the instance.")
4858

    
4859
    if self.op.mode == constants.REPLACE_DISK_PRI:
4860
      n1 = self.tgt_node = instance.primary_node
4861
      n2 = self.oth_node = self.sec_node
4862
    elif self.op.mode == constants.REPLACE_DISK_SEC:
4863
      n1 = self.tgt_node = self.sec_node
4864
      n2 = self.oth_node = instance.primary_node
4865
    elif self.op.mode == constants.REPLACE_DISK_CHG:
4866
      n1 = self.new_node = remote_node
4867
      n2 = self.oth_node = instance.primary_node
4868
      self.tgt_node = self.sec_node
4869
    else:
4870
      raise errors.ProgrammerError("Unhandled disk replace mode")
4871

    
4872
    _CheckNodeOnline(self, n1)
4873
    _CheckNodeOnline(self, n2)
4874

    
4875
    if not self.op.disks:
4876
      self.op.disks = range(len(instance.disks))
4877

    
4878
    for disk_idx in self.op.disks:
4879
      instance.FindDisk(disk_idx)
4880

    
4881
  def _ExecD8DiskOnly(self, feedback_fn):
4882
    """Replace a disk on the primary or secondary for dbrd8.
4883

4884
    The algorithm for replace is quite complicated:
4885

4886
      1. for each disk to be replaced:
4887

4888
        1. create new LVs on the target node with unique names
4889
        1. detach old LVs from the drbd device
4890
        1. rename old LVs to name_replaced.<time_t>
4891
        1. rename new LVs to old LVs
4892
        1. attach the new LVs (with the old names now) to the drbd device
4893

4894
      1. wait for sync across all devices
4895

4896
      1. for each modified disk:
4897

4898
        1. remove old LVs (which have the name name_replaces.<time_t>)
4899

4900
    Failures are not very well handled.
4901

4902
    """
4903
    steps_total = 6
4904
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
4905
    instance = self.instance
4906
    iv_names = {}
4907
    vgname = self.cfg.GetVGName()
4908
    # start of work
4909
    cfg = self.cfg
4910
    tgt_node = self.tgt_node
4911
    oth_node = self.oth_node
4912

    
4913
    # Step: check device activation
4914
    self.proc.LogStep(1, steps_total, "check device existence")
4915
    info("checking volume groups")
4916
    my_vg = cfg.GetVGName()
4917
    results = self.rpc.call_vg_list([oth_node, tgt_node])
4918
    if not results:
4919
      raise errors.OpExecError("Can't list volume groups on the nodes")
4920
    for node in oth_node, tgt_node:
4921
      res = results[node]
4922
      if res.failed or not res.data or my_vg not in res.data:
4923
        raise errors.OpExecError("Volume group '%s' not found on %s" %
4924
                                 (my_vg, node))
4925
    for idx, dev in enumerate(instance.disks):
4926
      if idx not in self.op.disks:
4927
        continue
4928
      for node in tgt_node, oth_node:
4929
        info("checking disk/%d on %s" % (idx, node))
4930
        cfg.SetDiskID(dev, node)
4931
        if not self.rpc.call_blockdev_find(node, dev):
4932
          raise errors.OpExecError("Can't find disk/%d on node %s" %
4933
                                   (idx, node))
4934

    
4935
    # Step: check other node consistency
4936
    self.proc.LogStep(2, steps_total, "check peer consistency")
4937
    for idx, dev in enumerate(instance.disks):
4938
      if idx not in self.op.disks:
4939
        continue
4940
      info("checking disk/%d consistency on %s" % (idx, oth_node))
4941
      if not _CheckDiskConsistency(self, dev, oth_node,
4942
                                   oth_node==instance.primary_node):
4943
        raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
4944
                                 " to replace disks on this node (%s)" %
4945
                                 (oth_node, tgt_node))
4946

    
4947
    # Step: create new storage
4948
    self.proc.LogStep(3, steps_total, "allocate new storage")
4949
    for idx, dev in enumerate(instance.disks):
4950
      if idx not in self.op.disks:
4951
        continue
4952
      size = dev.size
4953
      cfg.SetDiskID(dev, tgt_node)
4954
      lv_names = [".disk%d_%s" % (idx, suf)
4955
                  for suf in ["data", "meta"]]
4956
      names = _GenerateUniqueNames(self, lv_names)
4957
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
4958
                             logical_id=(vgname, names[0]))
4959
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
4960
                             logical_id=(vgname, names[1]))
4961
      new_lvs = [lv_data, lv_meta]
4962
      old_lvs = dev.children
4963
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
4964
      info("creating new local storage on %s for %s" %
4965
           (tgt_node, dev.iv_name))
4966
      # we pass force_create=True to force the LVM creation
4967
      for new_lv in new_lvs:
4968
        _CreateBlockDev(self, tgt_node, instance, new_lv, True,
4969
                        _GetInstanceInfoText(instance), False)
4970

    
4971
    # Step: for each lv, detach+rename*2+attach
4972
    self.proc.LogStep(4, steps_total, "change drbd configuration")
4973
    for dev, old_lvs, new_lvs in iv_names.itervalues():
4974
      info("detaching %s drbd from local storage" % dev.iv_name)
4975
      result = self.rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs)
4976
      result.Raise()
4977
      if not result.data:
4978
        raise errors.OpExecError("Can't detach drbd from local storage on node"
4979
                                 " %s for device %s" % (tgt_node, dev.iv_name))
4980
      #dev.children = []
4981
      #cfg.Update(instance)
4982

    
4983
      # ok, we created the new LVs, so now we know we have the needed
4984
      # storage; as such, we proceed on the target node to rename
4985
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
4986
      # using the assumption that logical_id == physical_id (which in
4987
      # turn is the unique_id on that node)
4988

    
4989
      # FIXME(iustin): use a better name for the replaced LVs
4990
      temp_suffix = int(time.time())
4991
      ren_fn = lambda d, suff: (d.physical_id[0],
4992
                                d.physical_id[1] + "_replaced-%s" % suff)
4993
      # build the rename list based on what LVs exist on the node
4994
      rlist = []
4995
      for to_ren in old_lvs:
4996
        find_res = self.rpc.call_blockdev_find(tgt_node, to_ren)
4997
        if not find_res.failed and find_res.data is not None: # device exists
4998
          rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
4999

    
5000
      info("renaming the old LVs on the target node")
5001
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5002
      result.Raise()
5003
      if not result.data:
5004
        raise errors.OpExecError("Can't rename old LVs on node %s" % tgt_node)
5005
      # now we rename the new LVs to the old LVs
5006
      info("renaming the new LVs on the target node")
5007
      rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
5008
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5009
      result.Raise()
5010
      if not result.data:
5011
        raise errors.OpExecError("Can't rename new LVs on node %s" % tgt_node)
5012

    
5013
      for old, new in zip(old_lvs, new_lvs):
5014
        new.logical_id = old.logical_id
5015
        cfg.SetDiskID(new, tgt_node)
5016

    
5017
      for disk in old_lvs:
5018
        disk.logical_id = ren_fn(disk, temp_suffix)
5019
        cfg.SetDiskID(disk, tgt_node)
5020

    
5021
      # now that the new lvs have the old name, we can add them to the device
5022
      info("adding new mirror component on %s" % tgt_node)
5023
      result = self.rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs)
5024
      if result.failed or not result.data:
5025
        for new_lv in new_lvs:
5026
          result = self.rpc.call_blockdev_remove(tgt_node, new_lv)
5027
          if result.failed or not result.data:
5028
            warning("Can't rollback device %s", hint="manually cleanup unused"
5029
                    " logical volumes")
5030
        raise errors.OpExecError("Can't add local storage to drbd")
5031

    
5032
      dev.children = new_lvs
5033
      cfg.Update(instance)
5034

    
5035
    # Step: wait for sync
5036

    
5037
    # this can fail as the old devices are degraded and _WaitForSync
5038
    # does a combined result over all disks, so we don't check its
5039
    # return value
5040
    self.proc.LogStep(5, steps_total, "sync devices")
5041
    _WaitForSync(self, instance, unlock=True)
5042

    
5043
    # so check manually all the devices
5044
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5045
      cfg.SetDiskID(dev, instance.primary_node)
5046
      result = self.rpc.call_blockdev_find(instance.primary_node, dev)
5047
      if result.failed or result.data[5]:
5048
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
5049

    
5050
    # Step: remove old storage
5051
    self.proc.LogStep(6, steps_total, "removing old storage")
5052
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5053
      info("remove logical volumes for %s" % name)
5054
      for lv in old_lvs:
5055
        cfg.SetDiskID(lv, tgt_node)
5056
        result = self.rpc.call_blockdev_remove(tgt_node, lv)
5057
        if result.failed or not result.data:
5058
          warning("Can't remove old LV", hint="manually remove unused LVs")
5059
          continue
5060

    
5061
  def _ExecD8Secondary(self, feedback_fn):
5062
    """Replace the secondary node for drbd8.
5063

5064
    The algorithm for replace is quite complicated:
5065
      - for all disks of the instance:
5066
        - create new LVs on the new node with same names
5067
        - shutdown the drbd device on the old secondary
5068
        - disconnect the drbd network on the primary
5069
        - create the drbd device on the new secondary
5070
        - network attach the drbd on the primary, using an artifice:
5071
          the drbd code for Attach() will connect to the network if it
5072
          finds a device which is connected to the good local disks but
5073
          not network enabled
5074
      - wait for sync across all devices
5075
      - remove all disks from the old secondary
5076

5077
    Failures are not very well handled.
5078

5079
    """
5080
    steps_total = 6
5081
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5082
    instance = self.instance
5083
    iv_names = {}
5084
    # start of work
5085
    cfg = self.cfg
5086
    old_node = self.tgt_node
5087
    new_node = self.new_node
5088
    pri_node = instance.primary_node
5089
    nodes_ip = {
5090
      old_node: self.cfg.GetNodeInfo(old_node).secondary_ip,
5091
      new_node: self.cfg.GetNodeInfo(new_node).secondary_ip,
5092
      pri_node: self.cfg.GetNodeInfo(pri_node).secondary_ip,
5093
      }
5094

    
5095
    # Step: check device activation
5096
    self.proc.LogStep(1, steps_total, "check device existence")
5097
    info("checking volume groups")
5098
    my_vg = cfg.GetVGName()
5099
    results = self.rpc.call_vg_list([pri_node, new_node])
5100
    for node in pri_node, new_node:
5101
      res = results[node]
5102
      if res.failed or not res.data or my_vg not in res.data:
5103
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5104
                                 (my_vg, node))
5105
    for idx, dev in enumerate(instance.disks):
5106
      if idx not in self.op.disks:
5107
        continue
5108
      info("checking disk/%d on %s" % (idx, pri_node))
5109
      cfg.SetDiskID(dev, pri_node)
5110
      result = self.rpc.call_blockdev_find(pri_node, dev)
5111
      result.Raise()
5112
      if not result.data:
5113
        raise errors.OpExecError("Can't find disk/%d on node %s" %
5114
                                 (idx, pri_node))
5115

    
5116
    # Step: check other node consistency
5117
    self.proc.LogStep(2, steps_total, "check peer consistency")
5118
    for idx, dev in enumerate(instance.disks):
5119
      if idx not in self.op.disks:
5120
        continue
5121
      info("checking disk/%d consistency on %s" % (idx, pri_node))
5122
      if not _CheckDiskConsistency(self, dev, pri_node, True, ldisk=True):
5123
        raise errors.OpExecError("Primary node (%s) has degraded storage,"
5124
                                 " unsafe to replace the secondary" %
5125
                                 pri_node)
5126

    
5127
    # Step: create new storage
5128
    self.proc.LogStep(3, steps_total, "allocate new storage")
5129
    for idx, dev in enumerate(instance.disks):
5130
      info("adding new local storage on %s for disk/%d" %
5131
           (new_node, idx))
5132
      # we pass force_create=True to force LVM creation
5133
      for new_lv in dev.children:
5134
        _CreateBlockDev(self, new_node, instance, new_lv, True,
5135
                        _GetInstanceInfoText(instance), False)
5136

    
5137
    # Step 4: dbrd minors and drbd setups changes
5138
    # after this, we must manually remove the drbd minors on both the
5139
    # error and the success paths
5140
    minors = cfg.AllocateDRBDMinor([new_node for dev in instance.disks],
5141
                                   instance.name)
5142
    logging.debug("Allocated minors %s" % (minors,))
5143
    self.proc.LogStep(4, steps_total, "changing drbd configuration")
5144
    for idx, (dev, new_minor) in enumerate(zip(instance.disks, minors)):
5145
      size = dev.size
5146
      info("activating a new drbd on %s for disk/%d" % (new_node, idx))
5147
      # create new devices on new_node; note that we create two IDs:
5148
      # one without port, so the drbd will be activated without
5149
      # networking information on the new node at this stage, and one
5150
      # with network, for the latter activation in step 4
5151
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
5152
      if pri_node == o_node1:
5153
        p_minor = o_minor1
5154
      else:
5155
        p_minor = o_minor2
5156

    
5157
      new_alone_id = (pri_node, new_node, None, p_minor, new_minor, o_secret)
5158
      new_net_id = (pri_node, new_node, o_port, p_minor, new_minor, o_secret)
5159

    
5160
      iv_names[idx] = (dev, dev.children, new_net_id)
5161
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
5162
                    new_net_id)
5163
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
5164
                              logical_id=new_alone_id,
5165
                              children=dev.children)
5166
      try:
5167
        _CreateSingleBlockDev(self, new_node, instance, new_drbd,
5168
                              _GetInstanceInfoText(instance), False)
5169
      except errors.BlockDeviceError:
5170
        self.cfg.ReleaseDRBDMinors(instance.name)
5171
        raise
5172

    
5173
    for idx, dev in enumerate(instance.disks):
5174
      # we have new devices, shutdown the drbd on the old secondary
5175
      info("shutting down drbd for disk/%d on old node" % idx)
5176
      cfg.SetDiskID(dev, old_node)
5177
      result = self.rpc.call_blockdev_shutdown(old_node, dev)
5178
      if result.failed or not result.data:
5179
        warning("Failed to shutdown drbd for disk/%d on old node" % idx,
5180
                hint="Please cleanup this device manually as soon as possible")
5181

    
5182
    info("detaching primary drbds from the network (=> standalone)")
5183
    result = self.rpc.call_drbd_disconnect_net([pri_node], nodes_ip,
5184
                                               instance.disks)[pri_node]
5185

    
5186
    msg = result.RemoteFailMsg()
5187
    if msg:
5188
      # detaches didn't succeed (unlikely)
5189
      self.cfg.ReleaseDRBDMinors(instance.name)
5190
      raise errors.OpExecError("Can't detach the disks from the network on"
5191
                               " old node: %s" % (msg,))
5192

    
5193
    # if we managed to detach at least one, we update all the disks of
5194
    # the instance to point to the new secondary
5195
    info("updating instance configuration")
5196
    for dev, _, new_logical_id in iv_names.itervalues():
5197
      dev.logical_id = new_logical_id
5198
      cfg.SetDiskID(dev, pri_node)
5199
    cfg.Update(instance)
5200

    
5201
    # and now perform the drbd attach
5202
    info("attaching primary drbds to new secondary (standalone => connected)")
5203
    result = self.rpc.call_drbd_attach_net([pri_node, new_node], nodes_ip,
5204
                                           instance.disks, instance.name,
5205
                                           False)
5206
    for to_node, to_result in result.items():
5207
      msg = to_result.RemoteFailMsg()
5208
      if msg:
5209
        warning("can't attach drbd disks on node %s: %s", to_node, msg,
5210
                hint="please do a gnt-instance info to see the"
5211
                " status of disks")
5212

    
5213
    # this can fail as the old devices are degraded and _WaitForSync
5214
    # does a combined result over all disks, so we don't check its
5215
    # return value
5216
    self.proc.LogStep(5, steps_total, "sync devices")
5217
    _WaitForSync(self, instance, unlock=True)
5218

    
5219
    # so check manually all the devices
5220
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5221
      cfg.SetDiskID(dev, pri_node)
5222
      result = self.rpc.call_blockdev_find(pri_node, dev)
5223
      result.Raise()
5224
      if result.data[5]:
5225
        raise errors.OpExecError("DRBD device disk/%d is degraded!" % idx)
5226

    
5227
    self.proc.LogStep(6, steps_total, "removing old storage")
5228
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5229
      info("remove logical volumes for disk/%d" % idx)
5230
      for lv in old_lvs:
5231
        cfg.SetDiskID(lv, old_node)
5232
        result = self.rpc.call_blockdev_remove(old_node, lv)
5233
        if result.failed or not result.data:
5234
          warning("Can't remove LV on old secondary",
5235
                  hint="Cleanup stale volumes by hand")
5236

    
5237
  def Exec(self, feedback_fn):
5238
    """Execute disk replacement.
5239

5240
    This dispatches the disk replacement to the appropriate handler.
5241

5242
    """
5243
    instance = self.instance
5244

    
5245
    # Activate the instance disks if we're replacing them on a down instance
5246
    if not instance.admin_up:
5247
      _StartInstanceDisks(self, instance, True)
5248

    
5249
    if self.op.mode == constants.REPLACE_DISK_CHG:
5250
      fn = self._ExecD8Secondary
5251
    else:
5252
      fn = self._ExecD8DiskOnly
5253

    
5254
    ret = fn(feedback_fn)
5255

    
5256
    # Deactivate the instance disks if we're replacing them on a down instance
5257
    if not instance.admin_up:
5258
      _SafeShutdownInstanceDisks(self, instance)
5259

    
5260
    return ret
5261

    
5262

    
5263
class LUGrowDisk(LogicalUnit):
5264
  """Grow a disk of an instance.
5265

5266
  """
5267
  HPATH = "disk-grow"
5268
  HTYPE = constants.HTYPE_INSTANCE
5269
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
5270
  REQ_BGL = False
5271

    
5272
  def ExpandNames(self):
5273
    self._ExpandAndLockInstance()
5274
    self.needed_locks[locking.LEVEL_NODE] = []
5275
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5276

    
5277
  def DeclareLocks(self, level):
5278
    if level == locking.LEVEL_NODE:
5279
      self._LockInstancesNodes()
5280

    
5281
  def BuildHooksEnv(self):
5282
    """Build hooks env.
5283

5284
    This runs on the master, the primary and all the secondaries.
5285

5286
    """
5287
    env = {
5288
      "DISK": self.op.disk,
5289
      "AMOUNT": self.op.amount,
5290
      }
5291
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5292
    nl = [
5293
      self.cfg.GetMasterNode(),
5294
      self.instance.primary_node,
5295
      ]
5296
    return env, nl, nl
5297

    
5298
  def CheckPrereq(self):
5299
    """Check prerequisites.
5300

5301
    This checks that the instance is in the cluster.
5302

5303
    """
5304
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5305
    assert instance is not None, \
5306
      "Cannot retrieve locked instance %s" % self.op.instance_name
5307
    nodenames = list(instance.all_nodes)
5308
    for node in nodenames:
5309
      _CheckNodeOnline(self, node)
5310

    
5311

    
5312
    self.instance = instance
5313

    
5314
    if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8):
5315
      raise errors.OpPrereqError("Instance's disk layout does not support"
5316
                                 " growing.")
5317

    
5318
    self.disk = instance.FindDisk(self.op.disk)
5319

    
5320
    nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
5321
                                       instance.hypervisor)
5322
    for node in nodenames:
5323
      info = nodeinfo[node]
5324
      if info.failed or not info.data:
5325
        raise errors.OpPrereqError("Cannot get current information"
5326
                                   " from node '%s'" % node)
5327
      vg_free = info.data.get('vg_free', None)
5328
      if not isinstance(vg_free, int):
5329
        raise errors.OpPrereqError("Can't compute free disk space on"
5330
                                   " node %s" % node)
5331
      if self.op.amount > vg_free:
5332
        raise errors.OpPrereqError("Not enough disk space on target node %s:"
5333
                                   " %d MiB available, %d MiB required" %
5334
                                   (node, vg_free, self.op.amount))
5335

    
5336
  def Exec(self, feedback_fn):
5337
    """Execute disk grow.
5338

5339
    """
5340
    instance = self.instance
5341
    disk = self.disk
5342
    for node in instance.all_nodes:
5343
      self.cfg.SetDiskID(disk, node)
5344
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
5345
      result.Raise()
5346
      if (not result.data or not isinstance(result.data, (list, tuple)) or
5347
          len(result.data) != 2):
5348
        raise errors.OpExecError("Grow request failed to node %s" % node)
5349
      elif not result.data[0]:
5350
        raise errors.OpExecError("Grow request failed to node %s: %s" %
5351
                                 (node, result.data[1]))
5352
    disk.RecordGrow(self.op.amount)
5353
    self.cfg.Update(instance)
5354
    if self.op.wait_for_sync:
5355
      disk_abort = not _WaitForSync(self, instance)
5356
      if disk_abort:
5357
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
5358
                             " status.\nPlease check the instance.")
5359

    
5360

    
5361
class LUQueryInstanceData(NoHooksLU):
5362
  """Query runtime instance data.
5363

5364
  """
5365
  _OP_REQP = ["instances", "static"]
5366
  REQ_BGL = False
5367

    
5368
  def ExpandNames(self):
5369
    self.needed_locks = {}
5370
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
5371

    
5372
    if not isinstance(self.op.instances, list):
5373
      raise errors.OpPrereqError("Invalid argument type 'instances'")
5374

    
5375
    if self.op.instances:
5376
      self.wanted_names = []
5377
      for name in self.op.instances:
5378
        full_name = self.cfg.ExpandInstanceName(name)
5379
        if full_name is None:
5380
          raise errors.OpPrereqError("Instance '%s' not known" % name)
5381
        self.wanted_names.append(full_name)
5382
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
5383
    else:
5384
      self.wanted_names = None
5385
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5386

    
5387
    self.needed_locks[locking.LEVEL_NODE] = []
5388
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5389

    
5390
  def DeclareLocks(self, level):
5391
    if level == locking.LEVEL_NODE:
5392
      self._LockInstancesNodes()
5393

    
5394
  def CheckPrereq(self):
5395
    """Check prerequisites.
5396

5397
    This only checks the optional instance list against the existing names.
5398

5399
    """
5400
    if self.wanted_names is None:
5401
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
5402

    
5403
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
5404
                             in self.wanted_names]
5405
    return
5406

    
5407
  def _ComputeDiskStatus(self, instance, snode, dev):
5408
    """Compute block device status.
5409

5410
    """
5411
    static = self.op.static
5412
    if not static:
5413
      self.cfg.SetDiskID(dev, instance.primary_node)
5414
      dev_pstatus = self.rpc.call_blockdev_find(instance.primary_node, dev)
5415
      dev_pstatus.Raise()
5416
      dev_pstatus = dev_pstatus.data
5417
    else:
5418
      dev_pstatus = None
5419

    
5420
    if dev.dev_type in constants.LDS_DRBD:
5421
      # we change the snode then (otherwise we use the one passed in)
5422
      if dev.logical_id[0] == instance.primary_node:
5423
        snode = dev.logical_id[1]
5424
      else:
5425
        snode = dev.logical_id[0]
5426

    
5427
    if snode and not static:
5428
      self.cfg.SetDiskID(dev, snode)
5429
      dev_sstatus = self.rpc.call_blockdev_find(snode, dev)
5430
      dev_sstatus.Raise()
5431
      dev_sstatus = dev_sstatus.data
5432
    else:
5433
      dev_sstatus = None
5434

    
5435
    if dev.children:
5436
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
5437
                      for child in dev.children]
5438
    else:
5439
      dev_children = []
5440

    
5441
    data = {
5442
      "iv_name": dev.iv_name,
5443
      "dev_type": dev.dev_type,
5444
      "logical_id": dev.logical_id,
5445
      "physical_id": dev.physical_id,
5446
      "pstatus": dev_pstatus,
5447
      "sstatus": dev_sstatus,
5448
      "children": dev_children,
5449
      "mode": dev.mode,
5450
      }
5451

    
5452
    return data
5453

    
5454
  def Exec(self, feedback_fn):
5455
    """Gather and return data"""
5456
    result = {}
5457

    
5458
    cluster = self.cfg.GetClusterInfo()
5459

    
5460
    for instance in self.wanted_instances:
5461
      if not self.op.static:
5462
        remote_info = self.rpc.call_instance_info(instance.primary_node,
5463
                                                  instance.name,
5464
                                                  instance.hypervisor)
5465
        remote_info.Raise()
5466
        remote_info = remote_info.data
5467
        if remote_info and "state" in remote_info:
5468
          remote_state = "up"
5469
        else:
5470
          remote_state = "down"
5471
      else:
5472
        remote_state = None
5473
      if instance.admin_up:
5474
        config_state = "up"
5475
      else:
5476
        config_state = "down"
5477

    
5478
      disks = [self._ComputeDiskStatus(instance, None, device)
5479
               for device in instance.disks]
5480

    
5481
      idict = {
5482
        "name": instance.name,
5483
        "config_state": config_state,
5484
        "run_state": remote_state,
5485
        "pnode": instance.primary_node,
5486
        "snodes": instance.secondary_nodes,
5487
        "os": instance.os,
5488
        "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
5489
        "disks": disks,
5490
        "hypervisor": instance.hypervisor,
5491
        "network_port": instance.network_port,
5492
        "hv_instance": instance.hvparams,
5493
        "hv_actual": cluster.FillHV(instance),
5494
        "be_instance": instance.beparams,
5495
        "be_actual": cluster.FillBE(instance),
5496
        }
5497

    
5498
      result[instance.name] = idict
5499

    
5500
    return result
5501

    
5502

    
5503
class LUSetInstanceParams(LogicalUnit):
5504
  """Modifies an instances's parameters.
5505

5506
  """
5507
  HPATH = "instance-modify"
5508
  HTYPE = constants.HTYPE_INSTANCE
5509
  _OP_REQP = ["instance_name"]
5510
  REQ_BGL = False
5511

    
5512
  def CheckArguments(self):
5513
    if not hasattr(self.op, 'nics'):
5514
      self.op.nics = []
5515
    if not hasattr(self.op, 'disks'):
5516
      self.op.disks = []
5517
    if not hasattr(self.op, 'beparams'):
5518
      self.op.beparams = {}
5519
    if not hasattr(self.op, 'hvparams'):
5520
      self.op.hvparams = {}
5521
    self.op.force = getattr(self.op, "force", False)
5522
    if not (self.op.nics or self.op.disks or
5523
            self.op.hvparams or self.op.beparams):
5524
      raise errors.OpPrereqError("No changes submitted")
5525

    
5526
    utils.CheckBEParams(self.op.beparams)
5527

    
5528
    # Disk validation
5529
    disk_addremove = 0
5530
    for disk_op, disk_dict in self.op.disks:
5531
      if disk_op == constants.DDM_REMOVE:
5532
        disk_addremove += 1
5533
        continue
5534
      elif disk_op == constants.DDM_ADD:
5535
        disk_addremove += 1
5536
      else:
5537
        if not isinstance(disk_op, int):
5538
          raise errors.OpPrereqError("Invalid disk index")
5539
      if disk_op == constants.DDM_ADD:
5540
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
5541
        if mode not in constants.DISK_ACCESS_SET:
5542
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
5543
        size = disk_dict.get('size', None)
5544
        if size is None:
5545
          raise errors.OpPrereqError("Required disk parameter size missing")
5546
        try:
5547
          size = int(size)
5548
        except ValueError, err:
5549
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
5550
                                     str(err))
5551
        disk_dict['size'] = size
5552
      else:
5553
        # modification of disk
5554
        if 'size' in disk_dict:
5555
          raise errors.OpPrereqError("Disk size change not possible, use"
5556
                                     " grow-disk")
5557

    
5558
    if disk_addremove > 1:
5559
      raise errors.OpPrereqError("Only one disk add or remove operation"
5560
                                 " supported at a time")
5561

    
5562
    # NIC validation
5563
    nic_addremove = 0
5564
    for nic_op, nic_dict in self.op.nics:
5565
      if nic_op == constants.DDM_REMOVE:
5566
        nic_addremove += 1
5567
        continue
5568
      elif nic_op == constants.DDM_ADD:
5569
        nic_addremove += 1
5570
      else:
5571
        if not isinstance(nic_op, int):
5572
          raise errors.OpPrereqError("Invalid nic index")
5573

    
5574
      # nic_dict should be a dict
5575
      nic_ip = nic_dict.get('ip', None)
5576
      if nic_ip is not None:
5577
        if nic_ip.lower() == "none":
5578
          nic_dict['ip'] = None
5579
        else:
5580
          if not utils.IsValidIP(nic_ip):
5581
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
5582
      # we can only check None bridges and assign the default one
5583
      nic_bridge = nic_dict.get('bridge', None)
5584
      if nic_bridge is None:
5585
        nic_dict['bridge'] = self.cfg.GetDefBridge()
5586
      # but we can validate MACs
5587
      nic_mac = nic_dict.get('mac', None)
5588
      if nic_mac is not None:
5589
        if self.cfg.IsMacInUse(nic_mac):
5590
          raise errors.OpPrereqError("MAC address %s already in use"
5591
                                     " in cluster" % nic_mac)
5592
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5593
          if not utils.IsValidMac(nic_mac):
5594
            raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
5595
    if nic_addremove > 1:
5596
      raise errors.OpPrereqError("Only one NIC add or remove operation"
5597
                                 " supported at a time")
5598

    
5599
  def ExpandNames(self):
5600
    self._ExpandAndLockInstance()
5601
    self.needed_locks[locking.LEVEL_NODE] = []
5602
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5603

    
5604
  def DeclareLocks(self, level):
5605
    if level == locking.LEVEL_NODE:
5606
      self._LockInstancesNodes()
5607

    
5608
  def BuildHooksEnv(self):
5609
    """Build hooks env.
5610

5611
    This runs on the master, primary and secondaries.
5612

5613
    """
5614
    args = dict()
5615
    if constants.BE_MEMORY in self.be_new:
5616
      args['memory'] = self.be_new[constants.BE_MEMORY]
5617
    if constants.BE_VCPUS in self.be_new:
5618
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
5619
    # FIXME: readd disk/nic changes
5620
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
5621
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5622
    return env, nl, nl
5623

    
5624
  def CheckPrereq(self):
5625
    """Check prerequisites.
5626

5627
    This only checks the instance list against the existing names.
5628

5629
    """
5630
    force = self.force = self.op.force
5631

    
5632
    # checking the new params on the primary/secondary nodes
5633

    
5634
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5635
    assert self.instance is not None, \
5636
      "Cannot retrieve locked instance %s" % self.op.instance_name
5637
    pnode = instance.primary_node
5638
    nodelist = list(instance.all_nodes)
5639

    
5640
    # hvparams processing
5641
    if self.op.hvparams:
5642
      i_hvdict = copy.deepcopy(instance.hvparams)
5643
      for key, val in self.op.hvparams.iteritems():
5644
        if val == constants.VALUE_DEFAULT:
5645
          try:
5646
            del i_hvdict[key]
5647
          except KeyError:
5648
            pass
5649
        elif val == constants.VALUE_NONE:
5650
          i_hvdict[key] = None
5651
        else:
5652
          i_hvdict[key] = val
5653
      cluster = self.cfg.GetClusterInfo()
5654
      hv_new = cluster.FillDict(cluster.hvparams[instance.hypervisor],
5655
                                i_hvdict)
5656
      # local check
5657
      hypervisor.GetHypervisor(
5658
        instance.hypervisor).CheckParameterSyntax(hv_new)
5659
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
5660
      self.hv_new = hv_new # the new actual values
5661
      self.hv_inst = i_hvdict # the new dict (without defaults)
5662
    else:
5663
      self.hv_new = self.hv_inst = {}
5664

    
5665
    # beparams processing
5666
    if self.op.beparams:
5667
      i_bedict = copy.deepcopy(instance.beparams)
5668
      for key, val in self.op.beparams.iteritems():
5669
        if val == constants.VALUE_DEFAULT:
5670
          try:
5671
            del i_bedict[key]
5672
          except KeyError:
5673
            pass
5674
        else:
5675
          i_bedict[key] = val
5676
      cluster = self.cfg.GetClusterInfo()
5677
      be_new = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
5678
                                i_bedict)
5679
      self.be_new = be_new # the new actual values
5680
      self.be_inst = i_bedict # the new dict (without defaults)
5681
    else:
5682
      self.be_new = self.be_inst = {}
5683

    
5684
    self.warn = []
5685

    
5686
    if constants.BE_MEMORY in self.op.beparams and not self.force:
5687
      mem_check_list = [pnode]
5688
      if be_new[constants.BE_AUTO_BALANCE]:
5689
        # either we changed auto_balance to yes or it was from before
5690
        mem_check_list.extend(instance.secondary_nodes)
5691
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
5692
                                                  instance.hypervisor)
5693
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
5694
                                         instance.hypervisor)
5695
      if nodeinfo[pnode].failed or not isinstance(nodeinfo[pnode].data, dict):
5696
        # Assume the primary node is unreachable and go ahead
5697
        self.warn.append("Can't get info from primary node %s" % pnode)
5698
      else:
5699
        if not instance_info.failed and instance_info.data:
5700
          current_mem = instance_info.data['memory']
5701
        else:
5702
          # Assume instance not running
5703
          # (there is a slight race condition here, but it's not very probable,
5704
          # and we have no other way to check)
5705
          current_mem = 0
5706
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
5707
                    nodeinfo[pnode].data['memory_free'])
5708
        if miss_mem > 0:
5709
          raise errors.OpPrereqError("This change will prevent the instance"
5710
                                     " from starting, due to %d MB of memory"
5711
                                     " missing on its primary node" % miss_mem)
5712

    
5713
      if be_new[constants.BE_AUTO_BALANCE]:
5714
        for node, nres in nodeinfo.iteritems():
5715
          if node not in instance.secondary_nodes:
5716
            continue
5717
          if nres.failed or not isinstance(nres.data, dict):
5718
            self.warn.append("Can't get info from secondary node %s" % node)
5719
          elif be_new[constants.BE_MEMORY] > nres.data['memory_free']:
5720
            self.warn.append("Not enough memory to failover instance to"
5721
                             " secondary node %s" % node)
5722

    
5723
    # NIC processing
5724
    for nic_op, nic_dict in self.op.nics:
5725
      if nic_op == constants.DDM_REMOVE:
5726
        if not instance.nics:
5727
          raise errors.OpPrereqError("Instance has no NICs, cannot remove")
5728
        continue
5729
      if nic_op != constants.DDM_ADD:
5730
        # an existing nic
5731
        if nic_op < 0 or nic_op >= len(instance.nics):
5732
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
5733
                                     " are 0 to %d" %
5734
                                     (nic_op, len(instance.nics)))
5735
      nic_bridge = nic_dict.get('bridge', None)
5736
      if nic_bridge is not None:
5737
        if not self.rpc.call_bridges_exist(pnode, [nic_bridge]):
5738
          msg = ("Bridge '%s' doesn't exist on one of"
5739
                 " the instance nodes" % nic_bridge)
5740
          if self.force:
5741
            self.warn.append(msg)
5742
          else:
5743
            raise errors.OpPrereqError(msg)
5744

    
5745
    # DISK processing
5746
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
5747
      raise errors.OpPrereqError("Disk operations not supported for"
5748
                                 " diskless instances")
5749
    for disk_op, disk_dict in self.op.disks:
5750
      if disk_op == constants.DDM_REMOVE:
5751
        if len(instance.disks) == 1:
5752
          raise errors.OpPrereqError("Cannot remove the last disk of"
5753
                                     " an instance")
5754
        ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
5755
        ins_l = ins_l[pnode]
5756
        if ins_l.failed or not isinstance(ins_l.data, list):
5757
          raise errors.OpPrereqError("Can't contact node '%s'" % pnode)
5758
        if instance.name in ins_l.data:
5759
          raise errors.OpPrereqError("Instance is running, can't remove"
5760
                                     " disks.")
5761

    
5762
      if (disk_op == constants.DDM_ADD and
5763
          len(instance.nics) >= constants.MAX_DISKS):
5764
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
5765
                                   " add more" % constants.MAX_DISKS)
5766
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
5767
        # an existing disk
5768
        if disk_op < 0 or disk_op >= len(instance.disks):
5769
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
5770
                                     " are 0 to %d" %
5771
                                     (disk_op, len(instance.disks)))
5772

    
5773
    return
5774

    
5775
  def Exec(self, feedback_fn):
5776
    """Modifies an instance.
5777

5778
    All parameters take effect only at the next restart of the instance.
5779

5780
    """
5781
    # Process here the warnings from CheckPrereq, as we don't have a
5782
    # feedback_fn there.
5783
    for warn in self.warn:
5784
      feedback_fn("WARNING: %s" % warn)
5785

    
5786
    result = []
5787
    instance = self.instance
5788
    # disk changes
5789
    for disk_op, disk_dict in self.op.disks:
5790
      if disk_op == constants.DDM_REMOVE:
5791
        # remove the last disk
5792
        device = instance.disks.pop()
5793
        device_idx = len(instance.disks)
5794
        for node, disk in device.ComputeNodeTree(instance.primary_node):
5795
          self.cfg.SetDiskID(disk, node)
5796
          rpc_result = self.rpc.call_blockdev_remove(node, disk)
5797
          if rpc_result.failed or not rpc_result.data:
5798
            self.proc.LogWarning("Could not remove disk/%d on node %s,"
5799
                                 " continuing anyway", device_idx, node)
5800
        result.append(("disk/%d" % device_idx, "remove"))
5801
      elif disk_op == constants.DDM_ADD:
5802
        # add a new disk
5803
        if instance.disk_template == constants.DT_FILE:
5804
          file_driver, file_path = instance.disks[0].logical_id
5805
          file_path = os.path.dirname(file_path)
5806
        else:
5807
          file_driver = file_path = None
5808
        disk_idx_base = len(instance.disks)
5809
        new_disk = _GenerateDiskTemplate(self,
5810
                                         instance.disk_template,
5811
                                         instance.name, instance.primary_node,
5812
                                         instance.secondary_nodes,
5813
                                         [disk_dict],
5814
                                         file_path,
5815
                                         file_driver,
5816
                                         disk_idx_base)[0]
5817
        instance.disks.append(new_disk)
5818
        info = _GetInstanceInfoText(instance)
5819

    
5820
        logging.info("Creating volume %s for instance %s",
5821
                     new_disk.iv_name, instance.name)
5822
        # Note: this needs to be kept in sync with _CreateDisks
5823
        #HARDCODE
5824
        for node in instance.all_nodes:
5825
          f_create = node == instance.primary_node
5826
          try:
5827
            _CreateBlockDev(self, node, instance, new_disk,
5828
                            f_create, info, f_create)
5829
          except errors.OpExecError, err:
5830
            self.LogWarning("Failed to create volume %s (%s) on"
5831
                            " node %s: %s",
5832
                            new_disk.iv_name, new_disk, node, err)
5833
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
5834
                       (new_disk.size, new_disk.mode)))
5835
      else:
5836
        # change a given disk
5837
        instance.disks[disk_op].mode = disk_dict['mode']
5838
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
5839
    # NIC changes
5840
    for nic_op, nic_dict in self.op.nics:
5841
      if nic_op == constants.DDM_REMOVE:
5842
        # remove the last nic
5843
        del instance.nics[-1]
5844
        result.append(("nic.%d" % len(instance.nics), "remove"))
5845
      elif nic_op == constants.DDM_ADD:
5846
        # add a new nic
5847
        if 'mac' not in nic_dict:
5848
          mac = constants.VALUE_GENERATE
5849
        else:
5850
          mac = nic_dict['mac']
5851
        if mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5852
          mac = self.cfg.GenerateMAC()
5853
        new_nic = objects.NIC(mac=mac, ip=nic_dict.get('ip', None),
5854
                              bridge=nic_dict.get('bridge', None))
5855
        instance.nics.append(new_nic)
5856
        result.append(("nic.%d" % (len(instance.nics) - 1),
5857
                       "add:mac=%s,ip=%s,bridge=%s" %
5858
                       (new_nic.mac, new_nic.ip, new_nic.bridge)))
5859
      else:
5860
        # change a given nic
5861
        for key in 'mac', 'ip', 'bridge':
5862
          if key in nic_dict:
5863
            setattr(instance.nics[nic_op], key, nic_dict[key])
5864
            result.append(("nic.%s/%d" % (key, nic_op), nic_dict[key]))
5865

    
5866
    # hvparams changes
5867
    if self.op.hvparams:
5868
      instance.hvparams = self.hv_new
5869
      for key, val in self.op.hvparams.iteritems():
5870
        result.append(("hv/%s" % key, val))
5871

    
5872
    # beparams changes
5873
    if self.op.beparams:
5874
      instance.beparams = self.be_inst
5875
      for key, val in self.op.beparams.iteritems():
5876
        result.append(("be/%s" % key, val))
5877

    
5878
    self.cfg.Update(instance)
5879

    
5880
    return result
5881

    
5882

    
5883
class LUQueryExports(NoHooksLU):
5884
  """Query the exports list
5885

5886
  """
5887
  _OP_REQP = ['nodes']
5888
  REQ_BGL = False
5889

    
5890
  def ExpandNames(self):
5891
    self.needed_locks = {}
5892
    self.share_locks[locking.LEVEL_NODE] = 1
5893
    if not self.op.nodes:
5894
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5895
    else:
5896
      self.needed_locks[locking.LEVEL_NODE] = \
5897
        _GetWantedNodes(self, self.op.nodes)
5898

    
5899
  def CheckPrereq(self):
5900
    """Check prerequisites.
5901

5902
    """
5903
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
5904

    
5905
  def Exec(self, feedback_fn):
5906
    """Compute the list of all the exported system images.
5907

5908
    @rtype: dict
5909
    @return: a dictionary with the structure node->(export-list)
5910
        where export-list is a list of the instances exported on
5911
        that node.
5912

5913
    """
5914
    rpcresult = self.rpc.call_export_list(self.nodes)
5915
    result = {}
5916
    for node in rpcresult:
5917
      if rpcresult[node].failed:
5918
        result[node] = False
5919
      else:
5920
        result[node] = rpcresult[node].data
5921

    
5922
    return result
5923

    
5924

    
5925
class LUExportInstance(LogicalUnit):
5926
  """Export an instance to an image in the cluster.
5927

5928
  """
5929
  HPATH = "instance-export"
5930
  HTYPE = constants.HTYPE_INSTANCE
5931
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
5932
  REQ_BGL = False
5933

    
5934
  def ExpandNames(self):
5935
    self._ExpandAndLockInstance()
5936
    # FIXME: lock only instance primary and destination node
5937
    #
5938
    # Sad but true, for now we have do lock all nodes, as we don't know where
5939
    # the previous export might be, and and in this LU we search for it and
5940
    # remove it from its current node. In the future we could fix this by:
5941
    #  - making a tasklet to search (share-lock all), then create the new one,
5942
    #    then one to remove, after
5943
    #  - removing the removal operation altoghether
5944
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5945

    
5946
  def DeclareLocks(self, level):
5947
    """Last minute lock declaration."""
5948
    # All nodes are locked anyway, so nothing to do here.
5949

    
5950
  def BuildHooksEnv(self):
5951
    """Build hooks env.
5952

5953
    This will run on the master, primary node and target node.
5954

5955
    """
5956
    env = {
5957
      "EXPORT_NODE": self.op.target_node,
5958
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
5959
      }
5960
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5961
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
5962
          self.op.target_node]
5963
    return env, nl, nl
5964

    
5965
  def CheckPrereq(self):
5966
    """Check prerequisites.
5967

5968
    This checks that the instance and node names are valid.
5969

5970
    """
5971
    instance_name = self.op.instance_name
5972
    self.instance = self.cfg.GetInstanceInfo(instance_name)
5973
    assert self.instance is not None, \
5974
          "Cannot retrieve locked instance %s" % self.op.instance_name
5975
    _CheckNodeOnline(self, self.instance.primary_node)
5976

    
5977
    self.dst_node = self.cfg.GetNodeInfo(
5978
      self.cfg.ExpandNodeName(self.op.target_node))
5979

    
5980
    if self.dst_node is None:
5981
      # This is wrong node name, not a non-locked node
5982
      raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
5983
    _CheckNodeOnline(self, self.dst_node.name)
5984

    
5985
    # instance disk type verification
5986
    for disk in self.instance.disks:
5987
      if disk.dev_type == constants.LD_FILE:
5988
        raise errors.OpPrereqError("Export not supported for instances with"
5989
                                   " file-based disks")
5990

    
5991
  def Exec(self, feedback_fn):
5992
    """Export an instance to an image in the cluster.
5993

5994
    """
5995
    instance = self.instance
5996
    dst_node = self.dst_node
5997
    src_node = instance.primary_node
5998
    if self.op.shutdown:
5999
      # shutdown the instance, but not the disks
6000
      result = self.rpc.call_instance_shutdown(src_node, instance)
6001
      result.Raise()
6002
      if not result.data:
6003
        raise errors.OpExecError("Could not shutdown instance %s on node %s" %
6004
                                 (instance.name, src_node))
6005

    
6006
    vgname = self.cfg.GetVGName()
6007

    
6008
    snap_disks = []
6009

    
6010
    # set the disks ID correctly since call_instance_start needs the
6011
    # correct drbd minor to create the symlinks
6012
    for disk in instance.disks:
6013
      self.cfg.SetDiskID(disk, src_node)
6014

    
6015
    try:
6016
      for disk in instance.disks:
6017
        # new_dev_name will be a snapshot of an lvm leaf of the one we passed
6018
        new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
6019
        if new_dev_name.failed or not new_dev_name.data:
6020
          self.LogWarning("Could not snapshot block device %s on node %s",
6021
                          disk.logical_id[1], src_node)
6022
          snap_disks.append(False)
6023
        else:
6024
          new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
6025
                                 logical_id=(vgname, new_dev_name.data),
6026
                                 physical_id=(vgname, new_dev_name.data),
6027
                                 iv_name=disk.iv_name)
6028
          snap_disks.append(new_dev)
6029

    
6030
    finally:
6031
      if self.op.shutdown and instance.admin_up:
6032
        result = self.rpc.call_instance_start(src_node, instance, None)
6033
        msg = result.RemoteFailMsg()
6034
        if msg:
6035
          _ShutdownInstanceDisks(self, instance)
6036
          raise errors.OpExecError("Could not start instance: %s" % msg)
6037

    
6038
    # TODO: check for size
6039

    
6040
    cluster_name = self.cfg.GetClusterName()
6041
    for idx, dev in enumerate(snap_disks):
6042
      if dev:
6043
        result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
6044
                                               instance, cluster_name, idx)
6045
        if result.failed or not result.data:
6046
          self.LogWarning("Could not export block device %s from node %s to"
6047
                          " node %s", dev.logical_id[1], src_node,
6048
                          dst_node.name)
6049
        result = self.rpc.call_blockdev_remove(src_node, dev)
6050
        if result.failed or not result.data:
6051
          self.LogWarning("Could not remove snapshot block device %s from node"
6052
                          " %s", dev.logical_id[1], src_node)
6053

    
6054
    result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
6055
    if result.failed or not result.data:
6056
      self.LogWarning("Could not finalize export for instance %s on node %s",
6057
                      instance.name, dst_node.name)
6058

    
6059
    nodelist = self.cfg.GetNodeList()
6060
    nodelist.remove(dst_node.name)
6061

    
6062
    # on one-node clusters nodelist will be empty after the removal
6063
    # if we proceed the backup would be removed because OpQueryExports
6064
    # substitutes an empty list with the full cluster node list.
6065
    if nodelist:
6066
      exportlist = self.rpc.call_export_list(nodelist)
6067
      for node in exportlist:
6068
        if exportlist[node].failed:
6069
          continue
6070
        if instance.name in exportlist[node].data:
6071
          if not self.rpc.call_export_remove(node, instance.name):
6072
            self.LogWarning("Could not remove older export for instance %s"
6073
                            " on node %s", instance.name, node)
6074

    
6075

    
6076
class LURemoveExport(NoHooksLU):
6077
  """Remove exports related to the named instance.
6078

6079
  """
6080
  _OP_REQP = ["instance_name"]
6081
  REQ_BGL = False
6082

    
6083
  def ExpandNames(self):
6084
    self.needed_locks = {}
6085
    # We need all nodes to be locked in order for RemoveExport to work, but we
6086
    # don't need to lock the instance itself, as nothing will happen to it (and
6087
    # we can remove exports also for a removed instance)
6088
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6089

    
6090
  def CheckPrereq(self):
6091
    """Check prerequisites.
6092
    """
6093
    pass
6094

    
6095
  def Exec(self, feedback_fn):
6096
    """Remove any export.
6097

6098
    """
6099
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
6100
    # If the instance was not found we'll try with the name that was passed in.
6101
    # This will only work if it was an FQDN, though.
6102
    fqdn_warn = False
6103
    if not instance_name:
6104
      fqdn_warn = True
6105
      instance_name = self.op.instance_name
6106

    
6107
    exportlist = self.rpc.call_export_list(self.acquired_locks[
6108
      locking.LEVEL_NODE])
6109
    found = False
6110
    for node in exportlist:
6111
      if exportlist[node].failed:
6112
        self.LogWarning("Failed to query node %s, continuing" % node)
6113
        continue
6114
      if instance_name in exportlist[node].data:
6115
        found = True
6116
        result = self.rpc.call_export_remove(node, instance_name)
6117
        if result.failed or not result.data:
6118
          logging.error("Could not remove export for instance %s"
6119
                        " on node %s", instance_name, node)
6120

    
6121
    if fqdn_warn and not found:
6122
      feedback_fn("Export not found. If trying to remove an export belonging"
6123
                  " to a deleted instance please use its Fully Qualified"
6124
                  " Domain Name.")
6125

    
6126

    
6127
class TagsLU(NoHooksLU):
6128
  """Generic tags LU.
6129

6130
  This is an abstract class which is the parent of all the other tags LUs.
6131

6132
  """
6133

    
6134
  def ExpandNames(self):
6135
    self.needed_locks = {}
6136
    if self.op.kind == constants.TAG_NODE:
6137
      name = self.cfg.ExpandNodeName(self.op.name)
6138
      if name is None:
6139
        raise errors.OpPrereqError("Invalid node name (%s)" %
6140
                                   (self.op.name,))
6141
      self.op.name = name
6142
      self.needed_locks[locking.LEVEL_NODE] = name
6143
    elif self.op.kind == constants.TAG_INSTANCE:
6144
      name = self.cfg.ExpandInstanceName(self.op.name)
6145
      if name is None:
6146
        raise errors.OpPrereqError("Invalid instance name (%s)" %
6147
                                   (self.op.name,))
6148
      self.op.name = name
6149
      self.needed_locks[locking.LEVEL_INSTANCE] = name
6150

    
6151
  def CheckPrereq(self):
6152
    """Check prerequisites.
6153

6154
    """
6155
    if self.op.kind == constants.TAG_CLUSTER:
6156
      self.target = self.cfg.GetClusterInfo()
6157
    elif self.op.kind == constants.TAG_NODE:
6158
      self.target = self.cfg.GetNodeInfo(self.op.name)
6159
    elif self.op.kind == constants.TAG_INSTANCE:
6160
      self.target = self.cfg.GetInstanceInfo(self.op.name)
6161
    else:
6162
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
6163
                                 str(self.op.kind))
6164

    
6165

    
6166
class LUGetTags(TagsLU):
6167
  """Returns the tags of a given object.
6168

6169
  """
6170
  _OP_REQP = ["kind", "name"]
6171
  REQ_BGL = False
6172

    
6173
  def Exec(self, feedback_fn):
6174
    """Returns the tag list.
6175

6176
    """
6177
    return list(self.target.GetTags())
6178

    
6179

    
6180
class LUSearchTags(NoHooksLU):
6181
  """Searches the tags for a given pattern.
6182

6183
  """
6184
  _OP_REQP = ["pattern"]
6185
  REQ_BGL = False
6186

    
6187
  def ExpandNames(self):
6188
    self.needed_locks = {}
6189

    
6190
  def CheckPrereq(self):
6191
    """Check prerequisites.
6192

6193
    This checks the pattern passed for validity by compiling it.
6194

6195
    """
6196
    try:
6197
      self.re = re.compile(self.op.pattern)
6198
    except re.error, err:
6199
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
6200
                                 (self.op.pattern, err))
6201

    
6202
  def Exec(self, feedback_fn):
6203
    """Returns the tag list.
6204

6205
    """
6206
    cfg = self.cfg
6207
    tgts = [("/cluster", cfg.GetClusterInfo())]
6208
    ilist = cfg.GetAllInstancesInfo().values()
6209
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
6210
    nlist = cfg.GetAllNodesInfo().values()
6211
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
6212
    results = []
6213
    for path, target in tgts:
6214
      for tag in target.GetTags():
6215
        if self.re.search(tag):
6216
          results.append((path, tag))
6217
    return results
6218

    
6219

    
6220
class LUAddTags(TagsLU):
6221
  """Sets a tag on a given object.
6222

6223
  """
6224
  _OP_REQP = ["kind", "name", "tags"]
6225
  REQ_BGL = False
6226

    
6227
  def CheckPrereq(self):
6228
    """Check prerequisites.
6229

6230
    This checks the type and length of the tag name and value.
6231

6232
    """
6233
    TagsLU.CheckPrereq(self)
6234
    for tag in self.op.tags:
6235
      objects.TaggableObject.ValidateTag(tag)
6236

    
6237
  def Exec(self, feedback_fn):
6238
    """Sets the tag.
6239

6240
    """
6241
    try:
6242
      for tag in self.op.tags:
6243
        self.target.AddTag(tag)
6244
    except errors.TagError, err:
6245
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
6246
    try:
6247
      self.cfg.Update(self.target)
6248
    except errors.ConfigurationError:
6249
      raise errors.OpRetryError("There has been a modification to the"
6250
                                " config file and the operation has been"
6251
                                " aborted. Please retry.")
6252

    
6253

    
6254
class LUDelTags(TagsLU):
6255
  """Delete a list of tags from a given object.
6256

6257
  """
6258
  _OP_REQP = ["kind", "name", "tags"]
6259
  REQ_BGL = False
6260

    
6261
  def CheckPrereq(self):
6262
    """Check prerequisites.
6263

6264
    This checks that we have the given tag.
6265

6266
    """
6267
    TagsLU.CheckPrereq(self)
6268
    for tag in self.op.tags:
6269
      objects.TaggableObject.ValidateTag(tag)
6270
    del_tags = frozenset(self.op.tags)
6271
    cur_tags = self.target.GetTags()
6272
    if not del_tags <= cur_tags:
6273
      diff_tags = del_tags - cur_tags
6274
      diff_names = ["'%s'" % tag for tag in diff_tags]
6275
      diff_names.sort()
6276
      raise errors.OpPrereqError("Tag(s) %s not found" %
6277
                                 (",".join(diff_names)))
6278

    
6279
  def Exec(self, feedback_fn):
6280
    """Remove the tag from the object.
6281

6282
    """
6283
    for tag in self.op.tags:
6284
      self.target.RemoveTag(tag)
6285
    try:
6286
      self.cfg.Update(self.target)
6287
    except errors.ConfigurationError:
6288
      raise errors.OpRetryError("There has been a modification to the"
6289
                                " config file and the operation has been"
6290
                                " aborted. Please retry.")
6291

    
6292

    
6293
class LUTestDelay(NoHooksLU):
6294
  """Sleep for a specified amount of time.
6295

6296
  This LU sleeps on the master and/or nodes for a specified amount of
6297
  time.
6298

6299
  """
6300
  _OP_REQP = ["duration", "on_master", "on_nodes"]
6301
  REQ_BGL = False
6302

    
6303
  def ExpandNames(self):
6304
    """Expand names and set required locks.
6305

6306
    This expands the node list, if any.
6307

6308
    """
6309
    self.needed_locks = {}
6310
    if self.op.on_nodes:
6311
      # _GetWantedNodes can be used here, but is not always appropriate to use
6312
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
6313
      # more information.
6314
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
6315
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
6316

    
6317
  def CheckPrereq(self):
6318
    """Check prerequisites.
6319

6320
    """
6321

    
6322
  def Exec(self, feedback_fn):
6323
    """Do the actual sleep.
6324

6325
    """
6326
    if self.op.on_master:
6327
      if not utils.TestDelay(self.op.duration):
6328
        raise errors.OpExecError("Error during master delay test")
6329
    if self.op.on_nodes:
6330
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
6331
      if not result:
6332
        raise errors.OpExecError("Complete failure from rpc call")
6333
      for node, node_result in result.items():
6334
        node_result.Raise()
6335
        if not node_result.data:
6336
          raise errors.OpExecError("Failure during rpc call to node %s,"
6337
                                   " result: %s" % (node, node_result.data))
6338

    
6339

    
6340
class IAllocator(object):
6341
  """IAllocator framework.
6342

6343
  An IAllocator instance has three sets of attributes:
6344
    - cfg that is needed to query the cluster
6345
    - input data (all members of the _KEYS class attribute are required)
6346
    - four buffer attributes (in|out_data|text), that represent the
6347
      input (to the external script) in text and data structure format,
6348
      and the output from it, again in two formats
6349
    - the result variables from the script (success, info, nodes) for
6350
      easy usage
6351

6352
  """
6353
  _ALLO_KEYS = [
6354
    "mem_size", "disks", "disk_template",
6355
    "os", "tags", "nics", "vcpus", "hypervisor",
6356
    ]
6357
  _RELO_KEYS = [
6358
    "relocate_from",
6359
    ]
6360

    
6361
  def __init__(self, lu, mode, name, **kwargs):
6362
    self.lu = lu
6363
    # init buffer variables
6364
    self.in_text = self.out_text = self.in_data = self.out_data = None
6365
    # init all input fields so that pylint is happy
6366
    self.mode = mode
6367
    self.name = name
6368
    self.mem_size = self.disks = self.disk_template = None
6369
    self.os = self.tags = self.nics = self.vcpus = None
6370
    self.hypervisor = None
6371
    self.relocate_from = None
6372
    # computed fields
6373
    self.required_nodes = None
6374
    # init result fields
6375
    self.success = self.info = self.nodes = None
6376
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6377
      keyset = self._ALLO_KEYS
6378
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6379
      keyset = self._RELO_KEYS
6380
    else:
6381
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
6382
                                   " IAllocator" % self.mode)
6383
    for key in kwargs:
6384
      if key not in keyset:
6385
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
6386
                                     " IAllocator" % key)
6387
      setattr(self, key, kwargs[key])
6388
    for key in keyset:
6389
      if key not in kwargs:
6390
        raise errors.ProgrammerError("Missing input parameter '%s' to"
6391
                                     " IAllocator" % key)
6392
    self._BuildInputData()
6393

    
6394
  def _ComputeClusterData(self):
6395
    """Compute the generic allocator input data.
6396

6397
    This is the data that is independent of the actual operation.
6398

6399
    """
6400
    cfg = self.lu.cfg
6401
    cluster_info = cfg.GetClusterInfo()
6402
    # cluster data
6403
    data = {
6404
      "version": 1,
6405
      "cluster_name": cfg.GetClusterName(),
6406
      "cluster_tags": list(cluster_info.GetTags()),
6407
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
6408
      # we don't have job IDs
6409
      }
6410
    iinfo = cfg.GetAllInstancesInfo().values()
6411
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
6412

    
6413
    # node data
6414
    node_results = {}
6415
    node_list = cfg.GetNodeList()
6416

    
6417
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6418
      hypervisor_name = self.hypervisor
6419
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6420
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
6421

    
6422
    node_data = self.lu.rpc.call_node_info(node_list, cfg.GetVGName(),
6423
                                           hypervisor_name)
6424
    node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
6425
                       cluster_info.enabled_hypervisors)
6426
    for nname, nresult in node_data.items():
6427
      # first fill in static (config-based) values
6428
      ninfo = cfg.GetNodeInfo(nname)
6429
      pnr = {
6430
        "tags": list(ninfo.GetTags()),
6431
        "primary_ip": ninfo.primary_ip,
6432
        "secondary_ip": ninfo.secondary_ip,
6433
        "offline": ninfo.offline,
6434
        "master_candidate": ninfo.master_candidate,
6435
        }
6436

    
6437
      if not ninfo.offline:
6438
        nresult.Raise()
6439
        if not isinstance(nresult.data, dict):
6440
          raise errors.OpExecError("Can't get data for node %s" % nname)
6441
        remote_info = nresult.data
6442
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
6443
                     'vg_size', 'vg_free', 'cpu_total']:
6444
          if attr not in remote_info:
6445
            raise errors.OpExecError("Node '%s' didn't return attribute"
6446
                                     " '%s'" % (nname, attr))
6447
          try:
6448
            remote_info[attr] = int(remote_info[attr])
6449
          except ValueError, err:
6450
            raise errors.OpExecError("Node '%s' returned invalid value"
6451
                                     " for '%s': %s" % (nname, attr, err))
6452
        # compute memory used by primary instances
6453
        i_p_mem = i_p_up_mem = 0
6454
        for iinfo, beinfo in i_list:
6455
          if iinfo.primary_node == nname:
6456
            i_p_mem += beinfo[constants.BE_MEMORY]
6457
            if iinfo.name not in node_iinfo[nname].data:
6458
              i_used_mem = 0
6459
            else:
6460
              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
6461
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
6462
            remote_info['memory_free'] -= max(0, i_mem_diff)
6463

    
6464
            if iinfo.admin_up:
6465
              i_p_up_mem += beinfo[constants.BE_MEMORY]
6466

    
6467
        # compute memory used by instances
6468
        pnr_dyn = {
6469
          "total_memory": remote_info['memory_total'],
6470
          "reserved_memory": remote_info['memory_dom0'],
6471
          "free_memory": remote_info['memory_free'],
6472
          "total_disk": remote_info['vg_size'],
6473
          "free_disk": remote_info['vg_free'],
6474
          "total_cpus": remote_info['cpu_total'],
6475
          "i_pri_memory": i_p_mem,
6476
          "i_pri_up_memory": i_p_up_mem,
6477
          }
6478
        pnr.update(pnr_dyn)
6479

    
6480
      node_results[nname] = pnr
6481
    data["nodes"] = node_results
6482

    
6483
    # instance data
6484
    instance_data = {}
6485
    for iinfo, beinfo in i_list:
6486
      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
6487
                  for n in iinfo.nics]
6488
      pir = {
6489
        "tags": list(iinfo.GetTags()),
6490
        "admin_up": iinfo.admin_up,
6491
        "vcpus": beinfo[constants.BE_VCPUS],
6492
        "memory": beinfo[constants.BE_MEMORY],
6493
        "os": iinfo.os,
6494
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
6495
        "nics": nic_data,
6496
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
6497
        "disk_template": iinfo.disk_template,
6498
        "hypervisor": iinfo.hypervisor,
6499
        }
6500
      instance_data[iinfo.name] = pir
6501

    
6502
    data["instances"] = instance_data
6503

    
6504
    self.in_data = data
6505

    
6506
  def _AddNewInstance(self):
6507
    """Add new instance data to allocator structure.
6508

6509
    This in combination with _AllocatorGetClusterData will create the
6510
    correct structure needed as input for the allocator.
6511

6512
    The checks for the completeness of the opcode must have already been
6513
    done.
6514

6515
    """
6516
    data = self.in_data
6517

    
6518
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
6519

    
6520
    if self.disk_template in constants.DTS_NET_MIRROR:
6521
      self.required_nodes = 2
6522
    else:
6523
      self.required_nodes = 1
6524
    request = {
6525
      "type": "allocate",
6526
      "name": self.name,
6527
      "disk_template": self.disk_template,
6528
      "tags": self.tags,
6529
      "os": self.os,
6530
      "vcpus": self.vcpus,
6531
      "memory": self.mem_size,
6532
      "disks": self.disks,
6533
      "disk_space_total": disk_space,
6534
      "nics": self.nics,
6535
      "required_nodes": self.required_nodes,
6536
      }
6537
    data["request"] = request
6538

    
6539
  def _AddRelocateInstance(self):
6540
    """Add relocate instance data to allocator structure.
6541

6542
    This in combination with _IAllocatorGetClusterData will create the
6543
    correct structure needed as input for the allocator.
6544

6545
    The checks for the completeness of the opcode must have already been
6546
    done.
6547

6548
    """
6549
    instance = self.lu.cfg.GetInstanceInfo(self.name)
6550
    if instance is None:
6551
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
6552
                                   " IAllocator" % self.name)
6553

    
6554
    if instance.disk_template not in constants.DTS_NET_MIRROR:
6555
      raise errors.OpPrereqError("Can't relocate non-mirrored instances")
6556

    
6557
    if len(instance.secondary_nodes) != 1:
6558
      raise errors.OpPrereqError("Instance has not exactly one secondary node")
6559

    
6560
    self.required_nodes = 1
6561
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
6562
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
6563

    
6564
    request = {
6565
      "type": "relocate",
6566
      "name": self.name,
6567
      "disk_space_total": disk_space,
6568
      "required_nodes": self.required_nodes,
6569
      "relocate_from": self.relocate_from,
6570
      }
6571
    self.in_data["request"] = request
6572

    
6573
  def _BuildInputData(self):
6574
    """Build input data structures.
6575

6576
    """
6577
    self._ComputeClusterData()
6578

    
6579
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6580
      self._AddNewInstance()
6581
    else:
6582
      self._AddRelocateInstance()
6583

    
6584
    self.in_text = serializer.Dump(self.in_data)
6585

    
6586
  def Run(self, name, validate=True, call_fn=None):
6587
    """Run an instance allocator and return the results.
6588

6589
    """
6590
    if call_fn is None:
6591
      call_fn = self.lu.rpc.call_iallocator_runner
6592
    data = self.in_text
6593

    
6594
    result = call_fn(self.lu.cfg.GetMasterNode(), name, self.in_text)
6595
    result.Raise()
6596

    
6597
    if not isinstance(result.data, (list, tuple)) or len(result.data) != 4:
6598
      raise errors.OpExecError("Invalid result from master iallocator runner")
6599

    
6600
    rcode, stdout, stderr, fail = result.data
6601

    
6602
    if rcode == constants.IARUN_NOTFOUND:
6603
      raise errors.OpExecError("Can't find allocator '%s'" % name)
6604
    elif rcode == constants.IARUN_FAILURE:
6605
      raise errors.OpExecError("Instance allocator call failed: %s,"
6606
                               " output: %s" % (fail, stdout+stderr))
6607
    self.out_text = stdout
6608
    if validate:
6609
      self._ValidateResult()
6610

    
6611
  def _ValidateResult(self):
6612
    """Process the allocator results.
6613

6614
    This will process and if successful save the result in
6615
    self.out_data and the other parameters.
6616

6617
    """
6618
    try:
6619
      rdict = serializer.Load(self.out_text)
6620
    except Exception, err:
6621
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
6622

    
6623
    if not isinstance(rdict, dict):
6624
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
6625

    
6626
    for key in "success", "info", "nodes":
6627
      if key not in rdict:
6628
        raise errors.OpExecError("Can't parse iallocator results:"
6629
                                 " missing key '%s'" % key)
6630
      setattr(self, key, rdict[key])
6631

    
6632
    if not isinstance(rdict["nodes"], list):
6633
      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
6634
                               " is not a list")
6635
    self.out_data = rdict
6636

    
6637

    
6638
class LUTestAllocator(NoHooksLU):
6639
  """Run allocator tests.
6640

6641
  This LU runs the allocator tests
6642

6643
  """
6644
  _OP_REQP = ["direction", "mode", "name"]
6645

    
6646
  def CheckPrereq(self):
6647
    """Check prerequisites.
6648

6649
    This checks the opcode parameters depending on the director and mode test.
6650

6651
    """
6652
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6653
      for attr in ["name", "mem_size", "disks", "disk_template",
6654
                   "os", "tags", "nics", "vcpus"]:
6655
        if not hasattr(self.op, attr):
6656
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
6657
                                     attr)
6658
      iname = self.cfg.ExpandInstanceName(self.op.name)
6659
      if iname is not None:
6660
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
6661
                                   iname)
6662
      if not isinstance(self.op.nics, list):
6663
        raise errors.OpPrereqError("Invalid parameter 'nics'")
6664
      for row in self.op.nics:
6665
        if (not isinstance(row, dict) or
6666
            "mac" not in row or
6667
            "ip" not in row or
6668
            "bridge" not in row):
6669
          raise errors.OpPrereqError("Invalid contents of the"
6670
                                     " 'nics' parameter")
6671
      if not isinstance(self.op.disks, list):
6672
        raise errors.OpPrereqError("Invalid parameter 'disks'")
6673
      for row in self.op.disks:
6674
        if (not isinstance(row, dict) or
6675
            "size" not in row or
6676
            not isinstance(row["size"], int) or
6677
            "mode" not in row or
6678
            row["mode"] not in ['r', 'w']):
6679
          raise errors.OpPrereqError("Invalid contents of the"
6680
                                     " 'disks' parameter")
6681
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
6682
        self.op.hypervisor = self.cfg.GetHypervisorType()
6683
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
6684
      if not hasattr(self.op, "name"):
6685
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
6686
      fname = self.cfg.ExpandInstanceName(self.op.name)
6687
      if fname is None:
6688
        raise errors.OpPrereqError("Instance '%s' not found for relocation" %
6689
                                   self.op.name)
6690
      self.op.name = fname
6691
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
6692
    else:
6693
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
6694
                                 self.op.mode)
6695

    
6696
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
6697
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
6698
        raise errors.OpPrereqError("Missing allocator name")
6699
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
6700
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
6701
                                 self.op.direction)
6702

    
6703
  def Exec(self, feedback_fn):
6704
    """Run the allocator test.
6705

6706
    """
6707
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6708
      ial = IAllocator(self,
6709
                       mode=self.op.mode,
6710
                       name=self.op.name,
6711
                       mem_size=self.op.mem_size,
6712
                       disks=self.op.disks,
6713
                       disk_template=self.op.disk_template,
6714
                       os=self.op.os,
6715
                       tags=self.op.tags,
6716
                       nics=self.op.nics,
6717
                       vcpus=self.op.vcpus,
6718
                       hypervisor=self.op.hypervisor,
6719
                       )
6720
    else:
6721
      ial = IAllocator(self,
6722
                       mode=self.op.mode,
6723
                       name=self.op.name,
6724
                       relocate_from=list(self.relocate_from),
6725
                       )
6726

    
6727
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
6728
      result = ial.in_text
6729
    else:
6730
      ial.Run(self.op.allocator, validate=False)
6731
      result = ial.out_text
6732
    return result