Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ c70d2d9b

History | View | Annotate | Download (253 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=W0201
25

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

    
29
import os
30
import os.path
31
import time
32
import re
33
import platform
34
import logging
35
import copy
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 serializer
45
from ganeti import ssconf
46

    
47

    
48
class LogicalUnit(object):
49
  """Logical Unit base class.
50

51
  Subclasses must follow these rules:
52
    - implement ExpandNames
53
    - implement CheckPrereq
54
    - implement Exec
55
    - implement BuildHooksEnv
56
    - redefine HPATH and HTYPE
57
    - optionally redefine their run requirements:
58
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
59

60
  Note that all commands require root permissions.
61

62
  """
63
  HPATH = None
64
  HTYPE = None
65
  _OP_REQP = []
66
  REQ_BGL = True
67

    
68
  def __init__(self, processor, op, context, rpc):
69
    """Constructor for LogicalUnit.
70

71
    This needs to be overridden in derived classes in order to check op
72
    validity.
73

74
    """
75
    self.proc = processor
76
    self.op = op
77
    self.cfg = context.cfg
78
    self.context = context
79
    self.rpc = rpc
80
    # Dicts used to declare locking needs to mcpu
81
    self.needed_locks = None
82
    self.acquired_locks = {}
83
    self.share_locks = dict(((i, 0) for i in locking.LEVELS))
84
    self.add_locks = {}
85
    self.remove_locks = {}
86
    # Used to force good behavior when calling helper functions
87
    self.recalculate_locks = {}
88
    self.__ssh = None
89
    # logging
90
    self.LogWarning = processor.LogWarning # pylint: disable-msg=C0103
91
    self.LogInfo = processor.LogInfo # pylint: disable-msg=C0103
92

    
93
    for attr_name in self._OP_REQP:
94
      attr_val = getattr(op, attr_name, None)
95
      if attr_val is None:
96
        raise errors.OpPrereqError("Required parameter '%s' missing" %
97
                                   attr_name)
98
    self.CheckArguments()
99

    
100
  def __GetSSH(self):
101
    """Returns the SshRunner object
102

103
    """
104
    if not self.__ssh:
105
      self.__ssh = ssh.SshRunner(self.cfg.GetClusterName())
106
    return self.__ssh
107

    
108
  ssh = property(fget=__GetSSH)
109

    
110
  def CheckArguments(self):
111
    """Check syntactic validity for the opcode arguments.
112

113
    This method is for doing a simple syntactic check and ensure
114
    validity of opcode parameters, without any cluster-related
115
    checks. While the same can be accomplished in ExpandNames and/or
116
    CheckPrereq, doing these separate is better because:
117

118
      - ExpandNames is left as as purely a lock-related function
119
      - CheckPrereq is run after we have acquired locks (and possible
120
        waited for them)
121

122
    The function is allowed to change the self.op attribute so that
123
    later methods can no longer worry about missing parameters.
124

125
    """
126
    pass
127

    
128
  def ExpandNames(self):
129
    """Expand names for this LU.
130

131
    This method is called before starting to execute the opcode, and it should
132
    update all the parameters of the opcode to their canonical form (e.g. a
133
    short node name must be fully expanded after this method has successfully
134
    completed). This way locking, hooks, logging, ecc. can work correctly.
135

136
    LUs which implement this method must also populate the self.needed_locks
137
    member, as a dict with lock levels as keys, and a list of needed lock names
138
    as values. Rules:
139

140
      - use an empty dict if you don't need any lock
141
      - if you don't need any lock at a particular level omit that level
142
      - don't put anything for the BGL level
143
      - if you want all locks at a level use locking.ALL_SET as a value
144

145
    If you need to share locks (rather than acquire them exclusively) at one
146
    level you can modify self.share_locks, setting a true value (usually 1) for
147
    that level. By default locks are not shared.
148

149
    Examples::
150

151
      # Acquire all nodes and one instance
152
      self.needed_locks = {
153
        locking.LEVEL_NODE: locking.ALL_SET,
154
        locking.LEVEL_INSTANCE: ['instance1.example.tld'],
155
      }
156
      # Acquire just two nodes
157
      self.needed_locks = {
158
        locking.LEVEL_NODE: ['node1.example.tld', 'node2.example.tld'],
159
      }
160
      # Acquire no locks
161
      self.needed_locks = {} # No, you can't leave it to the default value None
162

163
    """
164
    # The implementation of this method is mandatory only if the new LU is
165
    # concurrent, so that old LUs don't need to be changed all at the same
166
    # time.
167
    if self.REQ_BGL:
168
      self.needed_locks = {} # Exclusive LUs don't need locks.
169
    else:
170
      raise NotImplementedError
171

    
172
  def DeclareLocks(self, level):
173
    """Declare LU locking needs for a level
174

175
    While most LUs can just declare their locking needs at ExpandNames time,
176
    sometimes there's the need to calculate some locks after having acquired
177
    the ones before. This function is called just before acquiring locks at a
178
    particular level, but after acquiring the ones at lower levels, and permits
179
    such calculations. It can be used to modify self.needed_locks, and by
180
    default it does nothing.
181

182
    This function is only called if you have something already set in
183
    self.needed_locks for the level.
184

185
    @param level: Locking level which is going to be locked
186
    @type level: member of ganeti.locking.LEVELS
187

188
    """
189

    
190
  def CheckPrereq(self):
191
    """Check prerequisites for this LU.
192

193
    This method should check that the prerequisites for the execution
194
    of this LU are fulfilled. It can do internode communication, but
195
    it should be idempotent - no cluster or system changes are
196
    allowed.
197

198
    The method should raise errors.OpPrereqError in case something is
199
    not fulfilled. Its return value is ignored.
200

201
    This method should also update all the parameters of the opcode to
202
    their canonical form if it hasn't been done by ExpandNames before.
203

204
    """
205
    raise NotImplementedError
206

    
207
  def Exec(self, feedback_fn):
208
    """Execute the LU.
209

210
    This method should implement the actual work. It should raise
211
    errors.OpExecError for failures that are somewhat dealt with in
212
    code, or expected.
213

214
    """
215
    raise NotImplementedError
216

    
217
  def BuildHooksEnv(self):
218
    """Build hooks environment for this LU.
219

220
    This method should return a three-node tuple consisting of: a dict
221
    containing the environment that will be used for running the
222
    specific hook for this LU, a list of node names on which the hook
223
    should run before the execution, and a list of node names on which
224
    the hook should run after the execution.
225

226
    The keys of the dict must not have 'GANETI_' prefixed as this will
227
    be handled in the hooks runner. Also note additional keys will be
228
    added by the hooks runner. If the LU doesn't define any
229
    environment, an empty dict (and not None) should be returned.
230

231
    No nodes should be returned as an empty list (and not None).
232

233
    Note that if the HPATH for a LU class is None, this function will
234
    not be called.
235

236
    """
237
    raise NotImplementedError
238

    
239
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
240
    """Notify the LU about the results of its hooks.
241

242
    This method is called every time a hooks phase is executed, and notifies
243
    the Logical Unit about the hooks' result. The LU can then use it to alter
244
    its result based on the hooks.  By default the method does nothing and the
245
    previous result is passed back unchanged but any LU can define it if it
246
    wants to use the local cluster hook-scripts somehow.
247

248
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
249
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
250
    @param hook_results: the results of the multi-node hooks rpc call
251
    @param feedback_fn: function used send feedback back to the caller
252
    @param lu_result: the previous Exec result this LU had, or None
253
        in the PRE phase
254
    @return: the new Exec result, based on the previous result
255
        and hook results
256

257
    """
258
    return lu_result
259

    
260
  def _ExpandAndLockInstance(self):
261
    """Helper function to expand and lock an instance.
262

263
    Many LUs that work on an instance take its name in self.op.instance_name
264
    and need to expand it and then declare the expanded name for locking. This
265
    function does it, and then updates self.op.instance_name to the expanded
266
    name. It also initializes needed_locks as a dict, if this hasn't been done
267
    before.
268

269
    """
270
    if self.needed_locks is None:
271
      self.needed_locks = {}
272
    else:
273
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
274
        "_ExpandAndLockInstance called with instance-level locks set"
275
    expanded_name = self.cfg.ExpandInstanceName(self.op.instance_name)
276
    if expanded_name is None:
277
      raise errors.OpPrereqError("Instance '%s' not known" %
278
                                  self.op.instance_name)
279
    self.needed_locks[locking.LEVEL_INSTANCE] = expanded_name
280
    self.op.instance_name = expanded_name
281

    
282
  def _LockInstancesNodes(self, primary_only=False):
283
    """Helper function to declare instances' nodes for locking.
284

285
    This function should be called after locking one or more instances to lock
286
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
287
    with all primary or secondary nodes for instances already locked and
288
    present in self.needed_locks[locking.LEVEL_INSTANCE].
289

290
    It should be called from DeclareLocks, and for safety only works if
291
    self.recalculate_locks[locking.LEVEL_NODE] is set.
292

293
    In the future it may grow parameters to just lock some instance's nodes, or
294
    to just lock primaries or secondary nodes, if needed.
295

296
    If should be called in DeclareLocks in a way similar to::
297

298
      if level == locking.LEVEL_NODE:
299
        self._LockInstancesNodes()
300

301
    @type primary_only: boolean
302
    @param primary_only: only lock primary nodes of locked instances
303

304
    """
305
    assert locking.LEVEL_NODE in self.recalculate_locks, \
306
      "_LockInstancesNodes helper function called with no nodes to recalculate"
307

    
308
    # TODO: check if we're really been called with the instance locks held
309

    
310
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
311
    # future we might want to have different behaviors depending on the value
312
    # of self.recalculate_locks[locking.LEVEL_NODE]
313
    wanted_nodes = []
314
    for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
315
      instance = self.context.cfg.GetInstanceInfo(instance_name)
316
      wanted_nodes.append(instance.primary_node)
317
      if not primary_only:
318
        wanted_nodes.extend(instance.secondary_nodes)
319

    
320
    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
321
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
322
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
323
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
324

    
325
    del self.recalculate_locks[locking.LEVEL_NODE]
326

    
327

    
328
class NoHooksLU(LogicalUnit): # pylint: disable-msg=W0223
329
  """Simple LU which runs no hooks.
330

331
  This LU is intended as a parent for other LogicalUnits which will
332
  run no hooks, in order to reduce duplicate code.
333

334
  """
335
  HPATH = None
336
  HTYPE = None
337

    
338

    
339
def _GetWantedNodes(lu, nodes):
340
  """Returns list of checked and expanded node names.
341

342
  @type lu: L{LogicalUnit}
343
  @param lu: the logical unit on whose behalf we execute
344
  @type nodes: list
345
  @param nodes: list of node names or None for all nodes
346
  @rtype: list
347
  @return: the list of nodes, sorted
348
  @raise errors.OpProgrammerError: if the nodes parameter is wrong type
349

350
  """
351
  if not isinstance(nodes, list):
352
    raise errors.OpPrereqError("Invalid argument type 'nodes'")
353

    
354
  if not nodes:
355
    raise errors.ProgrammerError("_GetWantedNodes should only be called with a"
356
      " non-empty list of nodes whose name is to be expanded.")
357

    
358
  wanted = []
359
  for name in nodes:
360
    node = lu.cfg.ExpandNodeName(name)
361
    if node is None:
362
      raise errors.OpPrereqError("No such node name '%s'" % name)
363
    wanted.append(node)
364

    
365
  return utils.NiceSort(wanted)
366

    
367

    
368
def _GetWantedInstances(lu, instances):
369
  """Returns list of checked and expanded instance names.
370

371
  @type lu: L{LogicalUnit}
372
  @param lu: the logical unit on whose behalf we execute
373
  @type instances: list
374
  @param instances: list of instance names or None for all instances
375
  @rtype: list
376
  @return: the list of instances, sorted
377
  @raise errors.OpPrereqError: if the instances parameter is wrong type
378
  @raise errors.OpPrereqError: if any of the passed instances is not found
379

380
  """
381
  if not isinstance(instances, list):
382
    raise errors.OpPrereqError("Invalid argument type 'instances'")
383

    
384
  if instances:
385
    wanted = []
386

    
387
    for name in instances:
388
      instance = lu.cfg.ExpandInstanceName(name)
389
      if instance is None:
390
        raise errors.OpPrereqError("No such instance name '%s'" % name)
391
      wanted.append(instance)
392

    
393
  else:
394
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
395
  return wanted
396

    
397

    
398
def _CheckOutputFields(static, dynamic, selected):
399
  """Checks whether all selected fields are valid.
400

401
  @type static: L{utils.FieldSet}
402
  @param static: static fields set
403
  @type dynamic: L{utils.FieldSet}
404
  @param dynamic: dynamic fields set
405

406
  """
407
  f = utils.FieldSet()
408
  f.Extend(static)
409
  f.Extend(dynamic)
410

    
411
  delta = f.NonMatching(selected)
412
  if delta:
413
    raise errors.OpPrereqError("Unknown output fields selected: %s"
414
                               % ",".join(delta))
415

    
416

    
417
def _CheckBooleanOpField(op, name):
418
  """Validates boolean opcode parameters.
419

420
  This will ensure that an opcode parameter is either a boolean value,
421
  or None (but that it always exists).
422

423
  """
424
  val = getattr(op, name, None)
425
  if not (val is None or isinstance(val, bool)):
426
    raise errors.OpPrereqError("Invalid boolean parameter '%s' (%s)" %
427
                               (name, str(val)))
428
  setattr(op, name, val)
429

    
430

    
431
def _CheckNodeOnline(lu, node):
432
  """Ensure that a given node is online.
433

434
  @param lu: the LU on behalf of which we make the check
435
  @param node: the node to check
436
  @raise errors.OpPrereqError: if the node is offline
437

438
  """
439
  if lu.cfg.GetNodeInfo(node).offline:
440
    raise errors.OpPrereqError("Can't use offline node %s" % node)
441

    
442

    
443
def _CheckNodeNotDrained(lu, node):
444
  """Ensure that a given node is not drained.
445

446
  @param lu: the LU on behalf of which we make the check
447
  @param node: the node to check
448
  @raise errors.OpPrereqError: if the node is drained
449

450
  """
451
  if lu.cfg.GetNodeInfo(node).drained:
452
    raise errors.OpPrereqError("Can't use drained node %s" % node)
453

    
454

    
455
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
456
                          memory, vcpus, nics, disk_template, disks,
457
                          bep, hvp, hypervisor_name):
458
  """Builds instance related env variables for hooks
459

460
  This builds the hook environment from individual variables.
461

462
  @type name: string
463
  @param name: the name of the instance
464
  @type primary_node: string
465
  @param primary_node: the name of the instance's primary node
466
  @type secondary_nodes: list
467
  @param secondary_nodes: list of secondary nodes as strings
468
  @type os_type: string
469
  @param os_type: the name of the instance's OS
470
  @type status: boolean
471
  @param status: the should_run status of the instance
472
  @type memory: string
473
  @param memory: the memory size of the instance
474
  @type vcpus: string
475
  @param vcpus: the count of VCPUs the instance has
476
  @type nics: list
477
  @param nics: list of tuples (ip, bridge, mac) representing
478
      the NICs the instance  has
479
  @type disk_template: string
480
  @param disk_template: the disk template of the instance
481
  @type disks: list
482
  @param disks: the list of (size, mode) pairs
483
  @type bep: dict
484
  @param bep: the backend parameters for the instance
485
  @type hvp: dict
486
  @param hvp: the hypervisor parameters for the instance
487
  @type hypervisor_name: string
488
  @param hypervisor_name: the hypervisor for the instance
489
  @rtype: dict
490
  @return: the hook environment for this instance
491

492
  """
493
  if status:
494
    str_status = "up"
495
  else:
496
    str_status = "down"
497
  env = {
498
    "OP_TARGET": name,
499
    "INSTANCE_NAME": name,
500
    "INSTANCE_PRIMARY": primary_node,
501
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
502
    "INSTANCE_OS_TYPE": os_type,
503
    "INSTANCE_STATUS": str_status,
504
    "INSTANCE_MEMORY": memory,
505
    "INSTANCE_VCPUS": vcpus,
506
    "INSTANCE_DISK_TEMPLATE": disk_template,
507
    "INSTANCE_HYPERVISOR": hypervisor_name,
508
  }
509

    
510
  if nics:
511
    nic_count = len(nics)
512
    for idx, (ip, bridge, mac) in enumerate(nics):
513
      if ip is None:
514
        ip = ""
515
      env["INSTANCE_NIC%d_IP" % idx] = ip
516
      env["INSTANCE_NIC%d_BRIDGE" % idx] = bridge
517
      env["INSTANCE_NIC%d_MAC" % idx] = mac
518
  else:
519
    nic_count = 0
520

    
521
  env["INSTANCE_NIC_COUNT"] = nic_count
522

    
523
  if disks:
524
    disk_count = len(disks)
525
    for idx, (size, mode) in enumerate(disks):
526
      env["INSTANCE_DISK%d_SIZE" % idx] = size
527
      env["INSTANCE_DISK%d_MODE" % idx] = mode
528
  else:
529
    disk_count = 0
530

    
531
  env["INSTANCE_DISK_COUNT"] = disk_count
532

    
533
  for source, kind in [(bep, "BE"), (hvp, "HV")]:
534
    for key, value in source.items():
535
      env["INSTANCE_%s_%s" % (kind, key)] = value
536

    
537
  return env
538

    
539

    
540
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
541
  """Builds instance related env variables for hooks from an object.
542

543
  @type lu: L{LogicalUnit}
544
  @param lu: the logical unit on whose behalf we execute
545
  @type instance: L{objects.Instance}
546
  @param instance: the instance for which we should build the
547
      environment
548
  @type override: dict
549
  @param override: dictionary with key/values that will override
550
      our values
551
  @rtype: dict
552
  @return: the hook environment dictionary
553

554
  """
555
  cluster = lu.cfg.GetClusterInfo()
556
  bep = cluster.FillBE(instance)
557
  hvp = cluster.FillHV(instance)
558
  args = {
559
    'name': instance.name,
560
    'primary_node': instance.primary_node,
561
    'secondary_nodes': instance.secondary_nodes,
562
    'os_type': instance.os,
563
    'status': instance.admin_up,
564
    'memory': bep[constants.BE_MEMORY],
565
    'vcpus': bep[constants.BE_VCPUS],
566
    'nics': [(nic.ip, nic.bridge, nic.mac) for nic in instance.nics],
567
    'disk_template': instance.disk_template,
568
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
569
    'bep': bep,
570
    'hvp': hvp,
571
    'hypervisor_name': instance.hypervisor,
572
  }
573
  if override:
574
    args.update(override)
575
  return _BuildInstanceHookEnv(**args)
576

    
577

    
578
def _AdjustCandidatePool(lu):
579
  """Adjust the candidate pool after node operations.
580

581
  """
582
  mod_list = lu.cfg.MaintainCandidatePool()
583
  if mod_list:
584
    lu.LogInfo("Promoted nodes to master candidate role: %s",
585
               ", ".join(node.name for node in mod_list))
586
    for name in mod_list:
587
      lu.context.ReaddNode(name)
588
  mc_now, mc_max = lu.cfg.GetMasterCandidateStats()
589
  if mc_now > mc_max:
590
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
591
               (mc_now, mc_max))
592

    
593

    
594
def _CheckInstanceBridgesExist(lu, instance):
595
  """Check that the bridges needed by an instance exist.
596

597
  """
598
  # check bridges existence
599
  brlist = [nic.bridge for nic in instance.nics]
600
  result = lu.rpc.call_bridges_exist(instance.primary_node, brlist)
601
  result.Raise()
602
  if not result.data:
603
    raise errors.OpPrereqError("One or more target bridges %s does not"
604
                               " exist on destination node '%s'" %
605
                               (brlist, instance.primary_node))
606

    
607

    
608
class LUDestroyCluster(NoHooksLU):
609
  """Logical unit for destroying the cluster.
610

611
  """
612
  _OP_REQP = []
613

    
614
  def CheckPrereq(self):
615
    """Check prerequisites.
616

617
    This checks whether the cluster is empty.
618

619
    Any errors are signaled by raising errors.OpPrereqError.
620

621
    """
622
    master = self.cfg.GetMasterNode()
623

    
624
    nodelist = self.cfg.GetNodeList()
625
    if len(nodelist) != 1 or nodelist[0] != master:
626
      raise errors.OpPrereqError("There are still %d node(s) in"
627
                                 " this cluster." % (len(nodelist) - 1))
628
    instancelist = self.cfg.GetInstanceList()
629
    if instancelist:
630
      raise errors.OpPrereqError("There are still %d instance(s) in"
631
                                 " this cluster." % len(instancelist))
632

    
633
  def Exec(self, feedback_fn):
634
    """Destroys the cluster.
635

636
    """
637
    master = self.cfg.GetMasterNode()
638
    result = self.rpc.call_node_stop_master(master, False)
639
    result.Raise()
640
    if not result.data:
641
      raise errors.OpExecError("Could not disable the master role")
642
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
643
    utils.CreateBackup(priv_key)
644
    utils.CreateBackup(pub_key)
645
    return master
646

    
647

    
648
class LUVerifyCluster(LogicalUnit):
649
  """Verifies the cluster status.
650

651
  """
652
  HPATH = "cluster-verify"
653
  HTYPE = constants.HTYPE_CLUSTER
654
  _OP_REQP = ["skip_checks"]
655
  REQ_BGL = False
656

    
657
  def ExpandNames(self):
658
    self.needed_locks = {
659
      locking.LEVEL_NODE: locking.ALL_SET,
660
      locking.LEVEL_INSTANCE: locking.ALL_SET,
661
    }
662
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
663

    
664
  def _VerifyNode(self, nodeinfo, file_list, local_cksum,
665
                  node_result, feedback_fn, master_files,
666
                  drbd_map, vg_name):
667
    """Run multiple tests against a node.
668

669
    Test list:
670

671
      - compares ganeti version
672
      - checks vg existence and size > 20G
673
      - checks config file checksum
674
      - checks ssh to other nodes
675

676
    @type nodeinfo: L{objects.Node}
677
    @param nodeinfo: the node to check
678
    @param file_list: required list of files
679
    @param local_cksum: dictionary of local files and their checksums
680
    @param node_result: the results from the node
681
    @param feedback_fn: function used to accumulate results
682
    @param master_files: list of files that only masters should have
683
    @param drbd_map: the useddrbd minors for this node, in
684
        form of minor: (instance, must_exist) which correspond to instances
685
        and their running status
686
    @param vg_name: Ganeti Volume Group (result of self.cfg.GetVGName())
687

688
    """
689
    node = nodeinfo.name
690

    
691
    # main result, node_result should be a non-empty dict
692
    if not node_result or not isinstance(node_result, dict):
693
      feedback_fn("  - ERROR: unable to verify node %s." % (node,))
694
      return True
695

    
696
    # compares ganeti version
697
    local_version = constants.PROTOCOL_VERSION
698
    remote_version = node_result.get('version', None)
699
    if not (remote_version and isinstance(remote_version, (list, tuple)) and
700
            len(remote_version) == 2):
701
      feedback_fn("  - ERROR: connection to %s failed" % (node))
702
      return True
703

    
704
    if local_version != remote_version[0]:
705
      feedback_fn("  - ERROR: incompatible protocol versions: master %s,"
706
                  " node %s %s" % (local_version, node, remote_version[0]))
707
      return True
708

    
709
    # node seems compatible, we can actually try to look into its results
710

    
711
    bad = False
712

    
713
    # full package version
714
    if constants.RELEASE_VERSION != remote_version[1]:
715
      feedback_fn("  - WARNING: software version mismatch: master %s,"
716
                  " node %s %s" %
717
                  (constants.RELEASE_VERSION, node, remote_version[1]))
718

    
719
    # checks vg existence and size > 20G
720
    if vg_name is not None:
721
      vglist = node_result.get(constants.NV_VGLIST, None)
722
      if not vglist:
723
        feedback_fn("  - ERROR: unable to check volume groups on node %s." %
724
                        (node,))
725
        bad = True
726
      else:
727
        vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
728
                                              constants.MIN_VG_SIZE)
729
        if vgstatus:
730
          feedback_fn("  - ERROR: %s on node %s" % (vgstatus, node))
731
          bad = True
732

    
733
    # checks config file checksum
734

    
735
    remote_cksum = node_result.get(constants.NV_FILELIST, None)
736
    if not isinstance(remote_cksum, dict):
737
      bad = True
738
      feedback_fn("  - ERROR: node hasn't returned file checksum data")
739
    else:
740
      for file_name in file_list:
741
        node_is_mc = nodeinfo.master_candidate
742
        must_have_file = file_name not in master_files
743
        if file_name not in remote_cksum:
744
          if node_is_mc or must_have_file:
745
            bad = True
746
            feedback_fn("  - ERROR: file '%s' missing" % file_name)
747
        elif remote_cksum[file_name] != local_cksum[file_name]:
748
          if node_is_mc or must_have_file:
749
            bad = True
750
            feedback_fn("  - ERROR: file '%s' has wrong checksum" % file_name)
751
          else:
752
            # not candidate and this is not a must-have file
753
            bad = True
754
            feedback_fn("  - ERROR: file '%s' should not exist on non master"
755
                        " candidates (and the file is outdated)" % file_name)
756
        else:
757
          # all good, except non-master/non-must have combination
758
          if not node_is_mc and not must_have_file:
759
            feedback_fn("  - ERROR: file '%s' should not exist on non master"
760
                        " candidates" % file_name)
761

    
762
    # checks ssh to any
763

    
764
    if constants.NV_NODELIST not in node_result:
765
      bad = True
766
      feedback_fn("  - ERROR: node hasn't returned node ssh connectivity data")
767
    else:
768
      if node_result[constants.NV_NODELIST]:
769
        bad = True
770
        for node in node_result[constants.NV_NODELIST]:
771
          feedback_fn("  - ERROR: ssh communication with node '%s': %s" %
772
                          (node, node_result[constants.NV_NODELIST][node]))
773

    
774
    if constants.NV_NODENETTEST not in node_result:
775
      bad = True
776
      feedback_fn("  - ERROR: node hasn't returned node tcp connectivity data")
777
    else:
778
      if node_result[constants.NV_NODENETTEST]:
779
        bad = True
780
        nlist = utils.NiceSort(node_result[constants.NV_NODENETTEST].keys())
781
        for node in nlist:
782
          feedback_fn("  - ERROR: tcp communication with node '%s': %s" %
783
                          (node, node_result[constants.NV_NODENETTEST][node]))
784

    
785
    hyp_result = node_result.get(constants.NV_HYPERVISOR, None)
786
    if isinstance(hyp_result, dict):
787
      for hv_name, hv_result in hyp_result.iteritems():
788
        if hv_result is not None:
789
          feedback_fn("  - ERROR: hypervisor %s verify failure: '%s'" %
790
                      (hv_name, hv_result))
791

    
792
    # check used drbd list
793
    if vg_name is not None:
794
      used_minors = node_result.get(constants.NV_DRBDLIST, [])
795
      if not isinstance(used_minors, (tuple, list)):
796
        feedback_fn("  - ERROR: cannot parse drbd status file: %s" %
797
                    str(used_minors))
798
      else:
799
        for minor, (iname, must_exist) in drbd_map.items():
800
          if minor not in used_minors and must_exist:
801
            feedback_fn("  - ERROR: drbd minor %d of instance %s is"
802
                        " not active" % (minor, iname))
803
            bad = True
804
        for minor in used_minors:
805
          if minor not in drbd_map:
806
            feedback_fn("  - ERROR: unallocated drbd minor %d is in use" %
807
                        minor)
808
            bad = True
809

    
810
    return bad
811

    
812
  def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
813
                      node_instance, feedback_fn, n_offline):
814
    """Verify an instance.
815

816
    This function checks to see if the required block devices are
817
    available on the instance's node.
818

819
    """
820
    bad = False
821

    
822
    node_current = instanceconfig.primary_node
823

    
824
    node_vol_should = {}
825
    instanceconfig.MapLVsByNode(node_vol_should)
826

    
827
    for node in node_vol_should:
828
      if node in n_offline:
829
        # ignore missing volumes on offline nodes
830
        continue
831
      for volume in node_vol_should[node]:
832
        if node not in node_vol_is or volume not in node_vol_is[node]:
833
          feedback_fn("  - ERROR: volume %s missing on node %s" %
834
                          (volume, node))
835
          bad = True
836

    
837
    if instanceconfig.admin_up:
838
      if ((node_current not in node_instance or
839
          not instance in node_instance[node_current]) and
840
          node_current not in n_offline):
841
        feedback_fn("  - ERROR: instance %s not running on node %s" %
842
                        (instance, node_current))
843
        bad = True
844

    
845
    for node in node_instance:
846
      if (not node == node_current):
847
        if instance in node_instance[node]:
848
          feedback_fn("  - ERROR: instance %s should not run on node %s" %
849
                          (instance, node))
850
          bad = True
851

    
852
    return bad
853

    
854
  def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is, feedback_fn):
855
    """Verify if there are any unknown volumes in the cluster.
856

857
    The .os, .swap and backup volumes are ignored. All other volumes are
858
    reported as unknown.
859

860
    """
861
    bad = False
862

    
863
    for node in node_vol_is:
864
      for volume in node_vol_is[node]:
865
        if node not in node_vol_should or volume not in node_vol_should[node]:
866
          feedback_fn("  - ERROR: volume %s on node %s should not exist" %
867
                      (volume, node))
868
          bad = True
869
    return bad
870

    
871
  def _VerifyOrphanInstances(self, instancelist, node_instance, feedback_fn):
872
    """Verify the list of running instances.
873

874
    This checks what instances are running but unknown to the cluster.
875

876
    """
877
    bad = False
878
    for node in node_instance:
879
      for runninginstance in node_instance[node]:
880
        if runninginstance not in instancelist:
881
          feedback_fn("  - ERROR: instance %s on node %s should not exist" %
882
                          (runninginstance, node))
883
          bad = True
884
    return bad
885

    
886
  def _VerifyNPlusOneMemory(self, node_info, instance_cfg, feedback_fn):
887
    """Verify N+1 Memory Resilience.
888

889
    Check that if one single node dies we can still start all the instances it
890
    was primary for.
891

892
    """
893
    bad = False
894

    
895
    for node, nodeinfo in node_info.iteritems():
896
      # This code checks that every node which is now listed as secondary has
897
      # enough memory to host all instances it is supposed to should a single
898
      # other node in the cluster fail.
899
      # FIXME: not ready for failover to an arbitrary node
900
      # FIXME: does not support file-backed instances
901
      # WARNING: we currently take into account down instances as well as up
902
      # ones, considering that even if they're down someone might want to start
903
      # them even in the event of a node failure.
904
      for prinode, instances in nodeinfo['sinst-by-pnode'].iteritems():
905
        needed_mem = 0
906
        for instance in instances:
907
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
908
          if bep[constants.BE_AUTO_BALANCE]:
909
            needed_mem += bep[constants.BE_MEMORY]
910
        if nodeinfo['mfree'] < needed_mem:
911
          feedback_fn("  - ERROR: not enough memory on node %s to accommodate"
912
                      " failovers should node %s fail" % (node, prinode))
913
          bad = True
914
    return bad
915

    
916
  def CheckPrereq(self):
917
    """Check prerequisites.
918

919
    Transform the list of checks we're going to skip into a set and check that
920
    all its members are valid.
921

922
    """
923
    self.skip_set = frozenset(self.op.skip_checks)
924
    if not constants.VERIFY_OPTIONAL_CHECKS.issuperset(self.skip_set):
925
      raise errors.OpPrereqError("Invalid checks to be skipped specified")
926

    
927
  def BuildHooksEnv(self):
928
    """Build hooks env.
929

930
    Cluster-Verify hooks just ran in the post phase and their failure makes
931
    the output be logged in the verify output and the verification to fail.
932

933
    """
934
    all_nodes = self.cfg.GetNodeList()
935
    env = {
936
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
937
      }
938
    for node in self.cfg.GetAllNodesInfo().values():
939
      env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())
940

    
941
    return env, [], all_nodes
942

    
943
  def Exec(self, feedback_fn):
944
    """Verify integrity of cluster, performing various test on nodes.
945

946
    """
947
    bad = False
948
    feedback_fn("* Verifying global settings")
949
    for msg in self.cfg.VerifyConfig():
950
      feedback_fn("  - ERROR: %s" % msg)
951

    
952
    vg_name = self.cfg.GetVGName()
953
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
954
    nodelist = utils.NiceSort(self.cfg.GetNodeList())
955
    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
956
    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
957
    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
958
                        for iname in instancelist)
959
    i_non_redundant = [] # Non redundant instances
960
    i_non_a_balanced = [] # Non auto-balanced instances
961
    n_offline = [] # List of offline nodes
962
    n_drained = [] # List of nodes being drained
963
    node_volume = {}
964
    node_instance = {}
965
    node_info = {}
966
    instance_cfg = {}
967

    
968
    # FIXME: verify OS list
969
    # do local checksums
970
    master_files = [constants.CLUSTER_CONF_FILE]
971

    
972
    file_names = ssconf.SimpleStore().GetFileList()
973
    file_names.append(constants.SSL_CERT_FILE)
974
    file_names.append(constants.RAPI_CERT_FILE)
975
    file_names.extend(master_files)
976

    
977
    local_checksums = utils.FingerprintFiles(file_names)
978

    
979
    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
980
    node_verify_param = {
981
      constants.NV_FILELIST: file_names,
982
      constants.NV_NODELIST: [node.name for node in nodeinfo
983
                              if not node.offline],
984
      constants.NV_HYPERVISOR: hypervisors,
985
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
986
                                  node.secondary_ip) for node in nodeinfo
987
                                 if not node.offline],
988
      constants.NV_INSTANCELIST: hypervisors,
989
      constants.NV_VERSION: None,
990
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
991
      }
992
    if vg_name is not None:
993
      node_verify_param[constants.NV_VGLIST] = None
994
      node_verify_param[constants.NV_LVLIST] = vg_name
995
      node_verify_param[constants.NV_DRBDLIST] = None
996
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
997
                                           self.cfg.GetClusterName())
998

    
999
    cluster = self.cfg.GetClusterInfo()
1000
    master_node = self.cfg.GetMasterNode()
1001
    all_drbd_map = self.cfg.ComputeDRBDMap()
1002

    
1003
    for node_i in nodeinfo:
1004
      node = node_i.name
1005
      nresult = all_nvinfo[node].data
1006

    
1007
      if node_i.offline:
1008
        feedback_fn("* Skipping offline node %s" % (node,))
1009
        n_offline.append(node)
1010
        continue
1011

    
1012
      if node == master_node:
1013
        ntype = "master"
1014
      elif node_i.master_candidate:
1015
        ntype = "master candidate"
1016
      elif node_i.drained:
1017
        ntype = "drained"
1018
        n_drained.append(node)
1019
      else:
1020
        ntype = "regular"
1021
      feedback_fn("* Verifying node %s (%s)" % (node, ntype))
1022

    
1023
      if all_nvinfo[node].failed or not isinstance(nresult, dict):
1024
        feedback_fn("  - ERROR: connection to %s failed" % (node,))
1025
        bad = True
1026
        continue
1027

    
1028
      node_drbd = {}
1029
      for minor, instance in all_drbd_map[node].items():
1030
        if instance not in instanceinfo:
1031
          feedback_fn("  - ERROR: ghost instance '%s' in temporary DRBD map" %
1032
                      instance)
1033
          # ghost instance should not be running, but otherwise we
1034
          # don't give double warnings (both ghost instance and
1035
          # unallocated minor in use)
1036
          node_drbd[minor] = (instance, False)
1037
        else:
1038
          instance = instanceinfo[instance]
1039
          node_drbd[minor] = (instance.name, instance.admin_up)
1040
      result = self._VerifyNode(node_i, file_names, local_checksums,
1041
                                nresult, feedback_fn, master_files,
1042
                                node_drbd, vg_name)
1043
      bad = bad or result
1044

    
1045
      lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
1046
      if vg_name is None:
1047
        node_volume[node] = {}
1048
      elif isinstance(lvdata, basestring):
1049
        feedback_fn("  - ERROR: LVM problem on node %s: %s" %
1050
                    (node, utils.SafeEncode(lvdata)))
1051
        bad = True
1052
        node_volume[node] = {}
1053
      elif not isinstance(lvdata, dict):
1054
        feedback_fn("  - ERROR: connection to %s failed (lvlist)" % (node,))
1055
        bad = True
1056
        continue
1057
      else:
1058
        node_volume[node] = lvdata
1059

    
1060
      # node_instance
1061
      idata = nresult.get(constants.NV_INSTANCELIST, None)
1062
      if not isinstance(idata, list):
1063
        feedback_fn("  - ERROR: connection to %s failed (instancelist)" %
1064
                    (node,))
1065
        bad = True
1066
        continue
1067

    
1068
      node_instance[node] = idata
1069

    
1070
      # node_info
1071
      nodeinfo = nresult.get(constants.NV_HVINFO, None)
1072
      if not isinstance(nodeinfo, dict):
1073
        feedback_fn("  - ERROR: connection to %s failed (hvinfo)" % (node,))
1074
        bad = True
1075
        continue
1076

    
1077
      try:
1078
        node_info[node] = {
1079
          "mfree": int(nodeinfo['memory_free']),
1080
          "pinst": [],
1081
          "sinst": [],
1082
          # dictionary holding all instances this node is secondary for,
1083
          # grouped by their primary node. Each key is a cluster node, and each
1084
          # value is a list of instances which have the key as primary and the
1085
          # current node as secondary.  this is handy to calculate N+1 memory
1086
          # availability if you can only failover from a primary to its
1087
          # secondary.
1088
          "sinst-by-pnode": {},
1089
        }
1090
        # FIXME: devise a free space model for file based instances as well
1091
        if vg_name is not None:
1092
          if (constants.NV_VGLIST not in nresult or
1093
              vg_name not in nresult[constants.NV_VGLIST]):
1094
            feedback_fn("  - ERROR: node %s didn't return data for the"
1095
                        " volume group '%s' - it is either missing or broken" %
1096
                        (node, vg_name))
1097
            bad = True
1098
            continue
1099
          node_info[node]["dfree"] = int(nresult[constants.NV_VGLIST][vg_name])
1100
      except (ValueError, KeyError):
1101
        feedback_fn("  - ERROR: invalid nodeinfo value returned"
1102
                    " from node %s" % (node,))
1103
        bad = True
1104
        continue
1105

    
1106
    node_vol_should = {}
1107

    
1108
    for instance in instancelist:
1109
      feedback_fn("* Verifying instance %s" % instance)
1110
      inst_config = instanceinfo[instance]
1111
      result =  self._VerifyInstance(instance, inst_config, node_volume,
1112
                                     node_instance, feedback_fn, n_offline)
1113
      bad = bad or result
1114
      inst_nodes_offline = []
1115

    
1116
      inst_config.MapLVsByNode(node_vol_should)
1117

    
1118
      instance_cfg[instance] = inst_config
1119

    
1120
      pnode = inst_config.primary_node
1121
      if pnode in node_info:
1122
        node_info[pnode]['pinst'].append(instance)
1123
      elif pnode not in n_offline:
1124
        feedback_fn("  - ERROR: instance %s, connection to primary node"
1125
                    " %s failed" % (instance, pnode))
1126
        bad = True
1127

    
1128
      if pnode in n_offline:
1129
        inst_nodes_offline.append(pnode)
1130

    
1131
      # If the instance is non-redundant we cannot survive losing its primary
1132
      # node, so we are not N+1 compliant. On the other hand we have no disk
1133
      # templates with more than one secondary so that situation is not well
1134
      # supported either.
1135
      # FIXME: does not support file-backed instances
1136
      if len(inst_config.secondary_nodes) == 0:
1137
        i_non_redundant.append(instance)
1138
      elif len(inst_config.secondary_nodes) > 1:
1139
        feedback_fn("  - WARNING: multiple secondaries for instance %s"
1140
                    % instance)
1141

    
1142
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
1143
        i_non_a_balanced.append(instance)
1144

    
1145
      for snode in inst_config.secondary_nodes:
1146
        if snode in node_info:
1147
          node_info[snode]['sinst'].append(instance)
1148
          if pnode not in node_info[snode]['sinst-by-pnode']:
1149
            node_info[snode]['sinst-by-pnode'][pnode] = []
1150
          node_info[snode]['sinst-by-pnode'][pnode].append(instance)
1151
        elif snode not in n_offline:
1152
          feedback_fn("  - ERROR: instance %s, connection to secondary node"
1153
                      " %s failed" % (instance, snode))
1154
          bad = True
1155
        if snode in n_offline:
1156
          inst_nodes_offline.append(snode)
1157

    
1158
      if inst_nodes_offline:
1159
        # warn that the instance lives on offline nodes, and set bad=True
1160
        feedback_fn("  - ERROR: instance lives on offline node(s) %s" %
1161
                    ", ".join(inst_nodes_offline))
1162
        bad = True
1163

    
1164
    feedback_fn("* Verifying orphan volumes")
1165
    result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
1166
                                       feedback_fn)
1167
    bad = bad or result
1168

    
1169
    feedback_fn("* Verifying remaining instances")
1170
    result = self._VerifyOrphanInstances(instancelist, node_instance,
1171
                                         feedback_fn)
1172
    bad = bad or result
1173

    
1174
    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
1175
      feedback_fn("* Verifying N+1 Memory redundancy")
1176
      result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
1177
      bad = bad or result
1178

    
1179
    feedback_fn("* Other Notes")
1180
    if i_non_redundant:
1181
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
1182
                  % len(i_non_redundant))
1183

    
1184
    if i_non_a_balanced:
1185
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
1186
                  % len(i_non_a_balanced))
1187

    
1188
    if n_offline:
1189
      feedback_fn("  - NOTICE: %d offline node(s) found." % len(n_offline))
1190

    
1191
    if n_drained:
1192
      feedback_fn("  - NOTICE: %d drained node(s) found." % len(n_drained))
1193

    
1194
    return not bad
1195

    
1196
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
1197
    """Analyze the post-hooks' result
1198

1199
    This method analyses the hook result, handles it, and sends some
1200
    nicely-formatted feedback back to the user.
1201

1202
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
1203
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
1204
    @param hooks_results: the results of the multi-node hooks rpc call
1205
    @param feedback_fn: function used send feedback back to the caller
1206
    @param lu_result: previous Exec result
1207
    @return: the new Exec result, based on the previous result
1208
        and hook results
1209

1210
    """
1211
    # We only really run POST phase hooks, and are only interested in
1212
    # their results
1213
    if phase == constants.HOOKS_PHASE_POST:
1214
      # Used to change hooks' output to proper indentation
1215
      indent_re = re.compile('^', re.M)
1216
      feedback_fn("* Hooks Results")
1217
      if not hooks_results:
1218
        feedback_fn("  - ERROR: general communication failure")
1219
        lu_result = 1
1220
      else:
1221
        for node_name in hooks_results:
1222
          show_node_header = True
1223
          res = hooks_results[node_name]
1224
          if res.failed or res.data is False or not isinstance(res.data, list):
1225
            if res.offline:
1226
              # no need to warn or set fail return value
1227
              continue
1228
            feedback_fn("    Communication failure in hooks execution")
1229
            lu_result = 1
1230
            continue
1231
          for script, hkr, output in res.data:
1232
            if hkr == constants.HKR_FAIL:
1233
              # The node header is only shown once, if there are
1234
              # failing hooks on that node
1235
              if show_node_header:
1236
                feedback_fn("  Node %s:" % node_name)
1237
                show_node_header = False
1238
              feedback_fn("    ERROR: Script %s failed, output:" % script)
1239
              output = indent_re.sub('      ', output)
1240
              feedback_fn("%s" % output)
1241
              lu_result = 1
1242

    
1243
      return lu_result
1244

    
1245

    
1246
class LUVerifyDisks(NoHooksLU):
1247
  """Verifies the cluster disks status.
1248

1249
  """
1250
  _OP_REQP = []
1251
  REQ_BGL = False
1252

    
1253
  def ExpandNames(self):
1254
    self.needed_locks = {
1255
      locking.LEVEL_NODE: locking.ALL_SET,
1256
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1257
    }
1258
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
1259

    
1260
  def CheckPrereq(self):
1261
    """Check prerequisites.
1262

1263
    This has no prerequisites.
1264

1265
    """
1266
    pass
1267

    
1268
  def Exec(self, feedback_fn):
1269
    """Verify integrity of cluster disks.
1270

1271
    """
1272
    result = res_nodes, res_nlvm, res_instances, res_missing = [], {}, [], {}
1273

    
1274
    vg_name = self.cfg.GetVGName()
1275
    nodes = utils.NiceSort(self.cfg.GetNodeList())
1276
    instances = [self.cfg.GetInstanceInfo(name)
1277
                 for name in self.cfg.GetInstanceList()]
1278

    
1279
    nv_dict = {}
1280
    for inst in instances:
1281
      inst_lvs = {}
1282
      if (not inst.admin_up or
1283
          inst.disk_template not in constants.DTS_NET_MIRROR):
1284
        continue
1285
      inst.MapLVsByNode(inst_lvs)
1286
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
1287
      for node, vol_list in inst_lvs.iteritems():
1288
        for vol in vol_list:
1289
          nv_dict[(node, vol)] = inst
1290

    
1291
    if not nv_dict:
1292
      return result
1293

    
1294
    node_lvs = self.rpc.call_volume_list(nodes, vg_name)
1295

    
1296
    for node in nodes:
1297
      # node_volume
1298
      lvs = node_lvs[node]
1299
      if lvs.failed:
1300
        if not lvs.offline:
1301
          self.LogWarning("Connection to node %s failed: %s" %
1302
                          (node, lvs.data))
1303
        continue
1304
      lvs = lvs.data
1305
      if isinstance(lvs, basestring):
1306
        logging.warning("Error enumerating LVs on node %s: %s", node, lvs)
1307
        res_nlvm[node] = lvs
1308
        continue
1309
      elif not isinstance(lvs, dict):
1310
        logging.warning("Connection to node %s failed or invalid data"
1311
                        " returned", node)
1312
        res_nodes.append(node)
1313
        continue
1314

    
1315
      for lv_name, (_, lv_inactive, lv_online) in lvs.iteritems():
1316
        inst = nv_dict.pop((node, lv_name), None)
1317
        if (not lv_online and inst is not None
1318
            and inst.name not in res_instances):
1319
          res_instances.append(inst.name)
1320

    
1321
    # any leftover items in nv_dict are missing LVs, let's arrange the
1322
    # data better
1323
    for key, inst in nv_dict.iteritems():
1324
      if inst.name not in res_missing:
1325
        res_missing[inst.name] = []
1326
      res_missing[inst.name].append(key)
1327

    
1328
    return result
1329

    
1330

    
1331
class LURepairDiskSizes(NoHooksLU):
1332
  """Verifies the cluster disks sizes.
1333

1334
  """
1335
  _OP_REQP = ["instances"]
1336
  REQ_BGL = False
1337

    
1338
  def ExpandNames(self):
1339

    
1340
    if not isinstance(self.op.instances, list):
1341
      raise errors.OpPrereqError("Invalid argument type 'instances'")
1342

    
1343
    if self.op.instances:
1344
      self.wanted_names = []
1345
      for name in self.op.instances:
1346
        full_name = self.cfg.ExpandInstanceName(name)
1347
        if full_name is None:
1348
          raise errors.OpPrereqError("Instance '%s' not known" % name)
1349
        self.wanted_names.append(full_name)
1350
      self.needed_locks = {
1351
        locking.LEVEL_NODE: [],
1352
        locking.LEVEL_INSTANCE: self.wanted_names,
1353
        }
1354
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
1355
    else:
1356
      self.wanted_names = None
1357
      self.needed_locks = {
1358
        locking.LEVEL_NODE: locking.ALL_SET,
1359
        locking.LEVEL_INSTANCE: locking.ALL_SET,
1360
        }
1361
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
1362

    
1363
  def DeclareLocks(self, level):
1364
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
1365
      self._LockInstancesNodes(primary_only=True)
1366

    
1367
  def CheckPrereq(self):
1368
    """Check prerequisites.
1369

1370
    This only checks the optional instance list against the existing names.
1371

1372
    """
1373
    if self.wanted_names is None:
1374
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
1375

    
1376
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
1377
                             in self.wanted_names]
1378

    
1379
  def _EnsureChildSizes(self, disk):
1380
    """Ensure children of the disk have the needed disk size.
1381

1382
    This is valid mainly for DRBD8 and fixes an issue where the
1383
    children have smaller disk size.
1384

1385
    @param disk: an L{ganeti.objects.Disk} object
1386

1387
    """
1388
    if disk.dev_type == constants.LD_DRBD8:
1389
      assert disk.children, "Empty children for DRBD8?"
1390
      fchild = disk.children[0]
1391
      mismatch = fchild.size < disk.size
1392
      if mismatch:
1393
        self.LogInfo("Child disk has size %d, parent %d, fixing",
1394
                     fchild.size, disk.size)
1395
        fchild.size = disk.size
1396

    
1397
      # and we recurse on this child only, not on the metadev
1398
      return self._EnsureChildSizes(fchild) or mismatch
1399
    else:
1400
      return False
1401

    
1402
  def Exec(self, feedback_fn):
1403
    """Verify the size of cluster disks.
1404

1405
    """
1406
    # TODO: check child disks too
1407
    # TODO: check differences in size between primary/secondary nodes
1408
    per_node_disks = {}
1409
    for instance in self.wanted_instances:
1410
      pnode = instance.primary_node
1411
      if pnode not in per_node_disks:
1412
        per_node_disks[pnode] = []
1413
      for idx, disk in enumerate(instance.disks):
1414
        per_node_disks[pnode].append((instance, idx, disk))
1415

    
1416
    changed = []
1417
    for node, dskl in per_node_disks.items():
1418
      newl = [v[2].Copy() for v in dskl]
1419
      for dsk in newl:
1420
        self.cfg.SetDiskID(dsk, node)
1421
      result = self.rpc.call_blockdev_getsizes(node, newl)
1422
      if result.failed:
1423
        self.LogWarning("Failure in blockdev_getsizes call to node"
1424
                        " %s, ignoring", node)
1425
        continue
1426
      if len(result.data) != len(dskl):
1427
        self.LogWarning("Invalid result from node %s, ignoring node results",
1428
                        node)
1429
        continue
1430
      for ((instance, idx, disk), size) in zip(dskl, result.data):
1431
        if size is None:
1432
          self.LogWarning("Disk %d of instance %s did not return size"
1433
                          " information, ignoring", idx, instance.name)
1434
          continue
1435
        if not isinstance(size, (int, long)):
1436
          self.LogWarning("Disk %d of instance %s did not return valid"
1437
                          " size information, ignoring", idx, instance.name)
1438
          continue
1439
        size = size >> 20
1440
        if size != disk.size:
1441
          self.LogInfo("Disk %d of instance %s has mismatched size,"
1442
                       " correcting: recorded %d, actual %d", idx,
1443
                       instance.name, disk.size, size)
1444
          disk.size = size
1445
          self.cfg.Update(instance)
1446
          changed.append((instance.name, idx, size))
1447
        if self._EnsureChildSizes(disk):
1448
          self.cfg.Update(instance)
1449
          changed.append((instance.name, idx, disk.size))
1450
    return changed
1451

    
1452

    
1453
class LURenameCluster(LogicalUnit):
1454
  """Rename the cluster.
1455

1456
  """
1457
  HPATH = "cluster-rename"
1458
  HTYPE = constants.HTYPE_CLUSTER
1459
  _OP_REQP = ["name"]
1460

    
1461
  def BuildHooksEnv(self):
1462
    """Build hooks env.
1463

1464
    """
1465
    env = {
1466
      "OP_TARGET": self.cfg.GetClusterName(),
1467
      "NEW_NAME": self.op.name,
1468
      }
1469
    mn = self.cfg.GetMasterNode()
1470
    return env, [mn], [mn]
1471

    
1472
  def CheckPrereq(self):
1473
    """Verify that the passed name is a valid one.
1474

1475
    """
1476
    hostname = utils.HostInfo(self.op.name)
1477

    
1478
    new_name = hostname.name
1479
    self.ip = new_ip = hostname.ip
1480
    old_name = self.cfg.GetClusterName()
1481
    old_ip = self.cfg.GetMasterIP()
1482
    if new_name == old_name and new_ip == old_ip:
1483
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
1484
                                 " cluster has changed")
1485
    if new_ip != old_ip:
1486
      if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
1487
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
1488
                                   " reachable on the network. Aborting." %
1489
                                   new_ip)
1490

    
1491
    self.op.name = new_name
1492

    
1493
  def Exec(self, feedback_fn):
1494
    """Rename the cluster.
1495

1496
    """
1497
    clustername = self.op.name
1498
    ip = self.ip
1499

    
1500
    # shutdown the master IP
1501
    master = self.cfg.GetMasterNode()
1502
    result = self.rpc.call_node_stop_master(master, False)
1503
    if result.failed or not result.data:
1504
      raise errors.OpExecError("Could not disable the master role")
1505

    
1506
    try:
1507
      cluster = self.cfg.GetClusterInfo()
1508
      cluster.cluster_name = clustername
1509
      cluster.master_ip = ip
1510
      self.cfg.Update(cluster)
1511

    
1512
      # update the known hosts file
1513
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
1514
      node_list = self.cfg.GetNodeList()
1515
      try:
1516
        node_list.remove(master)
1517
      except ValueError:
1518
        pass
1519
      result = self.rpc.call_upload_file(node_list,
1520
                                         constants.SSH_KNOWN_HOSTS_FILE)
1521
      for to_node, to_result in result.iteritems():
1522
        if to_result.failed or not to_result.data:
1523
          logging.error("Copy of file %s to node %s failed",
1524
                        constants.SSH_KNOWN_HOSTS_FILE, to_node)
1525

    
1526
    finally:
1527
      result = self.rpc.call_node_start_master(master, False, False)
1528
      if result.failed or not result.data:
1529
        self.LogWarning("Could not re-enable the master role on"
1530
                        " the master, please restart manually.")
1531

    
1532

    
1533
def _RecursiveCheckIfLVMBased(disk):
1534
  """Check if the given disk or its children are lvm-based.
1535

1536
  @type disk: L{objects.Disk}
1537
  @param disk: the disk to check
1538
  @rtype: boolean
1539
  @return: boolean indicating whether a LD_LV dev_type was found or not
1540

1541
  """
1542
  if disk.children:
1543
    for chdisk in disk.children:
1544
      if _RecursiveCheckIfLVMBased(chdisk):
1545
        return True
1546
  return disk.dev_type == constants.LD_LV
1547

    
1548

    
1549
class LUSetClusterParams(LogicalUnit):
1550
  """Change the parameters of the cluster.
1551

1552
  """
1553
  HPATH = "cluster-modify"
1554
  HTYPE = constants.HTYPE_CLUSTER
1555
  _OP_REQP = []
1556
  REQ_BGL = False
1557

    
1558
  def CheckArguments(self):
1559
    """Check parameters
1560

1561
    """
1562
    if not hasattr(self.op, "candidate_pool_size"):
1563
      self.op.candidate_pool_size = None
1564
    if self.op.candidate_pool_size is not None:
1565
      try:
1566
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1567
      except (ValueError, TypeError), err:
1568
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1569
                                   str(err))
1570
      if self.op.candidate_pool_size < 1:
1571
        raise errors.OpPrereqError("At least one master candidate needed")
1572

    
1573
  def ExpandNames(self):
1574
    # FIXME: in the future maybe other cluster params won't require checking on
1575
    # all nodes to be modified.
1576
    self.needed_locks = {
1577
      locking.LEVEL_NODE: locking.ALL_SET,
1578
    }
1579
    self.share_locks[locking.LEVEL_NODE] = 1
1580

    
1581
  def BuildHooksEnv(self):
1582
    """Build hooks env.
1583

1584
    """
1585
    env = {
1586
      "OP_TARGET": self.cfg.GetClusterName(),
1587
      "NEW_VG_NAME": self.op.vg_name,
1588
      }
1589
    mn = self.cfg.GetMasterNode()
1590
    return env, [mn], [mn]
1591

    
1592
  def CheckPrereq(self):
1593
    """Check prerequisites.
1594

1595
    This checks whether the given params don't conflict and
1596
    if the given volume group is valid.
1597

1598
    """
1599
    if self.op.vg_name is not None and not self.op.vg_name:
1600
      instances = self.cfg.GetAllInstancesInfo().values()
1601
      for inst in instances:
1602
        for disk in inst.disks:
1603
          if _RecursiveCheckIfLVMBased(disk):
1604
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1605
                                       " lvm-based instances exist")
1606

    
1607
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1608

    
1609
    # if vg_name not None, checks given volume group on all nodes
1610
    if self.op.vg_name:
1611
      vglist = self.rpc.call_vg_list(node_list)
1612
      for node in node_list:
1613
        if vglist[node].failed:
1614
          # ignoring down node
1615
          self.LogWarning("Node %s unreachable/error, ignoring" % node)
1616
          continue
1617
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].data,
1618
                                              self.op.vg_name,
1619
                                              constants.MIN_VG_SIZE)
1620
        if vgstatus:
1621
          raise errors.OpPrereqError("Error on node '%s': %s" %
1622
                                     (node, vgstatus))
1623

    
1624
    self.cluster = cluster = self.cfg.GetClusterInfo()
1625
    # validate beparams changes
1626
    if self.op.beparams:
1627
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
1628
      self.new_beparams = cluster.FillDict(
1629
        cluster.beparams[constants.BEGR_DEFAULT], self.op.beparams)
1630

    
1631
    # hypervisor list/parameters
1632
    self.new_hvparams = cluster.FillDict(cluster.hvparams, {})
1633
    if self.op.hvparams:
1634
      if not isinstance(self.op.hvparams, dict):
1635
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1636
      for hv_name, hv_dict in self.op.hvparams.items():
1637
        if hv_name not in self.new_hvparams:
1638
          self.new_hvparams[hv_name] = hv_dict
1639
        else:
1640
          self.new_hvparams[hv_name].update(hv_dict)
1641

    
1642
    if self.op.enabled_hypervisors is not None:
1643
      self.hv_list = self.op.enabled_hypervisors
1644
      if not self.hv_list:
1645
        raise errors.OpPrereqError("Enabled hypervisors list must contain at"
1646
                                   " least one member")
1647
      invalid_hvs = set(self.hv_list) - constants.HYPER_TYPES
1648
      if invalid_hvs:
1649
        raise errors.OpPrereqError("Enabled hypervisors contains invalid"
1650
                                   " entries: %s" % invalid_hvs)
1651
    else:
1652
      self.hv_list = cluster.enabled_hypervisors
1653

    
1654
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1655
      # either the enabled list has changed, or the parameters have, validate
1656
      for hv_name, hv_params in self.new_hvparams.items():
1657
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1658
            (self.op.enabled_hypervisors and
1659
             hv_name in self.op.enabled_hypervisors)):
1660
          # either this is a new hypervisor, or its parameters have changed
1661
          hv_class = hypervisor.GetHypervisor(hv_name)
1662
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1663
          hv_class.CheckParameterSyntax(hv_params)
1664
          _CheckHVParams(self, node_list, hv_name, hv_params)
1665

    
1666
  def Exec(self, feedback_fn):
1667
    """Change the parameters of the cluster.
1668

1669
    """
1670
    if self.op.vg_name is not None:
1671
      new_volume = self.op.vg_name
1672
      if not new_volume:
1673
        new_volume = None
1674
      if new_volume != self.cfg.GetVGName():
1675
        self.cfg.SetVGName(new_volume)
1676
      else:
1677
        feedback_fn("Cluster LVM configuration already in desired"
1678
                    " state, not changing")
1679
    if self.op.hvparams:
1680
      self.cluster.hvparams = self.new_hvparams
1681
    if self.op.enabled_hypervisors is not None:
1682
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1683
    if self.op.beparams:
1684
      self.cluster.beparams[constants.BEGR_DEFAULT] = self.new_beparams
1685
    if self.op.candidate_pool_size is not None:
1686
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1687
      # we need to update the pool size here, otherwise the save will fail
1688
      _AdjustCandidatePool(self)
1689

    
1690
    self.cfg.Update(self.cluster)
1691

    
1692

    
1693
class LURedistributeConfig(NoHooksLU):
1694
  """Force the redistribution of cluster configuration.
1695

1696
  This is a very simple LU.
1697

1698
  """
1699
  _OP_REQP = []
1700
  REQ_BGL = False
1701

    
1702
  def ExpandNames(self):
1703
    self.needed_locks = {
1704
      locking.LEVEL_NODE: locking.ALL_SET,
1705
    }
1706
    self.share_locks[locking.LEVEL_NODE] = 1
1707

    
1708
  def CheckPrereq(self):
1709
    """Check prerequisites.
1710

1711
    """
1712

    
1713
  def Exec(self, feedback_fn):
1714
    """Redistribute the configuration.
1715

1716
    """
1717
    self.cfg.Update(self.cfg.GetClusterInfo())
1718

    
1719

    
1720
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1721
  """Sleep and poll for an instance's disk to sync.
1722

1723
  """
1724
  if not instance.disks:
1725
    return True
1726

    
1727
  if not oneshot:
1728
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1729

    
1730
  node = instance.primary_node
1731

    
1732
  for dev in instance.disks:
1733
    lu.cfg.SetDiskID(dev, node)
1734

    
1735
  retries = 0
1736
  degr_retries = 10 # in seconds, as we sleep 1 second each time
1737
  while True:
1738
    max_time = 0
1739
    done = True
1740
    cumul_degraded = False
1741
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1742
    if rstats.failed or not rstats.data:
1743
      lu.LogWarning("Can't get any data from node %s", node)
1744
      retries += 1
1745
      if retries >= 10:
1746
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1747
                                 " aborting." % node)
1748
      time.sleep(6)
1749
      continue
1750
    rstats = rstats.data
1751
    retries = 0
1752
    for i, mstat in enumerate(rstats):
1753
      if mstat is None:
1754
        lu.LogWarning("Can't compute data for node %s/%s",
1755
                           node, instance.disks[i].iv_name)
1756
        continue
1757
      # we ignore the ldisk parameter
1758
      perc_done, est_time, is_degraded, _ = mstat
1759
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1760
      if perc_done is not None:
1761
        done = False
1762
        if est_time is not None:
1763
          rem_time = "%d estimated seconds remaining" % est_time
1764
          max_time = est_time
1765
        else:
1766
          rem_time = "no time estimate"
1767
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1768
                        (instance.disks[i].iv_name, perc_done, rem_time))
1769

    
1770
    # if we're done but degraded, let's do a few small retries, to
1771
    # make sure we see a stable and not transient situation; therefore
1772
    # we force restart of the loop
1773
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
1774
      logging.info("Degraded disks found, %d retries left", degr_retries)
1775
      degr_retries -= 1
1776
      time.sleep(1)
1777
      continue
1778

    
1779
    if done or oneshot:
1780
      break
1781

    
1782
    time.sleep(min(60, max_time))
1783

    
1784
  if done:
1785
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1786
  return not cumul_degraded
1787

    
1788

    
1789
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1790
  """Check that mirrors are not degraded.
1791

1792
  The ldisk parameter, if True, will change the test from the
1793
  is_degraded attribute (which represents overall non-ok status for
1794
  the device(s)) to the ldisk (representing the local storage status).
1795

1796
  """
1797
  lu.cfg.SetDiskID(dev, node)
1798
  if ldisk:
1799
    idx = 6
1800
  else:
1801
    idx = 5
1802

    
1803
  result = True
1804
  if on_primary or dev.AssembleOnSecondary():
1805
    rstats = lu.rpc.call_blockdev_find(node, dev)
1806
    msg = rstats.RemoteFailMsg()
1807
    if msg:
1808
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1809
      result = False
1810
    elif not rstats.payload:
1811
      lu.LogWarning("Can't find disk on node %s", node)
1812
      result = False
1813
    else:
1814
      result = result and (not rstats.payload[idx])
1815
  if dev.children:
1816
    for child in dev.children:
1817
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1818

    
1819
  return result
1820

    
1821

    
1822
class LUDiagnoseOS(NoHooksLU):
1823
  """Logical unit for OS diagnose/query.
1824

1825
  """
1826
  _OP_REQP = ["output_fields", "names"]
1827
  REQ_BGL = False
1828
  _FIELDS_STATIC = utils.FieldSet()
1829
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1830

    
1831
  def ExpandNames(self):
1832
    if self.op.names:
1833
      raise errors.OpPrereqError("Selective OS query not supported")
1834

    
1835
    _CheckOutputFields(static=self._FIELDS_STATIC,
1836
                       dynamic=self._FIELDS_DYNAMIC,
1837
                       selected=self.op.output_fields)
1838

    
1839
    # Lock all nodes, in shared mode
1840
    # Temporary removal of locks, should be reverted later
1841
    # TODO: reintroduce locks when they are lighter-weight
1842
    self.needed_locks = {}
1843
    #self.share_locks[locking.LEVEL_NODE] = 1
1844
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1845

    
1846
  def CheckPrereq(self):
1847
    """Check prerequisites.
1848

1849
    """
1850

    
1851
  @staticmethod
1852
  def _DiagnoseByOS(node_list, rlist):
1853
    """Remaps a per-node return list into an a per-os per-node dictionary
1854

1855
    @param node_list: a list with the names of all nodes
1856
    @param rlist: a map with node names as keys and OS objects as values
1857

1858
    @rtype: dict
1859
    @return: a dictionary with osnames as keys and as value another map, with
1860
        nodes as keys and list of OS objects as values, eg::
1861

1862
          {"debian-etch": {"node1": [<object>,...],
1863
                           "node2": [<object>,]}
1864
          }
1865

1866
    """
1867
    all_os = {}
1868
    # we build here the list of nodes that didn't fail the RPC (at RPC
1869
    # level), so that nodes with a non-responding node daemon don't
1870
    # make all OSes invalid
1871
    good_nodes = [node_name for node_name in rlist
1872
                  if not rlist[node_name].failed]
1873
    for node_name, nr in rlist.iteritems():
1874
      if nr.failed or not nr.data:
1875
        continue
1876
      for os_obj in nr.data:
1877
        if os_obj.name not in all_os:
1878
          # build a list of nodes for this os containing empty lists
1879
          # for each node in node_list
1880
          all_os[os_obj.name] = {}
1881
          for nname in good_nodes:
1882
            all_os[os_obj.name][nname] = []
1883
        all_os[os_obj.name][node_name].append(os_obj)
1884
    return all_os
1885

    
1886
  def Exec(self, feedback_fn):
1887
    """Compute the list of OSes.
1888

1889
    """
1890
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
1891
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1892
    if node_data == False:
1893
      raise errors.OpExecError("Can't gather the list of OSes")
1894
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1895
    output = []
1896
    for os_name, os_data in pol.iteritems():
1897
      row = []
1898
      for field in self.op.output_fields:
1899
        if field == "name":
1900
          val = os_name
1901
        elif field == "valid":
1902
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1903
        elif field == "node_status":
1904
          val = {}
1905
          for node_name, nos_list in os_data.iteritems():
1906
            val[node_name] = [(v.status, v.path) for v in nos_list]
1907
        else:
1908
          raise errors.ParameterError(field)
1909
        row.append(val)
1910
      output.append(row)
1911

    
1912
    return output
1913

    
1914

    
1915
class LURemoveNode(LogicalUnit):
1916
  """Logical unit for removing a node.
1917

1918
  """
1919
  HPATH = "node-remove"
1920
  HTYPE = constants.HTYPE_NODE
1921
  _OP_REQP = ["node_name"]
1922

    
1923
  def BuildHooksEnv(self):
1924
    """Build hooks env.
1925

1926
    This doesn't run on the target node in the pre phase as a failed
1927
    node would then be impossible to remove.
1928

1929
    """
1930
    env = {
1931
      "OP_TARGET": self.op.node_name,
1932
      "NODE_NAME": self.op.node_name,
1933
      }
1934
    all_nodes = self.cfg.GetNodeList()
1935
    all_nodes.remove(self.op.node_name)
1936
    return env, all_nodes, all_nodes
1937

    
1938
  def CheckPrereq(self):
1939
    """Check prerequisites.
1940

1941
    This checks:
1942
     - the node exists in the configuration
1943
     - it does not have primary or secondary instances
1944
     - it's not the master
1945

1946
    Any errors are signaled by raising errors.OpPrereqError.
1947

1948
    """
1949
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1950
    if node is None:
1951
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1952

    
1953
    instance_list = self.cfg.GetInstanceList()
1954

    
1955
    masternode = self.cfg.GetMasterNode()
1956
    if node.name == masternode:
1957
      raise errors.OpPrereqError("Node is the master node,"
1958
                                 " you need to failover first.")
1959

    
1960
    for instance_name in instance_list:
1961
      instance = self.cfg.GetInstanceInfo(instance_name)
1962
      if node.name in instance.all_nodes:
1963
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1964
                                   " please remove first." % instance_name)
1965
    self.op.node_name = node.name
1966
    self.node = node
1967

    
1968
  def Exec(self, feedback_fn):
1969
    """Removes the node from the cluster.
1970

1971
    """
1972
    node = self.node
1973
    logging.info("Stopping the node daemon and removing configs from node %s",
1974
                 node.name)
1975

    
1976
    self.context.RemoveNode(node.name)
1977

    
1978
    self.rpc.call_node_leave_cluster(node.name)
1979

    
1980
    # Promote nodes to master candidate as needed
1981
    _AdjustCandidatePool(self)
1982

    
1983

    
1984
class LUQueryNodes(NoHooksLU):
1985
  """Logical unit for querying nodes.
1986

1987
  """
1988
  _OP_REQP = ["output_fields", "names", "use_locking"]
1989
  REQ_BGL = False
1990
  _FIELDS_DYNAMIC = utils.FieldSet(
1991
    "dtotal", "dfree",
1992
    "mtotal", "mnode", "mfree",
1993
    "bootid",
1994
    "ctotal", "cnodes", "csockets",
1995
    )
1996

    
1997
  _FIELDS_STATIC = utils.FieldSet(
1998
    "name", "pinst_cnt", "sinst_cnt",
1999
    "pinst_list", "sinst_list",
2000
    "pip", "sip", "tags",
2001
    "serial_no",
2002
    "master_candidate",
2003
    "master",
2004
    "offline",
2005
    "drained",
2006
    "role",
2007
    )
2008

    
2009
  def ExpandNames(self):
2010
    _CheckOutputFields(static=self._FIELDS_STATIC,
2011
                       dynamic=self._FIELDS_DYNAMIC,
2012
                       selected=self.op.output_fields)
2013

    
2014
    self.needed_locks = {}
2015
    self.share_locks[locking.LEVEL_NODE] = 1
2016

    
2017
    if self.op.names:
2018
      self.wanted = _GetWantedNodes(self, self.op.names)
2019
    else:
2020
      self.wanted = locking.ALL_SET
2021

    
2022
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
2023
    self.do_locking = self.do_node_query and self.op.use_locking
2024
    if self.do_locking:
2025
      # if we don't request only static fields, we need to lock the nodes
2026
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
2027

    
2028

    
2029
  def CheckPrereq(self):
2030
    """Check prerequisites.
2031

2032
    """
2033
    # The validation of the node list is done in the _GetWantedNodes,
2034
    # if non empty, and if empty, there's no validation to do
2035
    pass
2036

    
2037
  def Exec(self, feedback_fn):
2038
    """Computes the list of nodes and their attributes.
2039

2040
    """
2041
    all_info = self.cfg.GetAllNodesInfo()
2042
    if self.do_locking:
2043
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
2044
    elif self.wanted != locking.ALL_SET:
2045
      nodenames = self.wanted
2046
      missing = set(nodenames).difference(all_info.keys())
2047
      if missing:
2048
        raise errors.OpExecError(
2049
          "Some nodes were removed before retrieving their data: %s" % missing)
2050
    else:
2051
      nodenames = all_info.keys()
2052

    
2053
    nodenames = utils.NiceSort(nodenames)
2054
    nodelist = [all_info[name] for name in nodenames]
2055

    
2056
    # begin data gathering
2057

    
2058
    if self.do_node_query:
2059
      live_data = {}
2060
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
2061
                                          self.cfg.GetHypervisorType())
2062
      for name in nodenames:
2063
        nodeinfo = node_data[name]
2064
        if not nodeinfo.failed and nodeinfo.data:
2065
          nodeinfo = nodeinfo.data
2066
          fn = utils.TryConvert
2067
          live_data[name] = {
2068
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
2069
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
2070
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
2071
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
2072
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
2073
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
2074
            "bootid": nodeinfo.get('bootid', None),
2075
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
2076
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
2077
            }
2078
        else:
2079
          live_data[name] = {}
2080
    else:
2081
      live_data = dict.fromkeys(nodenames, {})
2082

    
2083
    node_to_primary = dict([(name, set()) for name in nodenames])
2084
    node_to_secondary = dict([(name, set()) for name in nodenames])
2085

    
2086
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
2087
                             "sinst_cnt", "sinst_list"))
2088
    if inst_fields & frozenset(self.op.output_fields):
2089
      inst_data = self.cfg.GetAllInstancesInfo()
2090

    
2091
      for instance_name, inst in inst_data.items():
2092
        if inst.primary_node in node_to_primary:
2093
          node_to_primary[inst.primary_node].add(inst.name)
2094
        for secnode in inst.secondary_nodes:
2095
          if secnode in node_to_secondary:
2096
            node_to_secondary[secnode].add(inst.name)
2097

    
2098
    master_node = self.cfg.GetMasterNode()
2099

    
2100
    # end data gathering
2101

    
2102
    output = []
2103
    for node in nodelist:
2104
      node_output = []
2105
      for field in self.op.output_fields:
2106
        if field == "name":
2107
          val = node.name
2108
        elif field == "pinst_list":
2109
          val = list(node_to_primary[node.name])
2110
        elif field == "sinst_list":
2111
          val = list(node_to_secondary[node.name])
2112
        elif field == "pinst_cnt":
2113
          val = len(node_to_primary[node.name])
2114
        elif field == "sinst_cnt":
2115
          val = len(node_to_secondary[node.name])
2116
        elif field == "pip":
2117
          val = node.primary_ip
2118
        elif field == "sip":
2119
          val = node.secondary_ip
2120
        elif field == "tags":
2121
          val = list(node.GetTags())
2122
        elif field == "serial_no":
2123
          val = node.serial_no
2124
        elif field == "master_candidate":
2125
          val = node.master_candidate
2126
        elif field == "master":
2127
          val = node.name == master_node
2128
        elif field == "offline":
2129
          val = node.offline
2130
        elif field == "drained":
2131
          val = node.drained
2132
        elif self._FIELDS_DYNAMIC.Matches(field):
2133
          val = live_data[node.name].get(field, None)
2134
        elif field == "role":
2135
          if node.name == master_node:
2136
            val = "M"
2137
          elif node.master_candidate:
2138
            val = "C"
2139
          elif node.drained:
2140
            val = "D"
2141
          elif node.offline:
2142
            val = "O"
2143
          else:
2144
            val = "R"
2145
        else:
2146
          raise errors.ParameterError(field)
2147
        node_output.append(val)
2148
      output.append(node_output)
2149

    
2150
    return output
2151

    
2152

    
2153
class LUQueryNodeVolumes(NoHooksLU):
2154
  """Logical unit for getting volumes on node(s).
2155

2156
  """
2157
  _OP_REQP = ["nodes", "output_fields"]
2158
  REQ_BGL = False
2159
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
2160
  _FIELDS_STATIC = utils.FieldSet("node")
2161

    
2162
  def ExpandNames(self):
2163
    _CheckOutputFields(static=self._FIELDS_STATIC,
2164
                       dynamic=self._FIELDS_DYNAMIC,
2165
                       selected=self.op.output_fields)
2166

    
2167
    self.needed_locks = {}
2168
    self.share_locks[locking.LEVEL_NODE] = 1
2169
    if not self.op.nodes:
2170
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2171
    else:
2172
      self.needed_locks[locking.LEVEL_NODE] = \
2173
        _GetWantedNodes(self, self.op.nodes)
2174

    
2175
  def CheckPrereq(self):
2176
    """Check prerequisites.
2177

2178
    This checks that the fields required are valid output fields.
2179

2180
    """
2181
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
2182

    
2183
  def Exec(self, feedback_fn):
2184
    """Computes the list of nodes and their attributes.
2185

2186
    """
2187
    nodenames = self.nodes
2188
    volumes = self.rpc.call_node_volumes(nodenames)
2189

    
2190
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
2191
             in self.cfg.GetInstanceList()]
2192

    
2193
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
2194

    
2195
    output = []
2196
    for node in nodenames:
2197
      if node not in volumes or volumes[node].failed or not volumes[node].data:
2198
        continue
2199

    
2200
      node_vols = volumes[node].data[:]
2201
      node_vols.sort(key=lambda vol: vol['dev'])
2202

    
2203
      for vol in node_vols:
2204
        node_output = []
2205
        for field in self.op.output_fields:
2206
          if field == "node":
2207
            val = node
2208
          elif field == "phys":
2209
            val = vol['dev']
2210
          elif field == "vg":
2211
            val = vol['vg']
2212
          elif field == "name":
2213
            val = vol['name']
2214
          elif field == "size":
2215
            val = int(float(vol['size']))
2216
          elif field == "instance":
2217
            for inst in ilist:
2218
              if node not in lv_by_node[inst]:
2219
                continue
2220
              if vol['name'] in lv_by_node[inst][node]:
2221
                val = inst.name
2222
                break
2223
            else:
2224
              val = '-'
2225
          else:
2226
            raise errors.ParameterError(field)
2227
          node_output.append(str(val))
2228

    
2229
        output.append(node_output)
2230

    
2231
    return output
2232

    
2233

    
2234
class LUAddNode(LogicalUnit):
2235
  """Logical unit for adding node to the cluster.
2236

2237
  """
2238
  HPATH = "node-add"
2239
  HTYPE = constants.HTYPE_NODE
2240
  _OP_REQP = ["node_name"]
2241

    
2242
  def BuildHooksEnv(self):
2243
    """Build hooks env.
2244

2245
    This will run on all nodes before, and on all nodes + the new node after.
2246

2247
    """
2248
    env = {
2249
      "OP_TARGET": self.op.node_name,
2250
      "NODE_NAME": self.op.node_name,
2251
      "NODE_PIP": self.op.primary_ip,
2252
      "NODE_SIP": self.op.secondary_ip,
2253
      }
2254
    nodes_0 = self.cfg.GetNodeList()
2255
    nodes_1 = nodes_0 + [self.op.node_name, ]
2256
    return env, nodes_0, nodes_1
2257

    
2258
  def CheckPrereq(self):
2259
    """Check prerequisites.
2260

2261
    This checks:
2262
     - the new node is not already in the config
2263
     - it is resolvable
2264
     - its parameters (single/dual homed) matches the cluster
2265

2266
    Any errors are signaled by raising errors.OpPrereqError.
2267

2268
    """
2269
    node_name = self.op.node_name
2270
    cfg = self.cfg
2271

    
2272
    dns_data = utils.HostInfo(node_name)
2273

    
2274
    node = dns_data.name
2275
    primary_ip = self.op.primary_ip = dns_data.ip
2276
    secondary_ip = getattr(self.op, "secondary_ip", None)
2277
    if secondary_ip is None:
2278
      secondary_ip = primary_ip
2279
    if not utils.IsValidIP(secondary_ip):
2280
      raise errors.OpPrereqError("Invalid secondary IP given")
2281
    self.op.secondary_ip = secondary_ip
2282

    
2283
    node_list = cfg.GetNodeList()
2284
    if not self.op.readd and node in node_list:
2285
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2286
                                 node)
2287
    elif self.op.readd and node not in node_list:
2288
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2289

    
2290
    for existing_node_name in node_list:
2291
      existing_node = cfg.GetNodeInfo(existing_node_name)
2292

    
2293
      if self.op.readd and node == existing_node_name:
2294
        if (existing_node.primary_ip != primary_ip or
2295
            existing_node.secondary_ip != secondary_ip):
2296
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2297
                                     " address configuration as before")
2298
        continue
2299

    
2300
      if (existing_node.primary_ip == primary_ip or
2301
          existing_node.secondary_ip == primary_ip or
2302
          existing_node.primary_ip == secondary_ip or
2303
          existing_node.secondary_ip == secondary_ip):
2304
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2305
                                   " existing node %s" % existing_node.name)
2306

    
2307
    # check that the type of the node (single versus dual homed) is the
2308
    # same as for the master
2309
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2310
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2311
    newbie_singlehomed = secondary_ip == primary_ip
2312
    if master_singlehomed != newbie_singlehomed:
2313
      if master_singlehomed:
2314
        raise errors.OpPrereqError("The master has no private ip but the"
2315
                                   " new node has one")
2316
      else:
2317
        raise errors.OpPrereqError("The master has a private ip but the"
2318
                                   " new node doesn't have one")
2319

    
2320
    # checks reachability
2321
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2322
      raise errors.OpPrereqError("Node not reachable by ping")
2323

    
2324
    if not newbie_singlehomed:
2325
      # check reachability from my secondary ip to newbie's secondary ip
2326
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2327
                           source=myself.secondary_ip):
2328
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2329
                                   " based ping to noded port")
2330

    
2331
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2332
    if self.op.readd:
2333
      exceptions = [node]
2334
    else:
2335
      exceptions = []
2336
    mc_now, mc_max = self.cfg.GetMasterCandidateStats(exceptions)
2337
    # the new node will increase mc_max with one, so:
2338
    mc_max = min(mc_max + 1, cp_size)
2339
    self.master_candidate = mc_now < mc_max
2340

    
2341
    if self.op.readd:
2342
      self.new_node = self.cfg.GetNodeInfo(node)
2343
      assert self.new_node is not None, "Can't retrieve locked node %s" % node
2344
    else:
2345
      self.new_node = objects.Node(name=node,
2346
                                   primary_ip=primary_ip,
2347
                                   secondary_ip=secondary_ip,
2348
                                   master_candidate=self.master_candidate,
2349
                                   offline=False, drained=False)
2350

    
2351
  def Exec(self, feedback_fn):
2352
    """Adds the new node to the cluster.
2353

2354
    """
2355
    new_node = self.new_node
2356
    node = new_node.name
2357

    
2358
    # for re-adds, reset the offline/drained/master-candidate flags;
2359
    # we need to reset here, otherwise offline would prevent RPC calls
2360
    # later in the procedure; this also means that if the re-add
2361
    # fails, we are left with a non-offlined, broken node
2362
    if self.op.readd:
2363
      new_node.drained = new_node.offline = False
2364
      self.LogInfo("Readding a node, the offline/drained flags were reset")
2365
      # if we demote the node, we do cleanup later in the procedure
2366
      new_node.master_candidate = self.master_candidate
2367

    
2368
    # notify the user about any possible mc promotion
2369
    if new_node.master_candidate:
2370
      self.LogInfo("Node will be a master candidate")
2371

    
2372
    # check connectivity
2373
    result = self.rpc.call_version([node])[node]
2374
    result.Raise()
2375
    if result.data:
2376
      if constants.PROTOCOL_VERSION == result.data:
2377
        logging.info("Communication to node %s fine, sw version %s match",
2378
                     node, result.data)
2379
      else:
2380
        raise errors.OpExecError("Version mismatch master version %s,"
2381
                                 " node version %s" %
2382
                                 (constants.PROTOCOL_VERSION, result.data))
2383
    else:
2384
      raise errors.OpExecError("Cannot get version from the new node")
2385

    
2386
    # setup ssh on node
2387
    logging.info("Copy ssh key to node %s", node)
2388
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2389
    keyarray = []
2390
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2391
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2392
                priv_key, pub_key]
2393

    
2394
    for i in keyfiles:
2395
      f = open(i, 'r')
2396
      try:
2397
        keyarray.append(f.read())
2398
      finally:
2399
        f.close()
2400

    
2401
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2402
                                    keyarray[2],
2403
                                    keyarray[3], keyarray[4], keyarray[5])
2404

    
2405
    msg = result.RemoteFailMsg()
2406
    if msg:
2407
      raise errors.OpExecError("Cannot transfer ssh keys to the"
2408
                               " new node: %s" % msg)
2409

    
2410
    # Add node to our /etc/hosts, and add key to known_hosts
2411
    if self.cfg.GetClusterInfo().modify_etc_hosts:
2412
      utils.AddHostToEtcHosts(new_node.name)
2413

    
2414
    if new_node.secondary_ip != new_node.primary_ip:
2415
      result = self.rpc.call_node_has_ip_address(new_node.name,
2416
                                                 new_node.secondary_ip)
2417
      if result.failed or not result.data:
2418
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2419
                                 " you gave (%s). Please fix and re-run this"
2420
                                 " command." % new_node.secondary_ip)
2421

    
2422
    node_verify_list = [self.cfg.GetMasterNode()]
2423
    node_verify_param = {
2424
      'nodelist': [node],
2425
      # TODO: do a node-net-test as well?
2426
    }
2427

    
2428
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2429
                                       self.cfg.GetClusterName())
2430
    for verifier in node_verify_list:
2431
      if result[verifier].failed or not result[verifier].data:
2432
        raise errors.OpExecError("Cannot communicate with %s's node daemon"
2433
                                 " for remote verification" % verifier)
2434
      if result[verifier].data['nodelist']:
2435
        for failed in result[verifier].data['nodelist']:
2436
          feedback_fn("ssh/hostname verification failed"
2437
                      " (checking from %s): %s" %
2438
                      (verifier, result[verifier].data['nodelist'][failed]))
2439
        raise errors.OpExecError("ssh/hostname verification failed.")
2440

    
2441
    # Distribute updated /etc/hosts and known_hosts to all nodes,
2442
    # including the node just added
2443
    myself = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
2444
    dist_nodes = self.cfg.GetNodeList()
2445
    if not self.op.readd:
2446
      dist_nodes.append(node)
2447
    if myself.name in dist_nodes:
2448
      dist_nodes.remove(myself.name)
2449

    
2450
    logging.debug("Copying hosts and known_hosts to all nodes")
2451
    for fname in (constants.ETC_HOSTS, constants.SSH_KNOWN_HOSTS_FILE):
2452
      result = self.rpc.call_upload_file(dist_nodes, fname)
2453
      for to_node, to_result in result.iteritems():
2454
        if to_result.failed or not to_result.data:
2455
          logging.error("Copy of file %s to node %s failed", fname, to_node)
2456

    
2457
    to_copy = []
2458
    enabled_hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
2459
    if constants.HTS_COPY_VNC_PASSWORD.intersection(enabled_hypervisors):
2460
      to_copy.append(constants.VNC_PASSWORD_FILE)
2461

    
2462
    for fname in to_copy:
2463
      result = self.rpc.call_upload_file([node], fname)
2464
      if result[node].failed or not result[node]:
2465
        logging.error("Could not copy file %s to node %s", fname, node)
2466

    
2467
    if self.op.readd:
2468
      self.context.ReaddNode(new_node)
2469
      # make sure we redistribute the config
2470
      self.cfg.Update(new_node)
2471
      # and make sure the new node will not have old files around
2472
      if not new_node.master_candidate:
2473
        result = self.rpc.call_node_demote_from_mc(new_node.name)
2474
        msg = result.RemoteFailMsg()
2475
        if msg:
2476
          self.LogWarning("Node failed to demote itself from master"
2477
                          " candidate status: %s" % msg)
2478
    else:
2479
      self.context.AddNode(new_node)
2480

    
2481

    
2482
class LUSetNodeParams(LogicalUnit):
2483
  """Modifies the parameters of a node.
2484

2485
  """
2486
  HPATH = "node-modify"
2487
  HTYPE = constants.HTYPE_NODE
2488
  _OP_REQP = ["node_name"]
2489
  REQ_BGL = False
2490

    
2491
  def CheckArguments(self):
2492
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2493
    if node_name is None:
2494
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2495
    self.op.node_name = node_name
2496
    _CheckBooleanOpField(self.op, 'master_candidate')
2497
    _CheckBooleanOpField(self.op, 'offline')
2498
    _CheckBooleanOpField(self.op, 'drained')
2499
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
2500
    if all_mods.count(None) == 3:
2501
      raise errors.OpPrereqError("Please pass at least one modification")
2502
    if all_mods.count(True) > 1:
2503
      raise errors.OpPrereqError("Can't set the node into more than one"
2504
                                 " state at the same time")
2505

    
2506
  def ExpandNames(self):
2507
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2508

    
2509
  def BuildHooksEnv(self):
2510
    """Build hooks env.
2511

2512
    This runs on the master node.
2513

2514
    """
2515
    env = {
2516
      "OP_TARGET": self.op.node_name,
2517
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2518
      "OFFLINE": str(self.op.offline),
2519
      "DRAINED": str(self.op.drained),
2520
      }
2521
    nl = [self.cfg.GetMasterNode(),
2522
          self.op.node_name]
2523
    return env, nl, nl
2524

    
2525
  def CheckPrereq(self):
2526
    """Check prerequisites.
2527

2528
    This only checks the instance list against the existing names.
2529

2530
    """
2531
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2532

    
2533
    if (self.op.master_candidate is not None or
2534
        self.op.drained is not None or
2535
        self.op.offline is not None):
2536
      # we can't change the master's node flags
2537
      if self.op.node_name == self.cfg.GetMasterNode():
2538
        raise errors.OpPrereqError("The master role can be changed"
2539
                                   " only via masterfailover")
2540

    
2541
    if ((self.op.master_candidate == False or self.op.offline == True or
2542
         self.op.drained == True) and node.master_candidate):
2543
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2544
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2545
      if num_candidates <= cp_size:
2546
        msg = ("Not enough master candidates (desired"
2547
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2548
        if self.op.force:
2549
          self.LogWarning(msg)
2550
        else:
2551
          raise errors.OpPrereqError(msg)
2552

    
2553
    if (self.op.master_candidate == True and
2554
        ((node.offline and not self.op.offline == False) or
2555
         (node.drained and not self.op.drained == False))):
2556
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
2557
                                 " to master_candidate" % node.name)
2558

    
2559
    return
2560

    
2561
  def Exec(self, feedback_fn):
2562
    """Modifies a node.
2563

2564
    """
2565
    node = self.node
2566

    
2567
    result = []
2568
    changed_mc = False
2569

    
2570
    if self.op.offline is not None:
2571
      node.offline = self.op.offline
2572
      result.append(("offline", str(self.op.offline)))
2573
      if self.op.offline == True:
2574
        if node.master_candidate:
2575
          node.master_candidate = False
2576
          changed_mc = True
2577
          result.append(("master_candidate", "auto-demotion due to offline"))
2578
        if node.drained:
2579
          node.drained = False
2580
          result.append(("drained", "clear drained status due to offline"))
2581

    
2582
    if self.op.master_candidate is not None:
2583
      node.master_candidate = self.op.master_candidate
2584
      changed_mc = True
2585
      result.append(("master_candidate", str(self.op.master_candidate)))
2586
      if self.op.master_candidate == False:
2587
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2588
        msg = rrc.RemoteFailMsg()
2589
        if msg:
2590
          self.LogWarning("Node failed to demote itself: %s" % msg)
2591

    
2592
    if self.op.drained is not None:
2593
      node.drained = self.op.drained
2594
      result.append(("drained", str(self.op.drained)))
2595
      if self.op.drained == True:
2596
        if node.master_candidate:
2597
          node.master_candidate = False
2598
          changed_mc = True
2599
          result.append(("master_candidate", "auto-demotion due to drain"))
2600
          rrc = self.rpc.call_node_demote_from_mc(node.name)
2601
          msg = rrc.RemoteFailMsg()
2602
          if msg:
2603
            self.LogWarning("Node failed to demote itself: %s" % msg)
2604
        if node.offline:
2605
          node.offline = False
2606
          result.append(("offline", "clear offline status due to drain"))
2607

    
2608
    # this will trigger configuration file update, if needed
2609
    self.cfg.Update(node)
2610
    # this will trigger job queue propagation or cleanup
2611
    if changed_mc:
2612
      self.context.ReaddNode(node)
2613

    
2614
    return result
2615

    
2616

    
2617
class LUQueryClusterInfo(NoHooksLU):
2618
  """Query cluster configuration.
2619

2620
  """
2621
  _OP_REQP = []
2622
  REQ_BGL = False
2623

    
2624
  def ExpandNames(self):
2625
    self.needed_locks = {}
2626

    
2627
  def CheckPrereq(self):
2628
    """No prerequsites needed for this LU.
2629

2630
    """
2631
    pass
2632

    
2633
  def Exec(self, feedback_fn):
2634
    """Return cluster config.
2635

2636
    """
2637
    cluster = self.cfg.GetClusterInfo()
2638
    result = {
2639
      "software_version": constants.RELEASE_VERSION,
2640
      "protocol_version": constants.PROTOCOL_VERSION,
2641
      "config_version": constants.CONFIG_VERSION,
2642
      "os_api_version": constants.OS_API_VERSION,
2643
      "export_version": constants.EXPORT_VERSION,
2644
      "architecture": (platform.architecture()[0], platform.machine()),
2645
      "name": cluster.cluster_name,
2646
      "master": cluster.master_node,
2647
      "default_hypervisor": cluster.default_hypervisor,
2648
      "enabled_hypervisors": cluster.enabled_hypervisors,
2649
      "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
2650
                        for hypervisor_name in cluster.enabled_hypervisors]),
2651
      "beparams": cluster.beparams,
2652
      "candidate_pool_size": cluster.candidate_pool_size,
2653
      "default_bridge": cluster.default_bridge,
2654
      "master_netdev": cluster.master_netdev,
2655
      "volume_group_name": cluster.volume_group_name,
2656
      "file_storage_dir": cluster.file_storage_dir,
2657
      "tags": list(cluster.GetTags()),
2658
      }
2659

    
2660
    return result
2661

    
2662

    
2663
class LUQueryConfigValues(NoHooksLU):
2664
  """Return configuration values.
2665

2666
  """
2667
  _OP_REQP = []
2668
  REQ_BGL = False
2669
  _FIELDS_DYNAMIC = utils.FieldSet()
2670
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2671

    
2672
  def ExpandNames(self):
2673
    self.needed_locks = {}
2674

    
2675
    _CheckOutputFields(static=self._FIELDS_STATIC,
2676
                       dynamic=self._FIELDS_DYNAMIC,
2677
                       selected=self.op.output_fields)
2678

    
2679
  def CheckPrereq(self):
2680
    """No prerequisites.
2681

2682
    """
2683
    pass
2684

    
2685
  def Exec(self, feedback_fn):
2686
    """Dump a representation of the cluster config to the standard output.
2687

2688
    """
2689
    values = []
2690
    for field in self.op.output_fields:
2691
      if field == "cluster_name":
2692
        entry = self.cfg.GetClusterName()
2693
      elif field == "master_node":
2694
        entry = self.cfg.GetMasterNode()
2695
      elif field == "drain_flag":
2696
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2697
      else:
2698
        raise errors.ParameterError(field)
2699
      values.append(entry)
2700
    return values
2701

    
2702

    
2703
class LUActivateInstanceDisks(NoHooksLU):
2704
  """Bring up an instance's disks.
2705

2706
  """
2707
  _OP_REQP = ["instance_name"]
2708
  REQ_BGL = False
2709

    
2710
  def ExpandNames(self):
2711
    self._ExpandAndLockInstance()
2712
    self.needed_locks[locking.LEVEL_NODE] = []
2713
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2714

    
2715
  def DeclareLocks(self, level):
2716
    if level == locking.LEVEL_NODE:
2717
      self._LockInstancesNodes()
2718

    
2719
  def CheckPrereq(self):
2720
    """Check prerequisites.
2721

2722
    This checks that the instance is in the cluster.
2723

2724
    """
2725
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2726
    assert self.instance is not None, \
2727
      "Cannot retrieve locked instance %s" % self.op.instance_name
2728
    _CheckNodeOnline(self, self.instance.primary_node)
2729
    if not hasattr(self.op, "ignore_size"):
2730
      self.op.ignore_size = False
2731

    
2732
  def Exec(self, feedback_fn):
2733
    """Activate the disks.
2734

2735
    """
2736
    disks_ok, disks_info = \
2737
              _AssembleInstanceDisks(self, self.instance,
2738
                                     ignore_size=self.op.ignore_size)
2739
    if not disks_ok:
2740
      raise errors.OpExecError("Cannot activate block devices")
2741

    
2742
    return disks_info
2743

    
2744

    
2745
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False,
2746
                           ignore_size=False):
2747
  """Prepare the block devices for an instance.
2748

2749
  This sets up the block devices on all nodes.
2750

2751
  @type lu: L{LogicalUnit}
2752
  @param lu: the logical unit on whose behalf we execute
2753
  @type instance: L{objects.Instance}
2754
  @param instance: the instance for whose disks we assemble
2755
  @type ignore_secondaries: boolean
2756
  @param ignore_secondaries: if true, errors on secondary nodes
2757
      won't result in an error return from the function
2758
  @type ignore_size: boolean
2759
  @param ignore_size: if true, the current known size of the disk
2760
      will not be used during the disk activation, useful for cases
2761
      when the size is wrong
2762
  @return: False if the operation failed, otherwise a list of
2763
      (host, instance_visible_name, node_visible_name)
2764
      with the mapping from node devices to instance devices
2765

2766
  """
2767
  device_info = []
2768
  disks_ok = True
2769
  iname = instance.name
2770
  # With the two passes mechanism we try to reduce the window of
2771
  # opportunity for the race condition of switching DRBD to primary
2772
  # before handshaking occured, but we do not eliminate it
2773

    
2774
  # The proper fix would be to wait (with some limits) until the
2775
  # connection has been made and drbd transitions from WFConnection
2776
  # into any other network-connected state (Connected, SyncTarget,
2777
  # SyncSource, etc.)
2778

    
2779
  # 1st pass, assemble on all nodes in secondary mode
2780
  for inst_disk in instance.disks:
2781
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2782
      if ignore_size:
2783
        node_disk = node_disk.Copy()
2784
        node_disk.UnsetSize()
2785
      lu.cfg.SetDiskID(node_disk, node)
2786
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2787
      msg = result.RemoteFailMsg()
2788
      if msg:
2789
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2790
                           " (is_primary=False, pass=1): %s",
2791
                           inst_disk.iv_name, node, msg)
2792
        if not ignore_secondaries:
2793
          disks_ok = False
2794

    
2795
  # FIXME: race condition on drbd migration to primary
2796

    
2797
  # 2nd pass, do only the primary node
2798
  for inst_disk in instance.disks:
2799
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2800
      if node != instance.primary_node:
2801
        continue
2802
      if ignore_size:
2803
        node_disk = node_disk.Copy()
2804
        node_disk.UnsetSize()
2805
      lu.cfg.SetDiskID(node_disk, node)
2806
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2807
      msg = result.RemoteFailMsg()
2808
      if msg:
2809
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2810
                           " (is_primary=True, pass=2): %s",
2811
                           inst_disk.iv_name, node, msg)
2812
        disks_ok = False
2813
    device_info.append((instance.primary_node, inst_disk.iv_name,
2814
                        result.payload))
2815

    
2816
  # leave the disks configured for the primary node
2817
  # this is a workaround that would be fixed better by
2818
  # improving the logical/physical id handling
2819
  for disk in instance.disks:
2820
    lu.cfg.SetDiskID(disk, instance.primary_node)
2821

    
2822
  return disks_ok, device_info
2823

    
2824

    
2825
def _StartInstanceDisks(lu, instance, force):
2826
  """Start the disks of an instance.
2827

2828
  """
2829
  disks_ok, _ = _AssembleInstanceDisks(lu, instance,
2830
                                           ignore_secondaries=force)
2831
  if not disks_ok:
2832
    _ShutdownInstanceDisks(lu, instance)
2833
    if force is not None and not force:
2834
      lu.proc.LogWarning("", hint="If the message above refers to a"
2835
                         " secondary node,"
2836
                         " you can retry the operation using '--force'.")
2837
    raise errors.OpExecError("Disk consistency error")
2838

    
2839

    
2840
class LUDeactivateInstanceDisks(NoHooksLU):
2841
  """Shutdown an instance's disks.
2842

2843
  """
2844
  _OP_REQP = ["instance_name"]
2845
  REQ_BGL = False
2846

    
2847
  def ExpandNames(self):
2848
    self._ExpandAndLockInstance()
2849
    self.needed_locks[locking.LEVEL_NODE] = []
2850
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2851

    
2852
  def DeclareLocks(self, level):
2853
    if level == locking.LEVEL_NODE:
2854
      self._LockInstancesNodes()
2855

    
2856
  def CheckPrereq(self):
2857
    """Check prerequisites.
2858

2859
    This checks that the instance is in the cluster.
2860

2861
    """
2862
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2863
    assert self.instance is not None, \
2864
      "Cannot retrieve locked instance %s" % self.op.instance_name
2865

    
2866
  def Exec(self, feedback_fn):
2867
    """Deactivate the disks
2868

2869
    """
2870
    instance = self.instance
2871
    _SafeShutdownInstanceDisks(self, instance)
2872

    
2873

    
2874
def _SafeShutdownInstanceDisks(lu, instance):
2875
  """Shutdown block devices of an instance.
2876

2877
  This function checks if an instance is running, before calling
2878
  _ShutdownInstanceDisks.
2879

2880
  """
2881
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2882
                                      [instance.hypervisor])
2883
  ins_l = ins_l[instance.primary_node]
2884
  if ins_l.failed or not isinstance(ins_l.data, list):
2885
    raise errors.OpExecError("Can't contact node '%s'" %
2886
                             instance.primary_node)
2887

    
2888
  if instance.name in ins_l.data:
2889
    raise errors.OpExecError("Instance is running, can't shutdown"
2890
                             " block devices.")
2891

    
2892
  _ShutdownInstanceDisks(lu, instance)
2893

    
2894

    
2895
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2896
  """Shutdown block devices of an instance.
2897

2898
  This does the shutdown on all nodes of the instance.
2899

2900
  If the ignore_primary is false, errors on the primary node are
2901
  ignored.
2902

2903
  """
2904
  all_result = True
2905
  for disk in instance.disks:
2906
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2907
      lu.cfg.SetDiskID(top_disk, node)
2908
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2909
      msg = result.RemoteFailMsg()
2910
      if msg:
2911
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2912
                      disk.iv_name, node, msg)
2913
        if not ignore_primary or node != instance.primary_node:
2914
          all_result = False
2915
  return all_result
2916

    
2917

    
2918
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2919
  """Checks if a node has enough free memory.
2920

2921
  This function check if a given node has the needed amount of free
2922
  memory. In case the node has less memory or we cannot get the
2923
  information from the node, this function raise an OpPrereqError
2924
  exception.
2925

2926
  @type lu: C{LogicalUnit}
2927
  @param lu: a logical unit from which we get configuration data
2928
  @type node: C{str}
2929
  @param node: the node to check
2930
  @type reason: C{str}
2931
  @param reason: string to use in the error message
2932
  @type requested: C{int}
2933
  @param requested: the amount of memory in MiB to check for
2934
  @type hypervisor_name: C{str}
2935
  @param hypervisor_name: the hypervisor to ask for memory stats
2936
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2937
      we cannot check the node
2938

2939
  """
2940
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2941
  nodeinfo[node].Raise()
2942
  free_mem = nodeinfo[node].data.get('memory_free')
2943
  if not isinstance(free_mem, int):
2944
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2945
                             " was '%s'" % (node, free_mem))
2946
  if requested > free_mem:
2947
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2948
                             " needed %s MiB, available %s MiB" %
2949
                             (node, reason, requested, free_mem))
2950

    
2951

    
2952
class LUStartupInstance(LogicalUnit):
2953
  """Starts an instance.
2954

2955
  """
2956
  HPATH = "instance-start"
2957
  HTYPE = constants.HTYPE_INSTANCE
2958
  _OP_REQP = ["instance_name", "force"]
2959
  REQ_BGL = False
2960

    
2961
  def ExpandNames(self):
2962
    self._ExpandAndLockInstance()
2963

    
2964
  def BuildHooksEnv(self):
2965
    """Build hooks env.
2966

2967
    This runs on master, primary and secondary nodes of the instance.
2968

2969
    """
2970
    env = {
2971
      "FORCE": self.op.force,
2972
      }
2973
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2974
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2975
    return env, nl, nl
2976

    
2977
  def CheckPrereq(self):
2978
    """Check prerequisites.
2979

2980
    This checks that the instance is in the cluster.
2981

2982
    """
2983
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2984
    assert self.instance is not None, \
2985
      "Cannot retrieve locked instance %s" % self.op.instance_name
2986

    
2987
    # extra beparams
2988
    self.beparams = getattr(self.op, "beparams", {})
2989
    if self.beparams:
2990
      if not isinstance(self.beparams, dict):
2991
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
2992
                                   " dict" % (type(self.beparams), ))
2993
      # fill the beparams dict
2994
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
2995
      self.op.beparams = self.beparams
2996

    
2997
    # extra hvparams
2998
    self.hvparams = getattr(self.op, "hvparams", {})
2999
    if self.hvparams:
3000
      if not isinstance(self.hvparams, dict):
3001
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
3002
                                   " dict" % (type(self.hvparams), ))
3003

    
3004
      # check hypervisor parameter syntax (locally)
3005
      cluster = self.cfg.GetClusterInfo()
3006
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
3007
      filled_hvp = cluster.FillDict(cluster.hvparams[instance.hypervisor],
3008
                                    instance.hvparams)
3009
      filled_hvp.update(self.hvparams)
3010
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
3011
      hv_type.CheckParameterSyntax(filled_hvp)
3012
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
3013
      self.op.hvparams = self.hvparams
3014

    
3015
    _CheckNodeOnline(self, instance.primary_node)
3016

    
3017
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3018
    # check bridges existence
3019
    _CheckInstanceBridgesExist(self, instance)
3020

    
3021
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3022
                                              instance.name,
3023
                                              instance.hypervisor)
3024
    remote_info.Raise()
3025
    if not remote_info.data:
3026
      _CheckNodeFreeMemory(self, instance.primary_node,
3027
                           "starting instance %s" % instance.name,
3028
                           bep[constants.BE_MEMORY], instance.hypervisor)
3029

    
3030
  def Exec(self, feedback_fn):
3031
    """Start the instance.
3032

3033
    """
3034
    instance = self.instance
3035
    force = self.op.force
3036

    
3037
    self.cfg.MarkInstanceUp(instance.name)
3038

    
3039
    node_current = instance.primary_node
3040

    
3041
    _StartInstanceDisks(self, instance, force)
3042

    
3043
    result = self.rpc.call_instance_start(node_current, instance,
3044
                                          self.hvparams, self.beparams)
3045
    msg = result.RemoteFailMsg()
3046
    if msg:
3047
      _ShutdownInstanceDisks(self, instance)
3048
      raise errors.OpExecError("Could not start instance: %s" % msg)
3049

    
3050

    
3051
class LURebootInstance(LogicalUnit):
3052
  """Reboot an instance.
3053

3054
  """
3055
  HPATH = "instance-reboot"
3056
  HTYPE = constants.HTYPE_INSTANCE
3057
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
3058
  REQ_BGL = False
3059

    
3060
  def ExpandNames(self):
3061
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
3062
                                   constants.INSTANCE_REBOOT_HARD,
3063
                                   constants.INSTANCE_REBOOT_FULL]:
3064
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
3065
                                  (constants.INSTANCE_REBOOT_SOFT,
3066
                                   constants.INSTANCE_REBOOT_HARD,
3067
                                   constants.INSTANCE_REBOOT_FULL))
3068
    self._ExpandAndLockInstance()
3069

    
3070
  def BuildHooksEnv(self):
3071
    """Build hooks env.
3072

3073
    This runs on master, primary and secondary nodes of the instance.
3074

3075
    """
3076
    env = {
3077
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
3078
      "REBOOT_TYPE": self.op.reboot_type,
3079
      }
3080
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3081
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3082
    return env, nl, nl
3083

    
3084
  def CheckPrereq(self):
3085
    """Check prerequisites.
3086

3087
    This checks that the instance is in the cluster.
3088

3089
    """
3090
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3091
    assert self.instance is not None, \
3092
      "Cannot retrieve locked instance %s" % self.op.instance_name
3093

    
3094
    _CheckNodeOnline(self, instance.primary_node)
3095

    
3096
    # check bridges existence
3097
    _CheckInstanceBridgesExist(self, instance)
3098

    
3099
  def Exec(self, feedback_fn):
3100
    """Reboot the instance.
3101

3102
    """
3103
    instance = self.instance
3104
    ignore_secondaries = self.op.ignore_secondaries
3105
    reboot_type = self.op.reboot_type
3106

    
3107
    node_current = instance.primary_node
3108

    
3109
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
3110
                       constants.INSTANCE_REBOOT_HARD]:
3111
      for disk in instance.disks:
3112
        self.cfg.SetDiskID(disk, node_current)
3113
      result = self.rpc.call_instance_reboot(node_current, instance,
3114
                                             reboot_type)
3115
      msg = result.RemoteFailMsg()
3116
      if msg:
3117
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
3118
    else:
3119
      result = self.rpc.call_instance_shutdown(node_current, instance)
3120
      msg = result.RemoteFailMsg()
3121
      if msg:
3122
        raise errors.OpExecError("Could not shutdown instance for"
3123
                                 " full reboot: %s" % msg)
3124
      _ShutdownInstanceDisks(self, instance)
3125
      _StartInstanceDisks(self, instance, ignore_secondaries)
3126
      result = self.rpc.call_instance_start(node_current, instance, None, None)
3127
      msg = result.RemoteFailMsg()
3128
      if msg:
3129
        _ShutdownInstanceDisks(self, instance)
3130
        raise errors.OpExecError("Could not start instance for"
3131
                                 " full reboot: %s" % msg)
3132

    
3133
    self.cfg.MarkInstanceUp(instance.name)
3134

    
3135

    
3136
class LUShutdownInstance(LogicalUnit):
3137
  """Shutdown an instance.
3138

3139
  """
3140
  HPATH = "instance-stop"
3141
  HTYPE = constants.HTYPE_INSTANCE
3142
  _OP_REQP = ["instance_name"]
3143
  REQ_BGL = False
3144

    
3145
  def ExpandNames(self):
3146
    self._ExpandAndLockInstance()
3147

    
3148
  def BuildHooksEnv(self):
3149
    """Build hooks env.
3150

3151
    This runs on master, primary and secondary nodes of the instance.
3152

3153
    """
3154
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3155
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3156
    return env, nl, nl
3157

    
3158
  def CheckPrereq(self):
3159
    """Check prerequisites.
3160

3161
    This checks that the instance is in the cluster.
3162

3163
    """
3164
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3165
    assert self.instance is not None, \
3166
      "Cannot retrieve locked instance %s" % self.op.instance_name
3167
    _CheckNodeOnline(self, self.instance.primary_node)
3168

    
3169
  def Exec(self, feedback_fn):
3170
    """Shutdown the instance.
3171

3172
    """
3173
    instance = self.instance
3174
    node_current = instance.primary_node
3175
    self.cfg.MarkInstanceDown(instance.name)
3176
    result = self.rpc.call_instance_shutdown(node_current, instance)
3177
    msg = result.RemoteFailMsg()
3178
    if msg:
3179
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
3180

    
3181
    _ShutdownInstanceDisks(self, instance)
3182

    
3183

    
3184
class LUReinstallInstance(LogicalUnit):
3185
  """Reinstall an instance.
3186

3187
  """
3188
  HPATH = "instance-reinstall"
3189
  HTYPE = constants.HTYPE_INSTANCE
3190
  _OP_REQP = ["instance_name"]
3191
  REQ_BGL = False
3192

    
3193
  def ExpandNames(self):
3194
    self._ExpandAndLockInstance()
3195

    
3196
  def BuildHooksEnv(self):
3197
    """Build hooks env.
3198

3199
    This runs on master, primary and secondary nodes of the instance.
3200

3201
    """
3202
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3203
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3204
    return env, nl, nl
3205

    
3206
  def CheckPrereq(self):
3207
    """Check prerequisites.
3208

3209
    This checks that the instance is in the cluster and is not running.
3210

3211
    """
3212
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3213
    assert instance is not None, \
3214
      "Cannot retrieve locked instance %s" % self.op.instance_name
3215
    _CheckNodeOnline(self, instance.primary_node)
3216

    
3217
    if instance.disk_template == constants.DT_DISKLESS:
3218
      raise errors.OpPrereqError("Instance '%s' has no disks" %
3219
                                 self.op.instance_name)
3220
    if instance.admin_up:
3221
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3222
                                 self.op.instance_name)
3223
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3224
                                              instance.name,
3225
                                              instance.hypervisor)
3226
    remote_info.Raise()
3227
    if remote_info.data:
3228
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3229
                                 (self.op.instance_name,
3230
                                  instance.primary_node))
3231

    
3232
    self.op.os_type = getattr(self.op, "os_type", None)
3233
    if self.op.os_type is not None:
3234
      # OS verification
3235
      pnode = self.cfg.GetNodeInfo(
3236
        self.cfg.ExpandNodeName(instance.primary_node))
3237
      if pnode is None:
3238
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
3239
                                   self.op.pnode)
3240
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
3241
      result.Raise()
3242
      if not isinstance(result.data, objects.OS):
3243
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
3244
                                   " primary node"  % self.op.os_type)
3245

    
3246
    self.instance = instance
3247

    
3248
  def Exec(self, feedback_fn):
3249
    """Reinstall the instance.
3250

3251
    """
3252
    inst = self.instance
3253

    
3254
    if self.op.os_type is not None:
3255
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3256
      inst.os = self.op.os_type
3257
      self.cfg.Update(inst)
3258

    
3259
    _StartInstanceDisks(self, inst, None)
3260
    try:
3261
      feedback_fn("Running the instance OS create scripts...")
3262
      result = self.rpc.call_instance_os_add(inst.primary_node, inst)
3263
      msg = result.RemoteFailMsg()
3264
      if msg:
3265
        raise errors.OpExecError("Could not install OS for instance %s"
3266
                                 " on node %s: %s" %
3267
                                 (inst.name, inst.primary_node, msg))
3268
    finally:
3269
      _ShutdownInstanceDisks(self, inst)
3270

    
3271

    
3272
class LURenameInstance(LogicalUnit):
3273
  """Rename an instance.
3274

3275
  """
3276
  HPATH = "instance-rename"
3277
  HTYPE = constants.HTYPE_INSTANCE
3278
  _OP_REQP = ["instance_name", "new_name"]
3279

    
3280
  def BuildHooksEnv(self):
3281
    """Build hooks env.
3282

3283
    This runs on master, primary and secondary nodes of the instance.
3284

3285
    """
3286
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3287
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3288
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3289
    return env, nl, nl
3290

    
3291
  def CheckPrereq(self):
3292
    """Check prerequisites.
3293

3294
    This checks that the instance is in the cluster and is not running.
3295

3296
    """
3297
    instance = self.cfg.GetInstanceInfo(
3298
      self.cfg.ExpandInstanceName(self.op.instance_name))
3299
    if instance is None:
3300
      raise errors.OpPrereqError("Instance '%s' not known" %
3301
                                 self.op.instance_name)
3302
    _CheckNodeOnline(self, instance.primary_node)
3303

    
3304
    if instance.admin_up:
3305
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3306
                                 self.op.instance_name)
3307
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3308
                                              instance.name,
3309
                                              instance.hypervisor)
3310
    remote_info.Raise()
3311
    if remote_info.data:
3312
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3313
                                 (self.op.instance_name,
3314
                                  instance.primary_node))
3315
    self.instance = instance
3316

    
3317
    # new name verification
3318
    name_info = utils.HostInfo(self.op.new_name)
3319

    
3320
    self.op.new_name = new_name = name_info.name
3321
    instance_list = self.cfg.GetInstanceList()
3322
    if new_name in instance_list:
3323
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3324
                                 new_name)
3325

    
3326
    if not getattr(self.op, "ignore_ip", False):
3327
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3328
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3329
                                   (name_info.ip, new_name))
3330

    
3331

    
3332
  def Exec(self, feedback_fn):
3333
    """Reinstall the instance.
3334

3335
    """
3336
    inst = self.instance
3337
    old_name = inst.name
3338

    
3339
    if inst.disk_template == constants.DT_FILE:
3340
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3341

    
3342
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3343
    # Change the instance lock. This is definitely safe while we hold the BGL
3344
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3345
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3346

    
3347
    # re-read the instance from the configuration after rename
3348
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3349

    
3350
    if inst.disk_template == constants.DT_FILE:
3351
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3352
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3353
                                                     old_file_storage_dir,
3354
                                                     new_file_storage_dir)
3355
      result.Raise()
3356
      if not result.data:
3357
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3358
                                 " directory '%s' to '%s' (but the instance"
3359
                                 " has been renamed in Ganeti)" % (
3360
                                 inst.primary_node, old_file_storage_dir,
3361
                                 new_file_storage_dir))
3362

    
3363
      if not result.data[0]:
3364
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3365
                                 " (but the instance has been renamed in"
3366
                                 " Ganeti)" % (old_file_storage_dir,
3367
                                               new_file_storage_dir))
3368

    
3369
    _StartInstanceDisks(self, inst, None)
3370
    try:
3371
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3372
                                                 old_name)
3373
      msg = result.RemoteFailMsg()
3374
      if msg:
3375
        msg = ("Could not run OS rename script for instance %s on node %s"
3376
               " (but the instance has been renamed in Ganeti): %s" %
3377
               (inst.name, inst.primary_node, msg))
3378
        self.proc.LogWarning(msg)
3379
    finally:
3380
      _ShutdownInstanceDisks(self, inst)
3381

    
3382

    
3383
class LURemoveInstance(LogicalUnit):
3384
  """Remove an instance.
3385

3386
  """
3387
  HPATH = "instance-remove"
3388
  HTYPE = constants.HTYPE_INSTANCE
3389
  _OP_REQP = ["instance_name", "ignore_failures"]
3390
  REQ_BGL = False
3391

    
3392
  def ExpandNames(self):
3393
    self._ExpandAndLockInstance()
3394
    self.needed_locks[locking.LEVEL_NODE] = []
3395
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3396

    
3397
  def DeclareLocks(self, level):
3398
    if level == locking.LEVEL_NODE:
3399
      self._LockInstancesNodes()
3400

    
3401
  def BuildHooksEnv(self):
3402
    """Build hooks env.
3403

3404
    This runs on master, primary and secondary nodes of the instance.
3405

3406
    """
3407
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3408
    nl = [self.cfg.GetMasterNode()]
3409
    return env, nl, nl
3410

    
3411
  def CheckPrereq(self):
3412
    """Check prerequisites.
3413

3414
    This checks that the instance is in the cluster.
3415

3416
    """
3417
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3418
    assert self.instance is not None, \
3419
      "Cannot retrieve locked instance %s" % self.op.instance_name
3420

    
3421
  def Exec(self, feedback_fn):
3422
    """Remove the instance.
3423

3424
    """
3425
    instance = self.instance
3426
    logging.info("Shutting down instance %s on node %s",
3427
                 instance.name, instance.primary_node)
3428

    
3429
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3430
    msg = result.RemoteFailMsg()
3431
    if msg:
3432
      if self.op.ignore_failures:
3433
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3434
      else:
3435
        raise errors.OpExecError("Could not shutdown instance %s on"
3436
                                 " node %s: %s" %
3437
                                 (instance.name, instance.primary_node, msg))
3438

    
3439
    logging.info("Removing block devices for instance %s", instance.name)
3440

    
3441
    if not _RemoveDisks(self, instance):
3442
      if self.op.ignore_failures:
3443
        feedback_fn("Warning: can't remove instance's disks")
3444
      else:
3445
        raise errors.OpExecError("Can't remove instance's disks")
3446

    
3447
    logging.info("Removing instance %s out of cluster config", instance.name)
3448

    
3449
    self.cfg.RemoveInstance(instance.name)
3450
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3451

    
3452

    
3453
class LUQueryInstances(NoHooksLU):
3454
  """Logical unit for querying instances.
3455

3456
  """
3457
  _OP_REQP = ["output_fields", "names", "use_locking"]
3458
  REQ_BGL = False
3459
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3460
                                    "admin_state",
3461
                                    "disk_template", "ip", "mac", "bridge",
3462
                                    "sda_size", "sdb_size", "vcpus", "tags",
3463
                                    "network_port", "beparams",
3464
                                    r"(disk)\.(size)/([0-9]+)",
3465
                                    r"(disk)\.(sizes)", "disk_usage",
3466
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3467
                                    r"(nic)\.(macs|ips|bridges)",
3468
                                    r"(disk|nic)\.(count)",
3469
                                    "serial_no", "hypervisor", "hvparams",] +
3470
                                  ["hv/%s" % name
3471
                                   for name in constants.HVS_PARAMETERS] +
3472
                                  ["be/%s" % name
3473
                                   for name in constants.BES_PARAMETERS])
3474
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3475

    
3476

    
3477
  def ExpandNames(self):
3478
    _CheckOutputFields(static=self._FIELDS_STATIC,
3479
                       dynamic=self._FIELDS_DYNAMIC,
3480
                       selected=self.op.output_fields)
3481

    
3482
    self.needed_locks = {}
3483
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3484
    self.share_locks[locking.LEVEL_NODE] = 1
3485

    
3486
    if self.op.names:
3487
      self.wanted = _GetWantedInstances(self, self.op.names)
3488
    else:
3489
      self.wanted = locking.ALL_SET
3490

    
3491
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3492
    self.do_locking = self.do_node_query and self.op.use_locking
3493
    if self.do_locking:
3494
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3495
      self.needed_locks[locking.LEVEL_NODE] = []
3496
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3497

    
3498
  def DeclareLocks(self, level):
3499
    if level == locking.LEVEL_NODE and self.do_locking:
3500
      self._LockInstancesNodes()
3501

    
3502
  def CheckPrereq(self):
3503
    """Check prerequisites.
3504

3505
    """
3506
    pass
3507

    
3508
  def Exec(self, feedback_fn):
3509
    """Computes the list of nodes and their attributes.
3510

3511
    """
3512
    all_info = self.cfg.GetAllInstancesInfo()
3513
    if self.wanted == locking.ALL_SET:
3514
      # caller didn't specify instance names, so ordering is not important
3515
      if self.do_locking:
3516
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3517
      else:
3518
        instance_names = all_info.keys()
3519
      instance_names = utils.NiceSort(instance_names)
3520
    else:
3521
      # caller did specify names, so we must keep the ordering
3522
      if self.do_locking:
3523
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3524
      else:
3525
        tgt_set = all_info.keys()
3526
      missing = set(self.wanted).difference(tgt_set)
3527
      if missing:
3528
        raise errors.OpExecError("Some instances were removed before"
3529
                                 " retrieving their data: %s" % missing)
3530
      instance_names = self.wanted
3531

    
3532
    instance_list = [all_info[iname] for iname in instance_names]
3533

    
3534
    # begin data gathering
3535

    
3536
    nodes = frozenset([inst.primary_node for inst in instance_list])
3537
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3538

    
3539
    bad_nodes = []
3540
    off_nodes = []
3541
    if self.do_node_query:
3542
      live_data = {}
3543
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3544
      for name in nodes:
3545
        result = node_data[name]
3546
        if result.offline:
3547
          # offline nodes will be in both lists
3548
          off_nodes.append(name)
3549
        if result.failed:
3550
          bad_nodes.append(name)
3551
        else:
3552
          if result.data:
3553
            live_data.update(result.data)
3554
            # else no instance is alive
3555
    else:
3556
      live_data = dict([(name, {}) for name in instance_names])
3557

    
3558
    # end data gathering
3559

    
3560
    HVPREFIX = "hv/"
3561
    BEPREFIX = "be/"
3562
    output = []
3563
    for instance in instance_list:
3564
      iout = []
3565
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3566
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3567
      for field in self.op.output_fields:
3568
        st_match = self._FIELDS_STATIC.Matches(field)
3569
        if field == "name":
3570
          val = instance.name
3571
        elif field == "os":
3572
          val = instance.os
3573
        elif field == "pnode":
3574
          val = instance.primary_node
3575
        elif field == "snodes":
3576
          val = list(instance.secondary_nodes)
3577
        elif field == "admin_state":
3578
          val = instance.admin_up
3579
        elif field == "oper_state":
3580
          if instance.primary_node in bad_nodes:
3581
            val = None
3582
          else:
3583
            val = bool(live_data.get(instance.name))
3584
        elif field == "status":
3585
          if instance.primary_node in off_nodes:
3586
            val = "ERROR_nodeoffline"
3587
          elif instance.primary_node in bad_nodes:
3588
            val = "ERROR_nodedown"
3589
          else:
3590
            running = bool(live_data.get(instance.name))
3591
            if running:
3592
              if instance.admin_up:
3593
                val = "running"
3594
              else:
3595
                val = "ERROR_up"
3596
            else:
3597
              if instance.admin_up:
3598
                val = "ERROR_down"
3599
              else:
3600
                val = "ADMIN_down"
3601
        elif field == "oper_ram":
3602
          if instance.primary_node in bad_nodes:
3603
            val = None
3604
          elif instance.name in live_data:
3605
            val = live_data[instance.name].get("memory", "?")
3606
          else:
3607
            val = "-"
3608
        elif field == "vcpus":
3609
          val = i_be[constants.BE_VCPUS]
3610
        elif field == "disk_template":
3611
          val = instance.disk_template
3612
        elif field == "ip":
3613
          if instance.nics:
3614
            val = instance.nics[0].ip
3615
          else:
3616
            val = None
3617
        elif field == "bridge":
3618
          if instance.nics:
3619
            val = instance.nics[0].bridge
3620
          else:
3621
            val = None
3622
        elif field == "mac":
3623
          if instance.nics:
3624
            val = instance.nics[0].mac
3625
          else:
3626
            val = None
3627
        elif field == "sda_size" or field == "sdb_size":
3628
          idx = ord(field[2]) - ord('a')
3629
          try:
3630
            val = instance.FindDisk(idx).size
3631
          except errors.OpPrereqError:
3632
            val = None
3633
        elif field == "disk_usage": # total disk usage per node
3634
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3635
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3636
        elif field == "tags":
3637
          val = list(instance.GetTags())
3638
        elif field == "serial_no":
3639
          val = instance.serial_no
3640
        elif field == "network_port":
3641
          val = instance.network_port
3642
        elif field == "hypervisor":
3643
          val = instance.hypervisor
3644
        elif field == "hvparams":
3645
          val = i_hv
3646
        elif (field.startswith(HVPREFIX) and
3647
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3648
          val = i_hv.get(field[len(HVPREFIX):], None)
3649
        elif field == "beparams":
3650
          val = i_be
3651
        elif (field.startswith(BEPREFIX) and
3652
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3653
          val = i_be.get(field[len(BEPREFIX):], None)
3654
        elif st_match and st_match.groups():
3655
          # matches a variable list
3656
          st_groups = st_match.groups()
3657
          if st_groups and st_groups[0] == "disk":
3658
            if st_groups[1] == "count":
3659
              val = len(instance.disks)
3660
            elif st_groups[1] == "sizes":
3661
              val = [disk.size for disk in instance.disks]
3662
            elif st_groups[1] == "size":
3663
              try:
3664
                val = instance.FindDisk(st_groups[2]).size
3665
              except errors.OpPrereqError:
3666
                val = None
3667
            else:
3668
              assert False, "Unhandled disk parameter"
3669
          elif st_groups[0] == "nic":
3670
            if st_groups[1] == "count":
3671
              val = len(instance.nics)
3672
            elif st_groups[1] == "macs":
3673
              val = [nic.mac for nic in instance.nics]
3674
            elif st_groups[1] == "ips":
3675
              val = [nic.ip for nic in instance.nics]
3676
            elif st_groups[1] == "bridges":
3677
              val = [nic.bridge for nic in instance.nics]
3678
            else:
3679
              # index-based item
3680
              nic_idx = int(st_groups[2])
3681
              if nic_idx >= len(instance.nics):
3682
                val = None
3683
              else:
3684
                if st_groups[1] == "mac":
3685
                  val = instance.nics[nic_idx].mac
3686
                elif st_groups[1] == "ip":
3687
                  val = instance.nics[nic_idx].ip
3688
                elif st_groups[1] == "bridge":
3689
                  val = instance.nics[nic_idx].bridge
3690
                else:
3691
                  assert False, "Unhandled NIC parameter"
3692
          else:
3693
            assert False, ("Declared but unhandled variable parameter '%s'" %
3694
                           field)
3695
        else:
3696
          assert False, "Declared but unhandled parameter '%s'" % field
3697
        iout.append(val)
3698
      output.append(iout)
3699

    
3700
    return output
3701

    
3702

    
3703
class LUFailoverInstance(LogicalUnit):
3704
  """Failover an instance.
3705

3706
  """
3707
  HPATH = "instance-failover"
3708
  HTYPE = constants.HTYPE_INSTANCE
3709
  _OP_REQP = ["instance_name", "ignore_consistency"]
3710
  REQ_BGL = False
3711

    
3712
  def ExpandNames(self):
3713
    self._ExpandAndLockInstance()
3714
    self.needed_locks[locking.LEVEL_NODE] = []
3715
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3716

    
3717
  def DeclareLocks(self, level):
3718
    if level == locking.LEVEL_NODE:
3719
      self._LockInstancesNodes()
3720

    
3721
  def BuildHooksEnv(self):
3722
    """Build hooks env.
3723

3724
    This runs on master, primary and secondary nodes of the instance.
3725

3726
    """
3727
    env = {
3728
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3729
      }
3730
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3731
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3732
    return env, nl, nl
3733

    
3734
  def CheckPrereq(self):
3735
    """Check prerequisites.
3736

3737
    This checks that the instance is in the cluster.
3738

3739
    """
3740
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3741
    assert self.instance is not None, \
3742
      "Cannot retrieve locked instance %s" % self.op.instance_name
3743

    
3744
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3745
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3746
      raise errors.OpPrereqError("Instance's disk layout is not"
3747
                                 " network mirrored, cannot failover.")
3748

    
3749
    secondary_nodes = instance.secondary_nodes
3750
    if not secondary_nodes:
3751
      raise errors.ProgrammerError("no secondary node but using "
3752
                                   "a mirrored disk template")
3753

    
3754
    target_node = secondary_nodes[0]
3755
    _CheckNodeOnline(self, target_node)
3756
    _CheckNodeNotDrained(self, target_node)
3757

    
3758
    if instance.admin_up:
3759
      # check memory requirements on the secondary node
3760
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3761
                           instance.name, bep[constants.BE_MEMORY],
3762
                           instance.hypervisor)
3763
    else:
3764
      self.LogInfo("Not checking memory on the secondary node as"
3765
                   " instance will not be started")
3766

    
3767
    # check bridge existence
3768
    brlist = [nic.bridge for nic in instance.nics]
3769
    result = self.rpc.call_bridges_exist(target_node, brlist)
3770
    result.Raise()
3771
    if not result.data:
3772
      raise errors.OpPrereqError("One or more target bridges %s does not"
3773
                                 " exist on destination node '%s'" %
3774
                                 (brlist, target_node))
3775

    
3776
  def Exec(self, feedback_fn):
3777
    """Failover an instance.
3778

3779
    The failover is done by shutting it down on its present node and
3780
    starting it on the secondary.
3781

3782
    """
3783
    instance = self.instance
3784

    
3785
    source_node = instance.primary_node
3786
    target_node = instance.secondary_nodes[0]
3787

    
3788
    feedback_fn("* checking disk consistency between source and target")
3789
    for dev in instance.disks:
3790
      # for drbd, these are drbd over lvm
3791
      if not _CheckDiskConsistency(self, dev, target_node, False):
3792
        if instance.admin_up and not self.op.ignore_consistency:
3793
          raise errors.OpExecError("Disk %s is degraded on target node,"
3794
                                   " aborting failover." % dev.iv_name)
3795

    
3796
    feedback_fn("* shutting down instance on source node")
3797
    logging.info("Shutting down instance %s on node %s",
3798
                 instance.name, source_node)
3799

    
3800
    result = self.rpc.call_instance_shutdown(source_node, instance)
3801
    msg = result.RemoteFailMsg()
3802
    if msg:
3803
      if self.op.ignore_consistency:
3804
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3805
                             " Proceeding anyway. Please make sure node"
3806
                             " %s is down. Error details: %s",
3807
                             instance.name, source_node, source_node, msg)
3808
      else:
3809
        raise errors.OpExecError("Could not shutdown instance %s on"
3810
                                 " node %s: %s" %
3811
                                 (instance.name, source_node, msg))
3812

    
3813
    feedback_fn("* deactivating the instance's disks on source node")
3814
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3815
      raise errors.OpExecError("Can't shut down the instance's disks.")
3816

    
3817
    instance.primary_node = target_node
3818
    # distribute new instance config to the other nodes
3819
    self.cfg.Update(instance)
3820

    
3821
    # Only start the instance if it's marked as up
3822
    if instance.admin_up:
3823
      feedback_fn("* activating the instance's disks on target node")
3824
      logging.info("Starting instance %s on node %s",
3825
                   instance.name, target_node)
3826

    
3827
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
3828
                                               ignore_secondaries=True)
3829
      if not disks_ok:
3830
        _ShutdownInstanceDisks(self, instance)
3831
        raise errors.OpExecError("Can't activate the instance's disks")
3832

    
3833
      feedback_fn("* starting the instance on the target node")
3834
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3835
      msg = result.RemoteFailMsg()
3836
      if msg:
3837
        _ShutdownInstanceDisks(self, instance)
3838
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3839
                                 (instance.name, target_node, msg))
3840

    
3841

    
3842
class LUMigrateInstance(LogicalUnit):
3843
  """Migrate an instance.
3844

3845
  This is migration without shutting down, compared to the failover,
3846
  which is done with shutdown.
3847

3848
  """
3849
  HPATH = "instance-migrate"
3850
  HTYPE = constants.HTYPE_INSTANCE
3851
  _OP_REQP = ["instance_name", "live", "cleanup"]
3852

    
3853
  REQ_BGL = False
3854

    
3855
  def ExpandNames(self):
3856
    self._ExpandAndLockInstance()
3857
    self.needed_locks[locking.LEVEL_NODE] = []
3858
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3859

    
3860
  def DeclareLocks(self, level):
3861
    if level == locking.LEVEL_NODE:
3862
      self._LockInstancesNodes()
3863

    
3864
  def BuildHooksEnv(self):
3865
    """Build hooks env.
3866

3867
    This runs on master, primary and secondary nodes of the instance.
3868

3869
    """
3870
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3871
    env["MIGRATE_LIVE"] = self.op.live
3872
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3873
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3874
    return env, nl, nl
3875

    
3876
  def CheckPrereq(self):
3877
    """Check prerequisites.
3878

3879
    This checks that the instance is in the cluster.
3880

3881
    """
3882
    instance = self.cfg.GetInstanceInfo(
3883
      self.cfg.ExpandInstanceName(self.op.instance_name))
3884
    if instance is None:
3885
      raise errors.OpPrereqError("Instance '%s' not known" %
3886
                                 self.op.instance_name)
3887

    
3888
    if instance.disk_template != constants.DT_DRBD8:
3889
      raise errors.OpPrereqError("Instance's disk layout is not"
3890
                                 " drbd8, cannot migrate.")
3891

    
3892
    secondary_nodes = instance.secondary_nodes
3893
    if not secondary_nodes:
3894
      raise errors.ConfigurationError("No secondary node but using"
3895
                                      " drbd8 disk template")
3896

    
3897
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3898

    
3899
    target_node = secondary_nodes[0]
3900
    # check memory requirements on the secondary node
3901
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3902
                         instance.name, i_be[constants.BE_MEMORY],
3903
                         instance.hypervisor)
3904

    
3905
    # check bridge existence
3906
    brlist = [nic.bridge for nic in instance.nics]
3907
    result = self.rpc.call_bridges_exist(target_node, brlist)
3908
    if result.failed or not result.data:
3909
      raise errors.OpPrereqError("One or more target bridges %s does not"
3910
                                 " exist on destination node '%s'" %
3911
                                 (brlist, target_node))
3912

    
3913
    if not self.op.cleanup:
3914
      _CheckNodeNotDrained(self, target_node)
3915
      result = self.rpc.call_instance_migratable(instance.primary_node,
3916
                                                 instance)
3917
      msg = result.RemoteFailMsg()
3918
      if msg:
3919
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3920
                                   msg)
3921

    
3922
    self.instance = instance
3923

    
3924
  def _WaitUntilSync(self):
3925
    """Poll with custom rpc for disk sync.
3926

3927
    This uses our own step-based rpc call.
3928

3929
    """
3930
    self.feedback_fn("* wait until resync is done")
3931
    all_done = False
3932
    while not all_done:
3933
      all_done = True
3934
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3935
                                            self.nodes_ip,
3936
                                            self.instance.disks)
3937
      min_percent = 100
3938
      for node, nres in result.items():
3939
        msg = nres.RemoteFailMsg()
3940
        if msg:
3941
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3942
                                   (node, msg))
3943
        node_done, node_percent = nres.payload
3944
        all_done = all_done and node_done
3945
        if node_percent is not None:
3946
          min_percent = min(min_percent, node_percent)
3947
      if not all_done:
3948
        if min_percent < 100:
3949
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3950
        time.sleep(2)
3951

    
3952
  def _EnsureSecondary(self, node):
3953
    """Demote a node to secondary.
3954

3955
    """
3956
    self.feedback_fn("* switching node %s to secondary mode" % node)
3957

    
3958
    for dev in self.instance.disks:
3959
      self.cfg.SetDiskID(dev, node)
3960

    
3961
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3962
                                          self.instance.disks)
3963
    msg = result.RemoteFailMsg()
3964
    if msg:
3965
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3966
                               " error %s" % (node, msg))
3967

    
3968
  def _GoStandalone(self):
3969
    """Disconnect from the network.
3970

3971
    """
3972
    self.feedback_fn("* changing into standalone mode")
3973
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3974
                                               self.instance.disks)
3975
    for node, nres in result.items():
3976
      msg = nres.RemoteFailMsg()
3977
      if msg:
3978
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3979
                                 " error %s" % (node, msg))
3980

    
3981
  def _GoReconnect(self, multimaster):
3982
    """Reconnect to the network.
3983

3984
    """
3985
    if multimaster:
3986
      msg = "dual-master"
3987
    else:
3988
      msg = "single-master"
3989
    self.feedback_fn("* changing disks into %s mode" % msg)
3990
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3991
                                           self.instance.disks,
3992
                                           self.instance.name, multimaster)
3993
    for node, nres in result.items():
3994
      msg = nres.RemoteFailMsg()
3995
      if msg:
3996
        raise errors.OpExecError("Cannot change disks config on node %s,"
3997
                                 " error: %s" % (node, msg))
3998

    
3999
  def _ExecCleanup(self):
4000
    """Try to cleanup after a failed migration.
4001

4002
    The cleanup is done by:
4003
      - check that the instance is running only on one node
4004
        (and update the config if needed)
4005
      - change disks on its secondary node to secondary
4006
      - wait until disks are fully synchronized
4007
      - disconnect from the network
4008
      - change disks into single-master mode
4009
      - wait again until disks are fully synchronized
4010

4011
    """
4012
    instance = self.instance
4013
    target_node = self.target_node
4014
    source_node = self.source_node
4015

    
4016
    # check running on only one node
4017
    self.feedback_fn("* checking where the instance actually runs"
4018
                     " (if this hangs, the hypervisor might be in"
4019
                     " a bad state)")
4020
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
4021
    for node, result in ins_l.items():
4022
      result.Raise()
4023
      if not isinstance(result.data, list):
4024
        raise errors.OpExecError("Can't contact node '%s'" % node)
4025

    
4026
    runningon_source = instance.name in ins_l[source_node].data
4027
    runningon_target = instance.name in ins_l[target_node].data
4028

    
4029
    if runningon_source and runningon_target:
4030
      raise errors.OpExecError("Instance seems to be running on two nodes,"
4031
                               " or the hypervisor is confused. You will have"
4032
                               " to ensure manually that it runs only on one"
4033
                               " and restart this operation.")
4034

    
4035
    if not (runningon_source or runningon_target):
4036
      raise errors.OpExecError("Instance does not seem to be running at all."
4037
                               " In this case, it's safer to repair by"
4038
                               " running 'gnt-instance stop' to ensure disk"
4039
                               " shutdown, and then restarting it.")
4040

    
4041
    if runningon_target:
4042
      # the migration has actually succeeded, we need to update the config
4043
      self.feedback_fn("* instance running on secondary node (%s),"
4044
                       " updating config" % target_node)
4045
      instance.primary_node = target_node
4046
      self.cfg.Update(instance)
4047
      demoted_node = source_node
4048
    else:
4049
      self.feedback_fn("* instance confirmed to be running on its"
4050
                       " primary node (%s)" % source_node)
4051
      demoted_node = target_node
4052

    
4053
    self._EnsureSecondary(demoted_node)
4054
    try:
4055
      self._WaitUntilSync()
4056
    except errors.OpExecError:
4057
      # we ignore here errors, since if the device is standalone, it
4058
      # won't be able to sync
4059
      pass
4060
    self._GoStandalone()
4061
    self._GoReconnect(False)
4062
    self._WaitUntilSync()
4063

    
4064
    self.feedback_fn("* done")
4065

    
4066
  def _RevertDiskStatus(self):
4067
    """Try to revert the disk status after a failed migration.
4068

4069
    """
4070
    target_node = self.target_node
4071
    try:
4072
      self._EnsureSecondary(target_node)
4073
      self._GoStandalone()
4074
      self._GoReconnect(False)
4075
      self._WaitUntilSync()
4076
    except errors.OpExecError, err:
4077
      self.LogWarning("Migration failed and I can't reconnect the"
4078
                      " drives: error '%s'\n"
4079
                      "Please look and recover the instance status" %
4080
                      str(err))
4081

    
4082
  def _AbortMigration(self):
4083
    """Call the hypervisor code to abort a started migration.
4084

4085
    """
4086
    instance = self.instance
4087
    target_node = self.target_node
4088
    migration_info = self.migration_info
4089

    
4090
    abort_result = self.rpc.call_finalize_migration(target_node,
4091
                                                    instance,
4092
                                                    migration_info,
4093
                                                    False)
4094
    abort_msg = abort_result.RemoteFailMsg()
4095
    if abort_msg:
4096
      logging.error("Aborting migration failed on target node %s: %s" %
4097
                    (target_node, abort_msg))
4098
      # Don't raise an exception here, as we stil have to try to revert the
4099
      # disk status, even if this step failed.
4100

    
4101
  def _ExecMigration(self):
4102
    """Migrate an instance.
4103

4104
    The migrate is done by:
4105
      - change the disks into dual-master mode
4106
      - wait until disks are fully synchronized again
4107
      - migrate the instance
4108
      - change disks on the new secondary node (the old primary) to secondary
4109
      - wait until disks are fully synchronized
4110
      - change disks into single-master mode
4111

4112
    """
4113
    instance = self.instance
4114
    target_node = self.target_node
4115
    source_node = self.source_node
4116

    
4117
    self.feedback_fn("* checking disk consistency between source and target")
4118
    for dev in instance.disks:
4119
      if not _CheckDiskConsistency(self, dev, target_node, False):
4120
        raise errors.OpExecError("Disk %s is degraded or not fully"
4121
                                 " synchronized on target node,"
4122
                                 " aborting migrate." % dev.iv_name)
4123

    
4124
    # First get the migration information from the remote node
4125
    result = self.rpc.call_migration_info(source_node, instance)
4126
    msg = result.RemoteFailMsg()
4127
    if msg:
4128
      log_err = ("Failed fetching source migration information from %s: %s" %
4129
                 (source_node, msg))
4130
      logging.error(log_err)
4131
      raise errors.OpExecError(log_err)
4132

    
4133
    self.migration_info = migration_info = result.payload
4134

    
4135
    # Then switch the disks to master/master mode
4136
    self._EnsureSecondary(target_node)
4137
    self._GoStandalone()
4138
    self._GoReconnect(True)
4139
    self._WaitUntilSync()
4140

    
4141
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
4142
    result = self.rpc.call_accept_instance(target_node,
4143
                                           instance,
4144
                                           migration_info,
4145
                                           self.nodes_ip[target_node])
4146

    
4147
    msg = result.RemoteFailMsg()
4148
    if msg:
4149
      logging.error("Instance pre-migration failed, trying to revert"
4150
                    " disk status: %s", msg)
4151
      self._AbortMigration()
4152
      self._RevertDiskStatus()
4153
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
4154
                               (instance.name, msg))
4155

    
4156
    self.feedback_fn("* migrating instance to %s" % target_node)
4157
    time.sleep(10)
4158
    result = self.rpc.call_instance_migrate(source_node, instance,
4159
                                            self.nodes_ip[target_node],
4160
                                            self.op.live)
4161
    msg = result.RemoteFailMsg()
4162
    if msg:
4163
      logging.error("Instance migration failed, trying to revert"
4164
                    " disk status: %s", msg)
4165
      self._AbortMigration()
4166
      self._RevertDiskStatus()
4167
      raise errors.OpExecError("Could not migrate instance %s: %s" %
4168
                               (instance.name, msg))
4169
    time.sleep(10)
4170

    
4171
    instance.primary_node = target_node
4172
    # distribute new instance config to the other nodes
4173
    self.cfg.Update(instance)
4174

    
4175
    result = self.rpc.call_finalize_migration(target_node,
4176
                                              instance,
4177
                                              migration_info,
4178
                                              True)
4179
    msg = result.RemoteFailMsg()
4180
    if msg:
4181
      logging.error("Instance migration succeeded, but finalization failed:"
4182
                    " %s" % msg)
4183
      raise errors.OpExecError("Could not finalize instance migration: %s" %
4184
                               msg)
4185

    
4186
    self._EnsureSecondary(source_node)
4187
    self._WaitUntilSync()
4188
    self._GoStandalone()
4189
    self._GoReconnect(False)
4190
    self._WaitUntilSync()
4191

    
4192
    self.feedback_fn("* done")
4193

    
4194
  def Exec(self, feedback_fn):
4195
    """Perform the migration.
4196

4197
    """
4198
    self.feedback_fn = feedback_fn
4199

    
4200
    self.source_node = self.instance.primary_node
4201
    self.target_node = self.instance.secondary_nodes[0]
4202
    self.all_nodes = [self.source_node, self.target_node]
4203
    self.nodes_ip = {
4204
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
4205
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
4206
      }
4207
    if self.op.cleanup:
4208
      return self._ExecCleanup()
4209
    else:
4210
      return self._ExecMigration()
4211

    
4212

    
4213
def _CreateBlockDev(lu, node, instance, device, force_create,
4214
                    info, force_open):
4215
  """Create a tree of block devices on a given node.
4216

4217
  If this device type has to be created on secondaries, create it and
4218
  all its children.
4219

4220
  If not, just recurse to children keeping the same 'force' value.
4221

4222
  @param lu: the lu on whose behalf we execute
4223
  @param node: the node on which to create the device
4224
  @type instance: L{objects.Instance}
4225
  @param instance: the instance which owns the device
4226
  @type device: L{objects.Disk}
4227
  @param device: the device to create
4228
  @type force_create: boolean
4229
  @param force_create: whether to force creation of this device; this
4230
      will be change to True whenever we find a device which has
4231
      CreateOnSecondary() attribute
4232
  @param info: the extra 'metadata' we should attach to the device
4233
      (this will be represented as a LVM tag)
4234
  @type force_open: boolean
4235
  @param force_open: this parameter will be passes to the
4236
      L{backend.BlockdevCreate} function where it specifies
4237
      whether we run on primary or not, and it affects both
4238
      the child assembly and the device own Open() execution
4239

4240
  """
4241
  if device.CreateOnSecondary():
4242
    force_create = True
4243

    
4244
  if device.children:
4245
    for child in device.children:
4246
      _CreateBlockDev(lu, node, instance, child, force_create,
4247
                      info, force_open)
4248

    
4249
  if not force_create:
4250
    return
4251

    
4252
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
4253

    
4254

    
4255
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
4256
  """Create a single block device on a given node.
4257

4258
  This will not recurse over children of the device, so they must be
4259
  created in advance.
4260

4261
  @param lu: the lu on whose behalf we execute
4262
  @param node: the node on which to create the device
4263
  @type instance: L{objects.Instance}
4264
  @param instance: the instance which owns the device
4265
  @type device: L{objects.Disk}
4266
  @param device: the device to create
4267
  @param info: the extra 'metadata' we should attach to the device
4268
      (this will be represented as a LVM tag)
4269
  @type force_open: boolean
4270
  @param force_open: this parameter will be passes to the
4271
      L{backend.BlockdevCreate} function where it specifies
4272
      whether we run on primary or not, and it affects both
4273
      the child assembly and the device own Open() execution
4274

4275
  """
4276
  lu.cfg.SetDiskID(device, node)
4277
  result = lu.rpc.call_blockdev_create(node, device, device.size,
4278
                                       instance.name, force_open, info)
4279
  msg = result.RemoteFailMsg()
4280
  if msg:
4281
    raise errors.OpExecError("Can't create block device %s on"
4282
                             " node %s for instance %s: %s" %
4283
                             (device, node, instance.name, msg))
4284
  if device.physical_id is None:
4285
    device.physical_id = result.payload
4286

    
4287

    
4288
def _GenerateUniqueNames(lu, exts):
4289
  """Generate a suitable LV name.
4290

4291
  This will generate a logical volume name for the given instance.
4292

4293
  """
4294
  results = []
4295
  for val in exts:
4296
    new_id = lu.cfg.GenerateUniqueID()
4297
    results.append("%s%s" % (new_id, val))
4298
  return results
4299

    
4300

    
4301
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
4302
                         p_minor, s_minor):
4303
  """Generate a drbd8 device complete with its children.
4304

4305
  """
4306
  port = lu.cfg.AllocatePort()
4307
  vgname = lu.cfg.GetVGName()
4308
  shared_secret = lu.cfg.GenerateDRBDSecret()
4309
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
4310
                          logical_id=(vgname, names[0]))
4311
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
4312
                          logical_id=(vgname, names[1]))
4313
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
4314
                          logical_id=(primary, secondary, port,
4315
                                      p_minor, s_minor,
4316
                                      shared_secret),
4317
                          children=[dev_data, dev_meta],
4318
                          iv_name=iv_name)
4319
  return drbd_dev
4320

    
4321

    
4322
def _GenerateDiskTemplate(lu, template_name,
4323
                          instance_name, primary_node,
4324
                          secondary_nodes, disk_info,
4325
                          file_storage_dir, file_driver,
4326
                          base_index):
4327
  """Generate the entire disk layout for a given template type.
4328

4329
  """
4330
  #TODO: compute space requirements
4331

    
4332
  vgname = lu.cfg.GetVGName()
4333
  disk_count = len(disk_info)
4334
  disks = []
4335
  if template_name == constants.DT_DISKLESS:
4336
    pass
4337
  elif template_name == constants.DT_PLAIN:
4338
    if len(secondary_nodes) != 0:
4339
      raise errors.ProgrammerError("Wrong template configuration")
4340

    
4341
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
4342
                                      for i in range(disk_count)])
4343
    for idx, disk in enumerate(disk_info):
4344
      disk_index = idx + base_index
4345
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
4346
                              logical_id=(vgname, names[idx]),
4347
                              iv_name="disk/%d" % disk_index,
4348
                              mode=disk["mode"])
4349
      disks.append(disk_dev)
4350
  elif template_name == constants.DT_DRBD8:
4351
    if len(secondary_nodes) != 1:
4352
      raise errors.ProgrammerError("Wrong template configuration")
4353
    remote_node = secondary_nodes[0]
4354
    minors = lu.cfg.AllocateDRBDMinor(
4355
      [primary_node, remote_node] * len(disk_info), instance_name)
4356

    
4357
    names = []
4358
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
4359
                                               for i in range(disk_count)]):
4360
      names.append(lv_prefix + "_data")
4361
      names.append(lv_prefix + "_meta")
4362
    for idx, disk in enumerate(disk_info):
4363
      disk_index = idx + base_index
4364
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
4365
                                      disk["size"], names[idx*2:idx*2+2],
4366
                                      "disk/%d" % disk_index,
4367
                                      minors[idx*2], minors[idx*2+1])
4368
      disk_dev.mode = disk["mode"]
4369
      disks.append(disk_dev)
4370
  elif template_name == constants.DT_FILE:
4371
    if len(secondary_nodes) != 0:
4372
      raise errors.ProgrammerError("Wrong template configuration")
4373

    
4374
    for idx, disk in enumerate(disk_info):
4375
      disk_index = idx + base_index
4376
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
4377
                              iv_name="disk/%d" % disk_index,
4378
                              logical_id=(file_driver,
4379
                                          "%s/disk%d" % (file_storage_dir,
4380
                                                         disk_index)),
4381
                              mode=disk["mode"])
4382
      disks.append(disk_dev)
4383
  else:
4384
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
4385
  return disks
4386

    
4387

    
4388
def _GetInstanceInfoText(instance):
4389
  """Compute that text that should be added to the disk's metadata.
4390

4391
  """
4392
  return "originstname+%s" % instance.name
4393

    
4394

    
4395
def _CreateDisks(lu, instance):
4396
  """Create all disks for an instance.
4397

4398
  This abstracts away some work from AddInstance.
4399

4400
  @type lu: L{LogicalUnit}
4401
  @param lu: the logical unit on whose behalf we execute
4402
  @type instance: L{objects.Instance}
4403
  @param instance: the instance whose disks we should create
4404
  @rtype: boolean
4405
  @return: the success of the creation
4406

4407
  """
4408
  info = _GetInstanceInfoText(instance)
4409
  pnode = instance.primary_node
4410

    
4411
  if instance.disk_template == constants.DT_FILE:
4412
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4413
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
4414

    
4415
    if result.failed or not result.data:
4416
      raise errors.OpExecError("Could not connect to node '%s'" % pnode)
4417

    
4418
    if not result.data[0]:
4419
      raise errors.OpExecError("Failed to create directory '%s'" %
4420
                               file_storage_dir)
4421

    
4422
  # Note: this needs to be kept in sync with adding of disks in
4423
  # LUSetInstanceParams
4424
  for device in instance.disks:
4425
    logging.info("Creating volume %s for instance %s",
4426
                 device.iv_name, instance.name)
4427
    #HARDCODE
4428
    for node in instance.all_nodes:
4429
      f_create = node == pnode
4430
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
4431

    
4432

    
4433
def _RemoveDisks(lu, instance):
4434
  """Remove all disks for an instance.
4435

4436
  This abstracts away some work from `AddInstance()` and
4437
  `RemoveInstance()`. Note that in case some of the devices couldn't
4438
  be removed, the removal will continue with the other ones (compare
4439
  with `_CreateDisks()`).
4440

4441
  @type lu: L{LogicalUnit}
4442
  @param lu: the logical unit on whose behalf we execute
4443
  @type instance: L{objects.Instance}
4444
  @param instance: the instance whose disks we should remove
4445
  @rtype: boolean
4446
  @return: the success of the removal
4447

4448
  """
4449
  logging.info("Removing block devices for instance %s", instance.name)
4450

    
4451
  all_result = True
4452
  for device in instance.disks:
4453
    for node, disk in device.ComputeNodeTree(instance.primary_node):
4454
      lu.cfg.SetDiskID(disk, node)
4455
      msg = lu.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
4456
      if msg:
4457
        lu.LogWarning("Could not remove block device %s on node %s,"
4458
                      " continuing anyway: %s", device.iv_name, node, msg)
4459
        all_result = False
4460

    
4461
  if instance.disk_template == constants.DT_FILE:
4462
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4463
    result = lu.rpc.call_file_storage_dir_remove(instance.primary_node,
4464
                                                 file_storage_dir)
4465
    if result.failed or not result.data:
4466
      logging.error("Could not remove directory '%s'", file_storage_dir)
4467
      all_result = False
4468

    
4469
  return all_result
4470

    
4471

    
4472
def _ComputeDiskSize(disk_template, disks):
4473
  """Compute disk size requirements in the volume group
4474

4475
  """
4476
  # Required free disk space as a function of disk and swap space
4477
  req_size_dict = {
4478
    constants.DT_DISKLESS: None,
4479
    constants.DT_PLAIN: sum(d["size"] for d in disks),
4480
    # 128 MB are added for drbd metadata for each disk
4481
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
4482
    constants.DT_FILE: None,
4483
  }
4484

    
4485
  if disk_template not in req_size_dict:
4486
    raise errors.ProgrammerError("Disk template '%s' size requirement"
4487
                                 " is unknown" %  disk_template)
4488

    
4489
  return req_size_dict[disk_template]
4490

    
4491

    
4492
def _CheckHVParams(lu, nodenames, hvname, hvparams):
4493
  """Hypervisor parameter validation.
4494

4495
  This function abstract the hypervisor parameter validation to be
4496
  used in both instance create and instance modify.
4497

4498
  @type lu: L{LogicalUnit}
4499
  @param lu: the logical unit for which we check
4500
  @type nodenames: list
4501
  @param nodenames: the list of nodes on which we should check
4502
  @type hvname: string
4503
  @param hvname: the name of the hypervisor we should use
4504
  @type hvparams: dict
4505
  @param hvparams: the parameters which we need to check
4506
  @raise errors.OpPrereqError: if the parameters are not valid
4507

4508
  """
4509
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
4510
                                                  hvname,
4511
                                                  hvparams)
4512
  for node in nodenames:
4513
    info = hvinfo[node]
4514
    if info.offline:
4515
      continue
4516
    msg = info.RemoteFailMsg()
4517
    if msg:
4518
      raise errors.OpPrereqError("Hypervisor parameter validation"
4519
                                 " failed on node %s: %s" % (node, msg))
4520

    
4521

    
4522
class LUCreateInstance(LogicalUnit):
4523
  """Create an instance.
4524

4525
  """
4526
  HPATH = "instance-add"
4527
  HTYPE = constants.HTYPE_INSTANCE
4528
  _OP_REQP = ["instance_name", "disks", "disk_template",
4529
              "mode", "start",
4530
              "wait_for_sync", "ip_check", "nics",
4531
              "hvparams", "beparams"]
4532
  REQ_BGL = False
4533

    
4534
  def _ExpandNode(self, node):
4535
    """Expands and checks one node name.
4536

4537
    """
4538
    node_full = self.cfg.ExpandNodeName(node)
4539
    if node_full is None:
4540
      raise errors.OpPrereqError("Unknown node %s" % node)
4541
    return node_full
4542

    
4543
  def ExpandNames(self):
4544
    """ExpandNames for CreateInstance.
4545

4546
    Figure out the right locks for instance creation.
4547

4548
    """
4549
    self.needed_locks = {}
4550

    
4551
    # set optional parameters to none if they don't exist
4552
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
4553
      if not hasattr(self.op, attr):
4554
        setattr(self.op, attr, None)
4555

    
4556
    # cheap checks, mostly valid constants given
4557

    
4558
    # verify creation mode
4559
    if self.op.mode not in (constants.INSTANCE_CREATE,
4560
                            constants.INSTANCE_IMPORT):
4561
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
4562
                                 self.op.mode)
4563

    
4564
    # disk template and mirror node verification
4565
    if self.op.disk_template not in constants.DISK_TEMPLATES:
4566
      raise errors.OpPrereqError("Invalid disk template name")
4567

    
4568
    if self.op.hypervisor is None:
4569
      self.op.hypervisor = self.cfg.GetHypervisorType()
4570

    
4571
    cluster = self.cfg.GetClusterInfo()
4572
    enabled_hvs = cluster.enabled_hypervisors
4573
    if self.op.hypervisor not in enabled_hvs:
4574
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
4575
                                 " cluster (%s)" % (self.op.hypervisor,
4576
                                  ",".join(enabled_hvs)))
4577

    
4578
    # check hypervisor parameter syntax (locally)
4579
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
4580
    filled_hvp = cluster.FillDict(cluster.hvparams[self.op.hypervisor],
4581
                                  self.op.hvparams)
4582
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
4583
    hv_type.CheckParameterSyntax(filled_hvp)
4584
    self.hv_full = filled_hvp
4585

    
4586
    # fill and remember the beparams dict
4587
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
4588
    self.be_full = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
4589
                                    self.op.beparams)
4590

    
4591
    #### instance parameters check
4592

    
4593
    # instance name verification
4594
    hostname1 = utils.HostInfo(self.op.instance_name)
4595
    self.op.instance_name = instance_name = hostname1.name
4596

    
4597
    # this is just a preventive check, but someone might still add this
4598
    # instance in the meantime, and creation will fail at lock-add time
4599
    if instance_name in self.cfg.GetInstanceList():
4600
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4601
                                 instance_name)
4602

    
4603
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
4604

    
4605
    # NIC buildup
4606
    self.nics = []
4607
    for nic in self.op.nics:
4608
      # ip validity checks
4609
      ip = nic.get("ip", None)
4610
      if ip is None or ip.lower() == "none":
4611
        nic_ip = None
4612
      elif ip.lower() == constants.VALUE_AUTO:
4613
        nic_ip = hostname1.ip
4614
      else:
4615
        if not utils.IsValidIP(ip):
4616
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
4617
                                     " like a valid IP" % ip)
4618
        nic_ip = ip
4619

    
4620
      # MAC address verification
4621
      mac = nic.get("mac", constants.VALUE_AUTO)
4622
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4623
        if not utils.IsValidMac(mac.lower()):
4624
          raise errors.OpPrereqError("Invalid MAC address specified: %s" %
4625
                                     mac)
4626
        else:
4627
          # or validate/reserve the current one
4628
          if self.cfg.IsMacInUse(mac):
4629
            raise errors.OpPrereqError("MAC address %s already in use"
4630
                                       " in cluster" % mac)
4631

    
4632
      # bridge verification
4633
      bridge = nic.get("bridge", None)
4634
      if bridge is None:
4635
        bridge = self.cfg.GetDefBridge()
4636
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
4637

    
4638
    # disk checks/pre-build
4639
    self.disks = []
4640
    for disk in self.op.disks:
4641
      mode = disk.get("mode", constants.DISK_RDWR)
4642
      if mode not in constants.DISK_ACCESS_SET:
4643
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
4644
                                   mode)
4645
      size = disk.get("size", None)
4646
      if size is None:
4647
        raise errors.OpPrereqError("Missing disk size")
4648
      try:
4649
        size = int(size)
4650
      except ValueError:
4651
        raise errors.OpPrereqError("Invalid disk size '%s'" % size)
4652
      self.disks.append({"size": size, "mode": mode})
4653

    
4654
    # used in CheckPrereq for ip ping check
4655
    self.check_ip = hostname1.ip
4656

    
4657
    # file storage checks
4658
    if (self.op.file_driver and
4659
        not self.op.file_driver in constants.FILE_DRIVER):
4660
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
4661
                                 self.op.file_driver)
4662

    
4663
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
4664
      raise errors.OpPrereqError("File storage directory path not absolute")
4665

    
4666
    ### Node/iallocator related checks
4667
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
4668
      raise errors.OpPrereqError("One and only one of iallocator and primary"
4669
                                 " node must be given")
4670

    
4671
    if self.op.iallocator:
4672
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4673
    else:
4674
      self.op.pnode = self._ExpandNode(self.op.pnode)
4675
      nodelist = [self.op.pnode]
4676
      if self.op.snode is not None:
4677
        self.op.snode = self._ExpandNode(self.op.snode)
4678
        nodelist.append(self.op.snode)
4679
      self.needed_locks[locking.LEVEL_NODE] = nodelist
4680

    
4681
    # in case of import lock the source node too
4682
    if self.op.mode == constants.INSTANCE_IMPORT:
4683
      src_node = getattr(self.op, "src_node", None)
4684
      src_path = getattr(self.op, "src_path", None)
4685

    
4686
      if src_path is None:
4687
        self.op.src_path = src_path = self.op.instance_name
4688

    
4689
      if src_node is None:
4690
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4691
        self.op.src_node = None
4692
        if os.path.isabs(src_path):
4693
          raise errors.OpPrereqError("Importing an instance from an absolute"
4694
                                     " path requires a source node option.")
4695
      else:
4696
        self.op.src_node = src_node = self._ExpandNode(src_node)
4697
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4698
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
4699
        if not os.path.isabs(src_path):
4700
          self.op.src_path = src_path = \
4701
            os.path.join(constants.EXPORT_DIR, src_path)
4702

    
4703
    else: # INSTANCE_CREATE
4704
      if getattr(self.op, "os_type", None) is None:
4705
        raise errors.OpPrereqError("No guest OS specified")
4706

    
4707
  def _RunAllocator(self):
4708
    """Run the allocator based on input opcode.
4709

4710
    """
4711
    nics = [n.ToDict() for n in self.nics]
4712
    ial = IAllocator(self,
4713
                     mode=constants.IALLOCATOR_MODE_ALLOC,
4714
                     name=self.op.instance_name,
4715
                     disk_template=self.op.disk_template,
4716
                     tags=[],
4717
                     os=self.op.os_type,
4718
                     vcpus=self.be_full[constants.BE_VCPUS],
4719
                     mem_size=self.be_full[constants.BE_MEMORY],
4720
                     disks=self.disks,
4721
                     nics=nics,
4722
                     hypervisor=self.op.hypervisor,
4723
                     )
4724

    
4725
    ial.Run(self.op.iallocator)
4726

    
4727
    if not ial.success:
4728
      raise errors.OpPrereqError("Can't compute nodes using"
4729
                                 " iallocator '%s': %s" % (self.op.iallocator,
4730
                                                           ial.info))
4731
    if len(ial.nodes) != ial.required_nodes:
4732
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4733
                                 " of nodes (%s), required %s" %
4734
                                 (self.op.iallocator, len(ial.nodes),
4735
                                  ial.required_nodes))
4736
    self.op.pnode = ial.nodes[0]
4737
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
4738
                 self.op.instance_name, self.op.iallocator,
4739
                 ", ".join(ial.nodes))
4740
    if ial.required_nodes == 2:
4741
      self.op.snode = ial.nodes[1]
4742

    
4743
  def BuildHooksEnv(self):
4744
    """Build hooks env.
4745

4746
    This runs on master, primary and secondary nodes of the instance.
4747

4748
    """
4749
    env = {
4750
      "ADD_MODE": self.op.mode,
4751
      }
4752
    if self.op.mode == constants.INSTANCE_IMPORT:
4753
      env["SRC_NODE"] = self.op.src_node
4754
      env["SRC_PATH"] = self.op.src_path
4755
      env["SRC_IMAGES"] = self.src_images
4756

    
4757
    env.update(_BuildInstanceHookEnv(
4758
      name=self.op.instance_name,
4759
      primary_node=self.op.pnode,
4760
      secondary_nodes=self.secondaries,
4761
      status=self.op.start,
4762
      os_type=self.op.os_type,
4763
      memory=self.be_full[constants.BE_MEMORY],
4764
      vcpus=self.be_full[constants.BE_VCPUS],
4765
      nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
4766
      disk_template=self.op.disk_template,
4767
      disks=[(d["size"], d["mode"]) for d in self.disks],
4768
      bep=self.be_full,
4769
      hvp=self.hv_full,
4770
      hypervisor_name=self.op.hypervisor,
4771
    ))
4772

    
4773
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
4774
          self.secondaries)
4775
    return env, nl, nl
4776

    
4777

    
4778
  def CheckPrereq(self):
4779
    """Check prerequisites.
4780

4781
    """
4782
    if (not self.cfg.GetVGName() and
4783
        self.op.disk_template not in constants.DTS_NOT_LVM):
4784
      raise errors.OpPrereqError("Cluster does not support lvm-based"
4785
                                 " instances")
4786

    
4787
    if self.op.mode == constants.INSTANCE_IMPORT:
4788
      src_node = self.op.src_node
4789
      src_path = self.op.src_path
4790

    
4791
      if src_node is None:
4792
        exp_list = self.rpc.call_export_list(
4793
          self.acquired_locks[locking.LEVEL_NODE])
4794
        found = False
4795
        for node in exp_list:
4796
          if not exp_list[node].failed and src_path in exp_list[node].data:
4797
            found = True
4798
            self.op.src_node = src_node = node
4799
            self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
4800
                                                       src_path)
4801
            break
4802
        if not found:
4803
          raise errors.OpPrereqError("No export found for relative path %s" %
4804
                                      src_path)
4805

    
4806
      _CheckNodeOnline(self, src_node)
4807
      result = self.rpc.call_export_info(src_node, src_path)
4808
      result.Raise()
4809
      if not result.data:
4810
        raise errors.OpPrereqError("No export found in dir %s" % src_path)
4811

    
4812
      export_info = result.data
4813
      if not export_info.has_section(constants.INISECT_EXP):
4814
        raise errors.ProgrammerError("Corrupted export config")
4815

    
4816
      ei_version = export_info.get(constants.INISECT_EXP, 'version')
4817
      if (int(ei_version) != constants.EXPORT_VERSION):
4818
        raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
4819
                                   (ei_version, constants.EXPORT_VERSION))
4820

    
4821
      # Check that the new instance doesn't have less disks than the export
4822
      instance_disks = len(self.disks)
4823
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
4824
      if instance_disks < export_disks:
4825
        raise errors.OpPrereqError("Not enough disks to import."
4826
                                   " (instance: %d, export: %d)" %
4827
                                   (instance_disks, export_disks))
4828

    
4829
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
4830
      disk_images = []
4831
      for idx in range(export_disks):
4832
        option = 'disk%d_dump' % idx
4833
        if export_info.has_option(constants.INISECT_INS, option):
4834
          # FIXME: are the old os-es, disk sizes, etc. useful?
4835
          export_name = export_info.get(constants.INISECT_INS, option)
4836
          image = os.path.join(src_path, export_name)
4837
          disk_images.append(image)
4838
        else:
4839
          disk_images.append(False)
4840

    
4841
      self.src_images = disk_images
4842

    
4843
      old_name = export_info.get(constants.INISECT_INS, 'name')
4844
      # FIXME: int() here could throw a ValueError on broken exports
4845
      exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
4846
      if self.op.instance_name == old_name:
4847
        for idx, nic in enumerate(self.nics):
4848
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
4849
            nic_mac_ini = 'nic%d_mac' % idx
4850
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
4851

    
4852
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
4853
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
4854
    if self.op.start and not self.op.ip_check:
4855
      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
4856
                                 " adding an instance in start mode")
4857

    
4858
    if self.op.ip_check:
4859
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
4860
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4861
                                   (self.check_ip, self.op.instance_name))
4862

    
4863
    #### mac address generation
4864
    # By generating here the mac address both the allocator and the hooks get
4865
    # the real final mac address rather than the 'auto' or 'generate' value.
4866
    # There is a race condition between the generation and the instance object
4867
    # creation, which means that we know the mac is valid now, but we're not
4868
    # sure it will be when we actually add the instance. If things go bad
4869
    # adding the instance will abort because of a duplicate mac, and the
4870
    # creation job will fail.
4871
    for nic in self.nics:
4872
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4873
        nic.mac = self.cfg.GenerateMAC()
4874

    
4875
    #### allocator run
4876

    
4877
    if self.op.iallocator is not None:
4878
      self._RunAllocator()
4879

    
4880
    #### node related checks
4881

    
4882
    # check primary node
4883
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
4884
    assert self.pnode is not None, \
4885
      "Cannot retrieve locked node %s" % self.op.pnode
4886
    if pnode.offline:
4887
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
4888
                                 pnode.name)
4889
    if pnode.drained:
4890
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
4891
                                 pnode.name)
4892

    
4893
    self.secondaries = []
4894

    
4895
    # mirror node verification
4896
    if self.op.disk_template in constants.DTS_NET_MIRROR:
4897
      if self.op.snode is None:
4898
        raise errors.OpPrereqError("The networked disk templates need"
4899
                                   " a mirror node")
4900
      if self.op.snode == pnode.name:
4901
        raise errors.OpPrereqError("The secondary node cannot be"
4902
                                   " the primary node.")
4903
      _CheckNodeOnline(self, self.op.snode)
4904
      _CheckNodeNotDrained(self, self.op.snode)
4905
      self.secondaries.append(self.op.snode)
4906

    
4907
    nodenames = [pnode.name] + self.secondaries
4908

    
4909
    req_size = _ComputeDiskSize(self.op.disk_template,
4910
                                self.disks)
4911

    
4912
    # Check lv size requirements
4913
    if req_size is not None:
4914
      nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
4915
                                         self.op.hypervisor)
4916
      for node in nodenames:
4917
        info = nodeinfo[node]
4918
        info.Raise()
4919
        info = info.data
4920
        if not info:
4921
          raise errors.OpPrereqError("Cannot get current information"
4922
                                     " from node '%s'" % node)
4923
        vg_free = info.get('vg_free', None)
4924
        if not isinstance(vg_free, int):
4925
          raise errors.OpPrereqError("Can't compute free disk space on"
4926
                                     " node %s" % node)
4927
        if req_size > info['vg_free']:
4928
          raise errors.OpPrereqError("Not enough disk space on target node %s."
4929
                                     " %d MB available, %d MB required" %
4930
                                     (node, info['vg_free'], req_size))
4931

    
4932
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
4933

    
4934
    # os verification
4935
    result = self.rpc.call_os_get(pnode.name, self.op.os_type)
4936
    result.Raise()
4937
    if not isinstance(result.data, objects.OS) or not result.data:
4938
      raise errors.OpPrereqError("OS '%s' not in supported os list for"
4939
                                 " primary node"  % self.op.os_type)
4940

    
4941
    # bridge check on primary node
4942
    bridges = [n.bridge for n in self.nics]
4943
    result = self.rpc.call_bridges_exist(self.pnode.name, bridges)
4944
    result.Raise()
4945
    if not result.data:
4946
      raise errors.OpPrereqError("One of the target bridges '%s' does not"
4947
                                 " exist on destination node '%s'" %
4948
                                 (",".join(bridges), pnode.name))
4949

    
4950
    # memory check on primary node
4951
    if self.op.start:
4952
      _CheckNodeFreeMemory(self, self.pnode.name,
4953
                           "creating instance %s" % self.op.instance_name,
4954
                           self.be_full[constants.BE_MEMORY],
4955
                           self.op.hypervisor)
4956

    
4957
  def Exec(self, feedback_fn):
4958
    """Create and add the instance to the cluster.
4959

4960
    """
4961
    instance = self.op.instance_name
4962
    pnode_name = self.pnode.name
4963

    
4964
    ht_kind = self.op.hypervisor
4965
    if ht_kind in constants.HTS_REQ_PORT:
4966
      network_port = self.cfg.AllocatePort()
4967
    else:
4968
      network_port = None
4969

    
4970
    ##if self.op.vnc_bind_address is None:
4971
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
4972

    
4973
    # this is needed because os.path.join does not accept None arguments
4974
    if self.op.file_storage_dir is None:
4975
      string_file_storage_dir = ""
4976
    else:
4977
      string_file_storage_dir = self.op.file_storage_dir
4978

    
4979
    # build the full file storage dir path
4980
    file_storage_dir = os.path.normpath(os.path.join(
4981
                                        self.cfg.GetFileStorageDir(),
4982
                                        string_file_storage_dir, instance))
4983

    
4984

    
4985
    disks = _GenerateDiskTemplate(self,
4986
                                  self.op.disk_template,
4987
                                  instance, pnode_name,
4988
                                  self.secondaries,
4989
                                  self.disks,
4990
                                  file_storage_dir,
4991
                                  self.op.file_driver,
4992
                                  0)
4993

    
4994
    iobj = objects.Instance(name=instance, os=self.op.os_type,
4995
                            primary_node=pnode_name,
4996
                            nics=self.nics, disks=disks,
4997
                            disk_template=self.op.disk_template,
4998
                            admin_up=False,
4999
                            network_port=network_port,
5000
                            beparams=self.op.beparams,
5001
                            hvparams=self.op.hvparams,
5002
                            hypervisor=self.op.hypervisor,
5003
                            )
5004

    
5005
    feedback_fn("* creating instance disks...")
5006
    try:
5007
      _CreateDisks(self, iobj)
5008
    except errors.OpExecError:
5009
      self.LogWarning("Device creation failed, reverting...")
5010
      try:
5011
        _RemoveDisks(self, iobj)
5012
      finally:
5013
        self.cfg.ReleaseDRBDMinors(instance)
5014
        raise
5015

    
5016
    feedback_fn("adding instance %s to cluster config" % instance)
5017

    
5018
    self.cfg.AddInstance(iobj)
5019
    # Declare that we don't want to remove the instance lock anymore, as we've
5020
    # added the instance to the config
5021
    del self.remove_locks[locking.LEVEL_INSTANCE]
5022
    # Unlock all the nodes
5023
    if self.op.mode == constants.INSTANCE_IMPORT:
5024
      nodes_keep = [self.op.src_node]
5025
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
5026
                       if node != self.op.src_node]
5027
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
5028
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
5029
    else:
5030
      self.context.glm.release(locking.LEVEL_NODE)
5031
      del self.acquired_locks[locking.LEVEL_NODE]
5032

    
5033
    if self.op.wait_for_sync:
5034
      disk_abort = not _WaitForSync(self, iobj)
5035
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
5036
      # make sure the disks are not degraded (still sync-ing is ok)
5037
      time.sleep(15)
5038
      feedback_fn("* checking mirrors status")
5039
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
5040
    else:
5041
      disk_abort = False
5042

    
5043
    if disk_abort:
5044
      _RemoveDisks(self, iobj)
5045
      self.cfg.RemoveInstance(iobj.name)
5046
      # Make sure the instance lock gets removed
5047
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
5048
      raise errors.OpExecError("There are some degraded disks for"
5049
                               " this instance")
5050

    
5051
    feedback_fn("creating os for instance %s on node %s" %
5052
                (instance, pnode_name))
5053

    
5054
    if iobj.disk_template != constants.DT_DISKLESS:
5055
      if self.op.mode == constants.INSTANCE_CREATE:
5056
        feedback_fn("* running the instance OS create scripts...")
5057
        result = self.rpc.call_instance_os_add(pnode_name, iobj)
5058
        msg = result.RemoteFailMsg()
5059
        if msg:
5060
          raise errors.OpExecError("Could not add os for instance %s"
5061
                                   " on node %s: %s" %
5062
                                   (instance, pnode_name, msg))
5063

    
5064
      elif self.op.mode == constants.INSTANCE_IMPORT:
5065
        feedback_fn("* running the instance OS import scripts...")
5066
        src_node = self.op.src_node
5067
        src_images = self.src_images
5068
        cluster_name = self.cfg.GetClusterName()
5069
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
5070
                                                         src_node, src_images,
5071
                                                         cluster_name)
5072
        import_result.Raise()
5073
        for idx, result in enumerate(import_result.data):
5074
          if not result:
5075
            self.LogWarning("Could not import the image %s for instance"
5076
                            " %s, disk %d, on node %s" %
5077
                            (src_images[idx], instance, idx, pnode_name))
5078
      else:
5079
        # also checked in the prereq part
5080
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
5081
                                     % self.op.mode)
5082

    
5083
    if self.op.start:
5084
      iobj.admin_up = True
5085
      self.cfg.Update(iobj)
5086
      logging.info("Starting instance %s on node %s", instance, pnode_name)
5087
      feedback_fn("* starting instance...")
5088
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
5089
      msg = result.RemoteFailMsg()
5090
      if msg:
5091
        raise errors.OpExecError("Could not start instance: %s" % msg)
5092

    
5093

    
5094
class LUConnectConsole(NoHooksLU):
5095
  """Connect to an instance's console.
5096

5097
  This is somewhat special in that it returns the command line that
5098
  you need to run on the master node in order to connect to the
5099
  console.
5100

5101
  """
5102
  _OP_REQP = ["instance_name"]
5103
  REQ_BGL = False
5104

    
5105
  def ExpandNames(self):
5106
    self._ExpandAndLockInstance()
5107

    
5108
  def CheckPrereq(self):
5109
    """Check prerequisites.
5110

5111
    This checks that the instance is in the cluster.
5112

5113
    """
5114
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5115
    assert self.instance is not None, \
5116
      "Cannot retrieve locked instance %s" % self.op.instance_name
5117
    _CheckNodeOnline(self, self.instance.primary_node)
5118

    
5119
  def Exec(self, feedback_fn):
5120
    """Connect to the console of an instance
5121

5122
    """
5123
    instance = self.instance
5124
    node = instance.primary_node
5125

    
5126
    node_insts = self.rpc.call_instance_list([node],
5127
                                             [instance.hypervisor])[node]
5128
    node_insts.Raise()
5129

    
5130
    if instance.name not in node_insts.data:
5131
      raise errors.OpExecError("Instance %s is not running." % instance.name)
5132

    
5133
    logging.debug("Connecting to console of %s on %s", instance.name, node)
5134

    
5135
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
5136
    cluster = self.cfg.GetClusterInfo()
5137
    # beparams and hvparams are passed separately, to avoid editing the
5138
    # instance and then saving the defaults in the instance itself.
5139
    hvparams = cluster.FillHV(instance)
5140
    beparams = cluster.FillBE(instance)
5141
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
5142

    
5143
    # build ssh cmdline
5144
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
5145

    
5146

    
5147
class LUReplaceDisks(LogicalUnit):
5148
  """Replace the disks of an instance.
5149

5150
  """
5151
  HPATH = "mirrors-replace"
5152
  HTYPE = constants.HTYPE_INSTANCE
5153
  _OP_REQP = ["instance_name", "mode", "disks"]
5154
  REQ_BGL = False
5155

    
5156
  def CheckArguments(self):
5157
    if not hasattr(self.op, "remote_node"):
5158
      self.op.remote_node = None
5159
    if not hasattr(self.op, "iallocator"):
5160
      self.op.iallocator = None
5161

    
5162
    # check for valid parameter combination
5163
    cnt = [self.op.remote_node, self.op.iallocator].count(None)
5164
    if self.op.mode == constants.REPLACE_DISK_CHG:
5165
      if cnt == 2:
5166
        raise errors.OpPrereqError("When changing the secondary either an"
5167
                                   " iallocator script must be used or the"
5168
                                   " new node given")
5169
      elif cnt == 0:
5170
        raise errors.OpPrereqError("Give either the iallocator or the new"
5171
                                   " secondary, not both")
5172
    else: # not replacing the secondary
5173
      if cnt != 2:
5174
        raise errors.OpPrereqError("The iallocator and new node options can"
5175
                                   " be used only when changing the"
5176
                                   " secondary node")
5177

    
5178
  def ExpandNames(self):
5179
    self._ExpandAndLockInstance()
5180

    
5181
    if self.op.iallocator is not None:
5182
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5183
    elif self.op.remote_node is not None:
5184
      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
5185
      if remote_node is None:
5186
        raise errors.OpPrereqError("Node '%s' not known" %
5187
                                   self.op.remote_node)
5188
      self.op.remote_node = remote_node
5189
      # Warning: do not remove the locking of the new secondary here
5190
      # unless DRBD8.AddChildren is changed to work in parallel;
5191
      # currently it doesn't since parallel invocations of
5192
      # FindUnusedMinor will conflict
5193
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
5194
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5195
    else:
5196
      self.needed_locks[locking.LEVEL_NODE] = []
5197
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5198

    
5199
  def DeclareLocks(self, level):
5200
    # If we're not already locking all nodes in the set we have to declare the
5201
    # instance's primary/secondary nodes.
5202
    if (level == locking.LEVEL_NODE and
5203
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
5204
      self._LockInstancesNodes()
5205

    
5206
  def _RunAllocator(self):
5207
    """Compute a new secondary node using an IAllocator.
5208

5209
    """
5210
    ial = IAllocator(self,
5211
                     mode=constants.IALLOCATOR_MODE_RELOC,
5212
                     name=self.op.instance_name,
5213
                     relocate_from=[self.sec_node])
5214

    
5215
    ial.Run(self.op.iallocator)
5216

    
5217
    if not ial.success:
5218
      raise errors.OpPrereqError("Can't compute nodes using"
5219
                                 " iallocator '%s': %s" % (self.op.iallocator,
5220
                                                           ial.info))
5221
    if len(ial.nodes) != ial.required_nodes:
5222
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
5223
                                 " of nodes (%s), required %s" %
5224
                                 (self.op.iallocator,
5225
                                  len(ial.nodes), ial.required_nodes))
5226
    self.op.remote_node = ial.nodes[0]
5227
    self.LogInfo("Selected new secondary for the instance: %s",
5228
                 self.op.remote_node)
5229

    
5230
  def BuildHooksEnv(self):
5231
    """Build hooks env.
5232

5233
    This runs on the master, the primary and all the secondaries.
5234

5235
    """
5236
    env = {
5237
      "MODE": self.op.mode,
5238
      "NEW_SECONDARY": self.op.remote_node,
5239
      "OLD_SECONDARY": self.instance.secondary_nodes[0],
5240
      }
5241
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5242
    nl = [
5243
      self.cfg.GetMasterNode(),
5244
      self.instance.primary_node,
5245
      ]
5246
    if self.op.remote_node is not None:
5247
      nl.append(self.op.remote_node)
5248
    return env, nl, nl
5249

    
5250
  def CheckPrereq(self):
5251
    """Check prerequisites.
5252

5253
    This checks that the instance is in the cluster.
5254

5255
    """
5256
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5257
    assert instance is not None, \
5258
      "Cannot retrieve locked instance %s" % self.op.instance_name
5259
    self.instance = instance
5260

    
5261
    if instance.disk_template != constants.DT_DRBD8:
5262
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
5263
                                 " instances")
5264

    
5265
    if len(instance.secondary_nodes) != 1:
5266
      raise errors.OpPrereqError("The instance has a strange layout,"
5267
                                 " expected one secondary but found %d" %
5268
                                 len(instance.secondary_nodes))
5269

    
5270
    self.sec_node = instance.secondary_nodes[0]
5271

    
5272
    if self.op.iallocator is not None:
5273
      self._RunAllocator()
5274

    
5275
    remote_node = self.op.remote_node
5276
    if remote_node is not None:
5277
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
5278
      assert self.remote_node_info is not None, \
5279
        "Cannot retrieve locked node %s" % remote_node
5280
    else:
5281
      self.remote_node_info = None
5282
    if remote_node == instance.primary_node:
5283
      raise errors.OpPrereqError("The specified node is the primary node of"
5284
                                 " the instance.")
5285
    elif remote_node == self.sec_node:
5286
      raise errors.OpPrereqError("The specified node is already the"
5287
                                 " secondary node of the instance.")
5288

    
5289
    if self.op.mode == constants.REPLACE_DISK_PRI:
5290
      n1 = self.tgt_node = instance.primary_node
5291
      n2 = self.oth_node = self.sec_node
5292
    elif self.op.mode == constants.REPLACE_DISK_SEC:
5293
      n1 = self.tgt_node = self.sec_node
5294
      n2 = self.oth_node = instance.primary_node
5295
    elif self.op.mode == constants.REPLACE_DISK_CHG:
5296
      n1 = self.new_node = remote_node
5297
      n2 = self.oth_node = instance.primary_node
5298
      self.tgt_node = self.sec_node
5299
      _CheckNodeNotDrained(self, remote_node)
5300
    else:
5301
      raise errors.ProgrammerError("Unhandled disk replace mode")
5302

    
5303
    _CheckNodeOnline(self, n1)
5304
    _CheckNodeOnline(self, n2)
5305

    
5306
    if not self.op.disks:
5307
      self.op.disks = range(len(instance.disks))
5308

    
5309
    for disk_idx in self.op.disks:
5310
      instance.FindDisk(disk_idx)
5311

    
5312
  def _ExecD8DiskOnly(self, feedback_fn):
5313
    """Replace a disk on the primary or secondary for dbrd8.
5314

5315
    The algorithm for replace is quite complicated:
5316

5317
      1. for each disk to be replaced:
5318

5319
        1. create new LVs on the target node with unique names
5320
        1. detach old LVs from the drbd device
5321
        1. rename old LVs to name_replaced.<time_t>
5322
        1. rename new LVs to old LVs
5323
        1. attach the new LVs (with the old names now) to the drbd device
5324

5325
      1. wait for sync across all devices
5326

5327
      1. for each modified disk:
5328

5329
        1. remove old LVs (which have the name name_replaces.<time_t>)
5330

5331
    Failures are not very well handled.
5332

5333
    """
5334
    steps_total = 6
5335
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5336
    instance = self.instance
5337
    iv_names = {}
5338
    vgname = self.cfg.GetVGName()
5339
    # start of work
5340
    cfg = self.cfg
5341
    tgt_node = self.tgt_node
5342
    oth_node = self.oth_node
5343

    
5344
    # Step: check device activation
5345
    self.proc.LogStep(1, steps_total, "check device existence")
5346
    info("checking volume groups")
5347
    my_vg = cfg.GetVGName()
5348
    results = self.rpc.call_vg_list([oth_node, tgt_node])
5349
    if not results:
5350
      raise errors.OpExecError("Can't list volume groups on the nodes")
5351
    for node in oth_node, tgt_node:
5352
      res = results[node]
5353
      if res.failed or not res.data or my_vg not in res.data:
5354
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5355
                                 (my_vg, node))
5356
    for idx, dev in enumerate(instance.disks):
5357
      if idx not in self.op.disks:
5358
        continue
5359
      for node in tgt_node, oth_node:
5360
        info("checking disk/%d on %s" % (idx, node))
5361
        cfg.SetDiskID(dev, node)
5362
        result = self.rpc.call_blockdev_find(node, dev)
5363
        msg = result.RemoteFailMsg()
5364
        if not msg and not result.payload:
5365
          msg = "disk not found"
5366
        if msg:
5367
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5368
                                   (idx, node, msg))
5369

    
5370
    # Step: check other node consistency
5371
    self.proc.LogStep(2, steps_total, "check peer consistency")
5372
    for idx, dev in enumerate(instance.disks):
5373
      if idx not in self.op.disks:
5374
        continue
5375
      info("checking disk/%d consistency on %s" % (idx, oth_node))
5376
      if not _CheckDiskConsistency(self, dev, oth_node,
5377
                                   oth_node==instance.primary_node):
5378
        raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
5379
                                 " to replace disks on this node (%s)" %
5380
                                 (oth_node, tgt_node))
5381

    
5382
    # Step: create new storage
5383
    self.proc.LogStep(3, steps_total, "allocate new storage")
5384
    for idx, dev in enumerate(instance.disks):
5385
      if idx not in self.op.disks:
5386
        continue
5387
      size = dev.size
5388
      cfg.SetDiskID(dev, tgt_node)
5389
      lv_names = [".disk%d_%s" % (idx, suf)
5390
                  for suf in ["data", "meta"]]
5391
      names = _GenerateUniqueNames(self, lv_names)
5392
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
5393
                             logical_id=(vgname, names[0]))
5394
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
5395
                             logical_id=(vgname, names[1]))
5396
      new_lvs = [lv_data, lv_meta]
5397
      old_lvs = dev.children
5398
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
5399
      info("creating new local storage on %s for %s" %
5400
           (tgt_node, dev.iv_name))
5401
      # we pass force_create=True to force the LVM creation
5402
      for new_lv in new_lvs:
5403
        _CreateBlockDev(self, tgt_node, instance, new_lv, True,
5404
                        _GetInstanceInfoText(instance), False)
5405

    
5406
    # Step: for each lv, detach+rename*2+attach
5407
    self.proc.LogStep(4, steps_total, "change drbd configuration")
5408
    for dev, old_lvs, new_lvs in iv_names.itervalues():
5409
      info("detaching %s drbd from local storage" % dev.iv_name)
5410
      result = self.rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs)
5411
      result.Raise()
5412
      if not result.data:
5413
        raise errors.OpExecError("Can't detach drbd from local storage on node"
5414
                                 " %s for device %s" % (tgt_node, dev.iv_name))
5415
      #dev.children = []
5416
      #cfg.Update(instance)
5417

    
5418
      # ok, we created the new LVs, so now we know we have the needed
5419
      # storage; as such, we proceed on the target node to rename
5420
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
5421
      # using the assumption that logical_id == physical_id (which in
5422
      # turn is the unique_id on that node)
5423

    
5424
      # FIXME(iustin): use a better name for the replaced LVs
5425
      temp_suffix = int(time.time())
5426
      ren_fn = lambda d, suff: (d.physical_id[0],
5427
                                d.physical_id[1] + "_replaced-%s" % suff)
5428
      # build the rename list based on what LVs exist on the node
5429
      rlist = []
5430
      for to_ren in old_lvs:
5431
        result = self.rpc.call_blockdev_find(tgt_node, to_ren)
5432
        if not result.RemoteFailMsg() and result.payload:
5433
          # device exists
5434
          rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
5435

    
5436
      info("renaming the old LVs on the target node")
5437
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5438
      result.Raise()
5439
      if not result.data:
5440
        raise errors.OpExecError("Can't rename old LVs on node %s" % tgt_node)
5441
      # now we rename the new LVs to the old LVs
5442
      info("renaming the new LVs on the target node")
5443
      rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
5444
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5445
      result.Raise()
5446
      if not result.data:
5447
        raise errors.OpExecError("Can't rename new LVs on node %s" % tgt_node)
5448

    
5449
      for old, new in zip(old_lvs, new_lvs):
5450
        new.logical_id = old.logical_id
5451
        cfg.SetDiskID(new, tgt_node)
5452

    
5453
      for disk in old_lvs:
5454
        disk.logical_id = ren_fn(disk, temp_suffix)
5455
        cfg.SetDiskID(disk, tgt_node)
5456

    
5457
      # now that the new lvs have the old name, we can add them to the device
5458
      info("adding new mirror component on %s" % tgt_node)
5459
      result = self.rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs)
5460
      if result.failed or not result.data:
5461
        for new_lv in new_lvs:
5462
          msg = self.rpc.call_blockdev_remove(tgt_node, new_lv).RemoteFailMsg()
5463
          if msg:
5464
            warning("Can't rollback device %s: %s", dev, msg,
5465
                    hint="cleanup manually the unused logical volumes")
5466
        raise errors.OpExecError("Can't add local storage to drbd")
5467

    
5468
      dev.children = new_lvs
5469
      cfg.Update(instance)
5470

    
5471
    # Step: wait for sync
5472

    
5473
    # this can fail as the old devices are degraded and _WaitForSync
5474
    # does a combined result over all disks, so we don't check its
5475
    # return value
5476
    self.proc.LogStep(5, steps_total, "sync devices")
5477
    _WaitForSync(self, instance, unlock=True)
5478

    
5479
    # so check manually all the devices
5480
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5481
      cfg.SetDiskID(dev, instance.primary_node)
5482
      result = self.rpc.call_blockdev_find(instance.primary_node, dev)
5483
      msg = result.RemoteFailMsg()
5484
      if not msg and not result.payload:
5485
        msg = "disk not found"
5486
      if msg:
5487
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
5488
                                 (name, msg))
5489
      if result.payload[5]:
5490
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
5491

    
5492
    # Step: remove old storage
5493
    self.proc.LogStep(6, steps_total, "removing old storage")
5494
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5495
      info("remove logical volumes for %s" % name)
5496
      for lv in old_lvs:
5497
        cfg.SetDiskID(lv, tgt_node)
5498
        msg = self.rpc.call_blockdev_remove(tgt_node, lv).RemoteFailMsg()
5499
        if msg:
5500
          warning("Can't remove old LV: %s" % msg,
5501
                  hint="manually remove unused LVs")
5502
          continue
5503

    
5504
  def _ExecD8Secondary(self, feedback_fn):
5505
    """Replace the secondary node for drbd8.
5506

5507
    The algorithm for replace is quite complicated:
5508
      - for all disks of the instance:
5509
        - create new LVs on the new node with same names
5510
        - shutdown the drbd device on the old secondary
5511
        - disconnect the drbd network on the primary
5512
        - create the drbd device on the new secondary
5513
        - network attach the drbd on the primary, using an artifice:
5514
          the drbd code for Attach() will connect to the network if it
5515
          finds a device which is connected to the good local disks but
5516
          not network enabled
5517
      - wait for sync across all devices
5518
      - remove all disks from the old secondary
5519

5520
    Failures are not very well handled.
5521

5522
    """
5523
    steps_total = 6
5524
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5525
    instance = self.instance
5526
    iv_names = {}
5527
    # start of work
5528
    cfg = self.cfg
5529
    old_node = self.tgt_node
5530
    new_node = self.new_node
5531
    pri_node = instance.primary_node
5532
    nodes_ip = {
5533
      old_node: self.cfg.GetNodeInfo(old_node).secondary_ip,
5534
      new_node: self.cfg.GetNodeInfo(new_node).secondary_ip,
5535
      pri_node: self.cfg.GetNodeInfo(pri_node).secondary_ip,
5536
      }
5537

    
5538
    # Step: check device activation
5539
    self.proc.LogStep(1, steps_total, "check device existence")
5540
    info("checking volume groups")
5541
    my_vg = cfg.GetVGName()
5542
    results = self.rpc.call_vg_list([pri_node, new_node])
5543
    for node in pri_node, new_node:
5544
      res = results[node]
5545
      if res.failed or not res.data or my_vg not in res.data:
5546
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5547
                                 (my_vg, node))
5548
    for idx, dev in enumerate(instance.disks):
5549
      if idx not in self.op.disks:
5550
        continue
5551
      info("checking disk/%d on %s" % (idx, pri_node))
5552
      cfg.SetDiskID(dev, pri_node)
5553
      result = self.rpc.call_blockdev_find(pri_node, dev)
5554
      msg = result.RemoteFailMsg()
5555
      if not msg and not result.payload:
5556
        msg = "disk not found"
5557
      if msg:
5558
        raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5559
                                 (idx, pri_node, msg))
5560

    
5561
    # Step: check other node consistency
5562
    self.proc.LogStep(2, steps_total, "check peer consistency")
5563
    for idx, dev in enumerate(instance.disks):
5564
      if idx not in self.op.disks:
5565
        continue
5566
      info("checking disk/%d consistency on %s" % (idx, pri_node))
5567
      if not _CheckDiskConsistency(self, dev, pri_node, True, ldisk=True):
5568
        raise errors.OpExecError("Primary node (%s) has degraded storage,"
5569
                                 " unsafe to replace the secondary" %
5570
                                 pri_node)
5571

    
5572
    # Step: create new storage
5573
    self.proc.LogStep(3, steps_total, "allocate new storage")
5574
    for idx, dev in enumerate(instance.disks):
5575
      info("adding new local storage on %s for disk/%d" %
5576
           (new_node, idx))
5577
      # we pass force_create=True to force LVM creation
5578
      for new_lv in dev.children:
5579
        _CreateBlockDev(self, new_node, instance, new_lv, True,
5580
                        _GetInstanceInfoText(instance), False)
5581

    
5582
    # Step 4: dbrd minors and drbd setups changes
5583
    # after this, we must manually remove the drbd minors on both the
5584
    # error and the success paths
5585
    minors = cfg.AllocateDRBDMinor([new_node for dev in instance.disks],
5586
                                   instance.name)
5587
    logging.debug("Allocated minors %s" % (minors,))
5588
    self.proc.LogStep(4, steps_total, "changing drbd configuration")
5589
    for idx, (dev, new_minor) in enumerate(zip(instance.disks, minors)):
5590
      info("activating a new drbd on %s for disk/%d" % (new_node, idx))
5591
      # create new devices on new_node; note that we create two IDs:
5592
      # one without port, so the drbd will be activated without
5593
      # networking information on the new node at this stage, and one
5594
      # with network, for the latter activation in step 4
5595
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
5596
      if pri_node == o_node1:
5597
        p_minor = o_minor1
5598
      else:
5599
        p_minor = o_minor2
5600

    
5601
      new_alone_id = (pri_node, new_node, None, p_minor, new_minor, o_secret)
5602
      new_net_id = (pri_node, new_node, o_port, p_minor, new_minor, o_secret)
5603

    
5604
      iv_names[idx] = (dev, dev.children, new_net_id)
5605
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
5606
                    new_net_id)
5607
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
5608
                              logical_id=new_alone_id,
5609
                              children=dev.children,
5610
                              size=dev.size)
5611
      try:
5612
        _CreateSingleBlockDev(self, new_node, instance, new_drbd,
5613
                              _GetInstanceInfoText(instance), False)
5614
      except errors.GenericError:
5615
        self.cfg.ReleaseDRBDMinors(instance.name)
5616
        raise
5617

    
5618
    for idx, dev in enumerate(instance.disks):
5619
      # we have new devices, shutdown the drbd on the old secondary
5620
      info("shutting down drbd for disk/%d on old node" % idx)
5621
      cfg.SetDiskID(dev, old_node)
5622
      msg = self.rpc.call_blockdev_shutdown(old_node, dev).RemoteFailMsg()
5623
      if msg:
5624
        warning("Failed to shutdown drbd for disk/%d on old node: %s" %
5625
                (idx, msg),
5626
                hint="Please cleanup this device manually as soon as possible")
5627

    
5628
    info("detaching primary drbds from the network (=> standalone)")
5629
    result = self.rpc.call_drbd_disconnect_net([pri_node], nodes_ip,
5630
                                               instance.disks)[pri_node]
5631

    
5632
    msg = result.RemoteFailMsg()
5633
    if msg:
5634
      # detaches didn't succeed (unlikely)
5635
      self.cfg.ReleaseDRBDMinors(instance.name)
5636
      raise errors.OpExecError("Can't detach the disks from the network on"
5637
                               " old node: %s" % (msg,))
5638

    
5639
    # if we managed to detach at least one, we update all the disks of
5640
    # the instance to point to the new secondary
5641
    info("updating instance configuration")
5642
    for dev, _, new_logical_id in iv_names.itervalues():
5643
      dev.logical_id = new_logical_id
5644
      cfg.SetDiskID(dev, pri_node)
5645
    cfg.Update(instance)
5646

    
5647
    # and now perform the drbd attach
5648
    info("attaching primary drbds to new secondary (standalone => connected)")
5649
    result = self.rpc.call_drbd_attach_net([pri_node, new_node], nodes_ip,
5650
                                           instance.disks, instance.name,
5651
                                           False)
5652
    for to_node, to_result in result.items():
5653
      msg = to_result.RemoteFailMsg()
5654
      if msg:
5655
        warning("can't attach drbd disks on node %s: %s", to_node, msg,
5656
                hint="please do a gnt-instance info to see the"
5657
                " status of disks")
5658

    
5659
    # this can fail as the old devices are degraded and _WaitForSync
5660
    # does a combined result over all disks, so we don't check its
5661
    # return value
5662
    self.proc.LogStep(5, steps_total, "sync devices")
5663
    _WaitForSync(self, instance, unlock=True)
5664

    
5665
    # so check manually all the devices
5666
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5667
      cfg.SetDiskID(dev, pri_node)
5668
      result = self.rpc.call_blockdev_find(pri_node, dev)
5669
      msg = result.RemoteFailMsg()
5670
      if not msg and not result.payload:
5671
        msg = "disk not found"
5672
      if msg:
5673
        raise errors.OpExecError("Can't find DRBD device disk/%d: %s" %
5674
                                 (idx, msg))
5675
      if result.payload[5]:
5676
        raise errors.OpExecError("DRBD device disk/%d is degraded!" % idx)
5677

    
5678
    self.proc.LogStep(6, steps_total, "removing old storage")
5679
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5680
      info("remove logical volumes for disk/%d" % idx)
5681
      for lv in old_lvs:
5682
        cfg.SetDiskID(lv, old_node)
5683
        msg = self.rpc.call_blockdev_remove(old_node, lv).RemoteFailMsg()
5684
        if msg:
5685
          warning("Can't remove LV on old secondary: %s", msg,
5686
                  hint="Cleanup stale volumes by hand")
5687

    
5688
  def Exec(self, feedback_fn):
5689
    """Execute disk replacement.
5690

5691
    This dispatches the disk replacement to the appropriate handler.
5692

5693
    """
5694
    instance = self.instance
5695

    
5696
    # Activate the instance disks if we're replacing them on a down instance
5697
    if not instance.admin_up:
5698
      _StartInstanceDisks(self, instance, True)
5699

    
5700
    if self.op.mode == constants.REPLACE_DISK_CHG:
5701
      fn = self._ExecD8Secondary
5702
    else:
5703
      fn = self._ExecD8DiskOnly
5704

    
5705
    ret = fn(feedback_fn)
5706

    
5707
    # Deactivate the instance disks if we're replacing them on a down instance
5708
    if not instance.admin_up:
5709
      _SafeShutdownInstanceDisks(self, instance)
5710

    
5711
    return ret
5712

    
5713

    
5714
class LUGrowDisk(LogicalUnit):
5715
  """Grow a disk of an instance.
5716

5717
  """
5718
  HPATH = "disk-grow"
5719
  HTYPE = constants.HTYPE_INSTANCE
5720
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
5721
  REQ_BGL = False
5722

    
5723
  def ExpandNames(self):
5724
    self._ExpandAndLockInstance()
5725
    self.needed_locks[locking.LEVEL_NODE] = []
5726
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5727

    
5728
  def DeclareLocks(self, level):
5729
    if level == locking.LEVEL_NODE:
5730
      self._LockInstancesNodes()
5731

    
5732
  def BuildHooksEnv(self):
5733
    """Build hooks env.
5734

5735
    This runs on the master, the primary and all the secondaries.
5736

5737
    """
5738
    env = {
5739
      "DISK": self.op.disk,
5740
      "AMOUNT": self.op.amount,
5741
      }
5742
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5743
    nl = [
5744
      self.cfg.GetMasterNode(),
5745
      self.instance.primary_node,
5746
      ]
5747
    return env, nl, nl
5748

    
5749
  def CheckPrereq(self):
5750
    """Check prerequisites.
5751

5752
    This checks that the instance is in the cluster.
5753

5754
    """
5755
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5756
    assert instance is not None, \
5757
      "Cannot retrieve locked instance %s" % self.op.instance_name
5758
    nodenames = list(instance.all_nodes)
5759
    for node in nodenames:
5760
      _CheckNodeOnline(self, node)
5761

    
5762

    
5763
    self.instance = instance
5764

    
5765
    if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8):
5766
      raise errors.OpPrereqError("Instance's disk layout does not support"
5767
                                 " growing.")
5768

    
5769
    self.disk = instance.FindDisk(self.op.disk)
5770

    
5771
    nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
5772
                                       instance.hypervisor)
5773
    for node in nodenames:
5774
      info = nodeinfo[node]
5775
      if info.failed or not info.data:
5776
        raise errors.OpPrereqError("Cannot get current information"
5777
                                   " from node '%s'" % node)
5778
      vg_free = info.data.get('vg_free', None)
5779
      if not isinstance(vg_free, int):
5780
        raise errors.OpPrereqError("Can't compute free disk space on"
5781
                                   " node %s" % node)
5782
      if self.op.amount > vg_free:
5783
        raise errors.OpPrereqError("Not enough disk space on target node %s:"
5784
                                   " %d MiB available, %d MiB required" %
5785
                                   (node, vg_free, self.op.amount))
5786

    
5787
  def Exec(self, feedback_fn):
5788
    """Execute disk grow.
5789

5790
    """
5791
    instance = self.instance
5792
    disk = self.disk
5793
    for node in instance.all_nodes:
5794
      self.cfg.SetDiskID(disk, node)
5795
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
5796
      msg = result.RemoteFailMsg()
5797
      if msg:
5798
        raise errors.OpExecError("Grow request failed to node %s: %s" %
5799
                                 (node, msg))
5800
    disk.RecordGrow(self.op.amount)
5801
    self.cfg.Update(instance)
5802
    if self.op.wait_for_sync:
5803
      disk_abort = not _WaitForSync(self, instance)
5804
      if disk_abort:
5805
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
5806
                             " status.\nPlease check the instance.")
5807

    
5808

    
5809
class LUQueryInstanceData(NoHooksLU):
5810
  """Query runtime instance data.
5811

5812
  """
5813
  _OP_REQP = ["instances", "static"]
5814
  REQ_BGL = False
5815

    
5816
  def ExpandNames(self):
5817
    self.needed_locks = {}
5818
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
5819

    
5820
    if not isinstance(self.op.instances, list):
5821
      raise errors.OpPrereqError("Invalid argument type 'instances'")
5822

    
5823
    if self.op.instances:
5824
      self.wanted_names = []
5825
      for name in self.op.instances:
5826
        full_name = self.cfg.ExpandInstanceName(name)
5827
        if full_name is None:
5828
          raise errors.OpPrereqError("Instance '%s' not known" % name)
5829
        self.wanted_names.append(full_name)
5830
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
5831
    else:
5832
      self.wanted_names = None
5833
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5834

    
5835
    self.needed_locks[locking.LEVEL_NODE] = []
5836
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5837

    
5838
  def DeclareLocks(self, level):
5839
    if level == locking.LEVEL_NODE:
5840
      self._LockInstancesNodes()
5841

    
5842
  def CheckPrereq(self):
5843
    """Check prerequisites.
5844

5845
    This only checks the optional instance list against the existing names.
5846

5847
    """
5848
    if self.wanted_names is None:
5849
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
5850

    
5851
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
5852
                             in self.wanted_names]
5853
    return
5854

    
5855
  def _ComputeDiskStatus(self, instance, snode, dev):
5856
    """Compute block device status.
5857

5858
    """
5859
    static = self.op.static
5860
    if not static:
5861
      self.cfg.SetDiskID(dev, instance.primary_node)
5862
      dev_pstatus = self.rpc.call_blockdev_find(instance.primary_node, dev)
5863
      if dev_pstatus.offline:
5864
        dev_pstatus = None
5865
      else:
5866
        msg = dev_pstatus.RemoteFailMsg()
5867
        if msg:
5868
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5869
                                   (instance.name, msg))
5870
        dev_pstatus = dev_pstatus.payload
5871
    else:
5872
      dev_pstatus = None
5873

    
5874
    if dev.dev_type in constants.LDS_DRBD:
5875
      # we change the snode then (otherwise we use the one passed in)
5876
      if dev.logical_id[0] == instance.primary_node:
5877
        snode = dev.logical_id[1]
5878
      else:
5879
        snode = dev.logical_id[0]
5880

    
5881
    if snode and not static:
5882
      self.cfg.SetDiskID(dev, snode)
5883
      dev_sstatus = self.rpc.call_blockdev_find(snode, dev)
5884
      if dev_sstatus.offline:
5885
        dev_sstatus = None
5886
      else:
5887
        msg = dev_sstatus.RemoteFailMsg()
5888
        if msg:
5889
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5890
                                   (instance.name, msg))
5891
        dev_sstatus = dev_sstatus.payload
5892
    else:
5893
      dev_sstatus = None
5894

    
5895
    if dev.children:
5896
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
5897
                      for child in dev.children]
5898
    else:
5899
      dev_children = []
5900

    
5901
    data = {
5902
      "iv_name": dev.iv_name,
5903
      "dev_type": dev.dev_type,
5904
      "logical_id": dev.logical_id,
5905
      "physical_id": dev.physical_id,
5906
      "pstatus": dev_pstatus,
5907
      "sstatus": dev_sstatus,
5908
      "children": dev_children,
5909
      "mode": dev.mode,
5910
      "size": dev.size,
5911
      }
5912

    
5913
    return data
5914

    
5915
  def Exec(self, feedback_fn):
5916
    """Gather and return data"""
5917
    result = {}
5918

    
5919
    cluster = self.cfg.GetClusterInfo()
5920

    
5921
    for instance in self.wanted_instances:
5922
      if not self.op.static:
5923
        remote_info = self.rpc.call_instance_info(instance.primary_node,
5924
                                                  instance.name,
5925
                                                  instance.hypervisor)
5926
        remote_info.Raise()
5927
        remote_info = remote_info.data
5928
        if remote_info and "state" in remote_info:
5929
          remote_state = "up"
5930
        else:
5931
          remote_state = "down"
5932
      else:
5933
        remote_state = None
5934
      if instance.admin_up:
5935
        config_state = "up"
5936
      else:
5937
        config_state = "down"
5938

    
5939
      disks = [self._ComputeDiskStatus(instance, None, device)
5940
               for device in instance.disks]
5941

    
5942
      idict = {
5943
        "name": instance.name,
5944
        "config_state": config_state,
5945
        "run_state": remote_state,
5946
        "pnode": instance.primary_node,
5947
        "snodes": instance.secondary_nodes,
5948
        "os": instance.os,
5949
        "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
5950
        "disks": disks,
5951
        "hypervisor": instance.hypervisor,
5952
        "network_port": instance.network_port,
5953
        "hv_instance": instance.hvparams,
5954
        "hv_actual": cluster.FillHV(instance),
5955
        "be_instance": instance.beparams,
5956
        "be_actual": cluster.FillBE(instance),
5957
        }
5958

    
5959
      result[instance.name] = idict
5960

    
5961
    return result
5962

    
5963

    
5964
class LUSetInstanceParams(LogicalUnit):
5965
  """Modifies an instances's parameters.
5966

5967
  """
5968
  HPATH = "instance-modify"
5969
  HTYPE = constants.HTYPE_INSTANCE
5970
  _OP_REQP = ["instance_name"]
5971
  REQ_BGL = False
5972

    
5973
  def CheckArguments(self):
5974
    if not hasattr(self.op, 'nics'):
5975
      self.op.nics = []
5976
    if not hasattr(self.op, 'disks'):
5977
      self.op.disks = []
5978
    if not hasattr(self.op, 'beparams'):
5979
      self.op.beparams = {}
5980
    if not hasattr(self.op, 'hvparams'):
5981
      self.op.hvparams = {}
5982
    self.op.force = getattr(self.op, "force", False)
5983
    if not (self.op.nics or self.op.disks or
5984
            self.op.hvparams or self.op.beparams):
5985
      raise errors.OpPrereqError("No changes submitted")
5986

    
5987
    # Disk validation
5988
    disk_addremove = 0
5989
    for disk_op, disk_dict in self.op.disks:
5990
      if disk_op == constants.DDM_REMOVE:
5991
        disk_addremove += 1
5992
        continue
5993
      elif disk_op == constants.DDM_ADD:
5994
        disk_addremove += 1
5995
      else:
5996
        if not isinstance(disk_op, int):
5997
          raise errors.OpPrereqError("Invalid disk index")
5998
      if disk_op == constants.DDM_ADD:
5999
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
6000
        if mode not in constants.DISK_ACCESS_SET:
6001
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
6002
        size = disk_dict.get('size', None)
6003
        if size is None:
6004
          raise errors.OpPrereqError("Required disk parameter size missing")
6005
        try:
6006
          size = int(size)
6007
        except ValueError, err:
6008
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
6009
                                     str(err))
6010
        disk_dict['size'] = size
6011
      else:
6012
        # modification of disk
6013
        if 'size' in disk_dict:
6014
          raise errors.OpPrereqError("Disk size change not possible, use"
6015
                                     " grow-disk")
6016

    
6017
    if disk_addremove > 1:
6018
      raise errors.OpPrereqError("Only one disk add or remove operation"
6019
                                 " supported at a time")
6020

    
6021
    # NIC validation
6022
    nic_addremove = 0
6023
    for nic_op, nic_dict in self.op.nics:
6024
      if nic_op == constants.DDM_REMOVE:
6025
        nic_addremove += 1
6026
        continue
6027
      elif nic_op == constants.DDM_ADD:
6028
        nic_addremove += 1
6029
      else:
6030
        if not isinstance(nic_op, int):
6031
          raise errors.OpPrereqError("Invalid nic index")
6032

    
6033
      # nic_dict should be a dict
6034
      nic_ip = nic_dict.get('ip', None)
6035
      if nic_ip is not None:
6036
        if nic_ip.lower() == constants.VALUE_NONE:
6037
          nic_dict['ip'] = None
6038
        else:
6039
          if not utils.IsValidIP(nic_ip):
6040
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
6041

    
6042
      if nic_op == constants.DDM_ADD:
6043
        nic_bridge = nic_dict.get('bridge', None)
6044
        if nic_bridge is None:
6045
          nic_dict['bridge'] = self.cfg.GetDefBridge()
6046
        nic_mac = nic_dict.get('mac', None)
6047
        if nic_mac is None:
6048
          nic_dict['mac'] = constants.VALUE_AUTO
6049

    
6050
      if 'mac' in nic_dict:
6051
        nic_mac = nic_dict['mac']
6052
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
6053
          if not utils.IsValidMac(nic_mac):
6054
            raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
6055
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
6056
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
6057
                                     " modifying an existing nic")
6058

    
6059
    if nic_addremove > 1:
6060
      raise errors.OpPrereqError("Only one NIC add or remove operation"
6061
                                 " supported at a time")
6062

    
6063
  def ExpandNames(self):
6064
    self._ExpandAndLockInstance()
6065
    self.needed_locks[locking.LEVEL_NODE] = []
6066
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6067

    
6068
  def DeclareLocks(self, level):
6069
    if level == locking.LEVEL_NODE:
6070
      self._LockInstancesNodes()
6071

    
6072
  def BuildHooksEnv(self):
6073
    """Build hooks env.
6074

6075
    This runs on the master, primary and secondaries.
6076

6077
    """
6078
    args = dict()
6079
    if constants.BE_MEMORY in self.be_new:
6080
      args['memory'] = self.be_new[constants.BE_MEMORY]
6081
    if constants.BE_VCPUS in self.be_new:
6082
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
6083
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
6084
    # information at all.
6085
    if self.op.nics:
6086
      args['nics'] = []
6087
      nic_override = dict(self.op.nics)
6088
      for idx, nic in enumerate(self.instance.nics):
6089
        if idx in nic_override:
6090
          this_nic_override = nic_override[idx]
6091
        else:
6092
          this_nic_override = {}
6093
        if 'ip' in this_nic_override:
6094
          ip = this_nic_override['ip']
6095
        else:
6096
          ip = nic.ip
6097
        if 'bridge' in this_nic_override:
6098
          bridge = this_nic_override['bridge']
6099
        else:
6100
          bridge = nic.bridge
6101
        if 'mac' in this_nic_override:
6102
          mac = this_nic_override['mac']
6103
        else:
6104
          mac = nic.mac
6105
        args['nics'].append((ip, bridge, mac))
6106
      if constants.DDM_ADD in nic_override:
6107
        ip = nic_override[constants.DDM_ADD].get('ip', None)
6108
        bridge = nic_override[constants.DDM_ADD]['bridge']
6109
        mac = nic_override[constants.DDM_ADD]['mac']
6110
        args['nics'].append((ip, bridge, mac))
6111
      elif constants.DDM_REMOVE in nic_override:
6112
        del args['nics'][-1]
6113

    
6114
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
6115
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
6116
    return env, nl, nl
6117

    
6118
  def CheckPrereq(self):
6119
    """Check prerequisites.
6120

6121
    This only checks the instance list against the existing names.
6122

6123
    """
6124
    self.force = self.op.force
6125

    
6126
    # checking the new params on the primary/secondary nodes
6127

    
6128
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6129
    assert self.instance is not None, \
6130
      "Cannot retrieve locked instance %s" % self.op.instance_name
6131
    pnode = instance.primary_node
6132
    nodelist = list(instance.all_nodes)
6133

    
6134
    # hvparams processing
6135
    if self.op.hvparams:
6136
      i_hvdict = copy.deepcopy(instance.hvparams)
6137
      for key, val in self.op.hvparams.iteritems():
6138
        if val == constants.VALUE_DEFAULT:
6139
          try:
6140
            del i_hvdict[key]
6141
          except KeyError:
6142
            pass
6143
        else:
6144
          i_hvdict[key] = val
6145
      cluster = self.cfg.GetClusterInfo()
6146
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
6147
      hv_new = cluster.FillDict(cluster.hvparams[instance.hypervisor],
6148
                                i_hvdict)
6149
      # local check
6150
      hypervisor.GetHypervisor(
6151
        instance.hypervisor).CheckParameterSyntax(hv_new)
6152
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
6153
      self.hv_new = hv_new # the new actual values
6154
      self.hv_inst = i_hvdict # the new dict (without defaults)
6155
    else:
6156
      self.hv_new = self.hv_inst = {}
6157

    
6158
    # beparams processing
6159
    if self.op.beparams:
6160
      i_bedict = copy.deepcopy(instance.beparams)
6161
      for key, val in self.op.beparams.iteritems():
6162
        if val == constants.VALUE_DEFAULT:
6163
          try:
6164
            del i_bedict[key]
6165
          except KeyError:
6166
            pass
6167
        else:
6168
          i_bedict[key] = val
6169
      cluster = self.cfg.GetClusterInfo()
6170
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
6171
      be_new = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
6172
                                i_bedict)
6173
      self.be_new = be_new # the new actual values
6174
      self.be_inst = i_bedict # the new dict (without defaults)
6175
    else:
6176
      self.be_new = self.be_inst = {}
6177

    
6178
    self.warn = []
6179

    
6180
    if constants.BE_MEMORY in self.op.beparams and not self.force:
6181
      mem_check_list = [pnode]
6182
      if be_new[constants.BE_AUTO_BALANCE]:
6183
        # either we changed auto_balance to yes or it was from before
6184
        mem_check_list.extend(instance.secondary_nodes)
6185
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
6186
                                                  instance.hypervisor)
6187
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
6188
                                         instance.hypervisor)
6189
      if nodeinfo[pnode].failed or not isinstance(nodeinfo[pnode].data, dict):
6190
        # Assume the primary node is unreachable and go ahead
6191
        self.warn.append("Can't get info from primary node %s" % pnode)
6192
      else:
6193
        if not instance_info.failed and instance_info.data:
6194
          current_mem = int(instance_info.data['memory'])
6195
        else:
6196
          # Assume instance not running
6197
          # (there is a slight race condition here, but it's not very probable,
6198
          # and we have no other way to check)
6199
          current_mem = 0
6200
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
6201
                    nodeinfo[pnode].data['memory_free'])
6202
        if miss_mem > 0:
6203
          raise errors.OpPrereqError("This change will prevent the instance"
6204
                                     " from starting, due to %d MB of memory"
6205
                                     " missing on its primary node" % miss_mem)
6206

    
6207
      if be_new[constants.BE_AUTO_BALANCE]:
6208
        for node, nres in nodeinfo.iteritems():
6209
          if node not in instance.secondary_nodes:
6210
            continue
6211
          if nres.failed or not isinstance(nres.data, dict):
6212
            self.warn.append("Can't get info from secondary node %s" % node)
6213
          elif be_new[constants.BE_MEMORY] > nres.data['memory_free']:
6214
            self.warn.append("Not enough memory to failover instance to"
6215
                             " secondary node %s" % node)
6216

    
6217
    # NIC processing
6218
    for nic_op, nic_dict in self.op.nics:
6219
      if nic_op == constants.DDM_REMOVE:
6220
        if not instance.nics:
6221
          raise errors.OpPrereqError("Instance has no NICs, cannot remove")
6222
        continue
6223
      if nic_op != constants.DDM_ADD:
6224
        # an existing nic
6225
        if nic_op < 0 or nic_op >= len(instance.nics):
6226
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
6227
                                     " are 0 to %d" %
6228
                                     (nic_op, len(instance.nics)))
6229
      if 'bridge' in nic_dict:
6230
        nic_bridge = nic_dict['bridge']
6231
        if nic_bridge is None:
6232
          raise errors.OpPrereqError('Cannot set the nic bridge to None')
6233
        if not self.rpc.call_bridges_exist(pnode, [nic_bridge]):
6234
          msg = ("Bridge '%s' doesn't exist on one of"
6235
                 " the instance nodes" % nic_bridge)
6236
          if self.force:
6237
            self.warn.append(msg)
6238
          else:
6239
            raise errors.OpPrereqError(msg)
6240
      if 'mac' in nic_dict:
6241
        nic_mac = nic_dict['mac']
6242
        if nic_mac is None:
6243
          raise errors.OpPrereqError('Cannot set the nic mac to None')
6244
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
6245
          # otherwise generate the mac
6246
          nic_dict['mac'] = self.cfg.GenerateMAC()
6247
        else:
6248
          # or validate/reserve the current one
6249
          if self.cfg.IsMacInUse(nic_mac):
6250
            raise errors.OpPrereqError("MAC address %s already in use"
6251
                                       " in cluster" % nic_mac)
6252

    
6253
    # DISK processing
6254
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
6255
      raise errors.OpPrereqError("Disk operations not supported for"
6256
                                 " diskless instances")
6257
    for disk_op, disk_dict in self.op.disks:
6258
      if disk_op == constants.DDM_REMOVE:
6259
        if len(instance.disks) == 1:
6260
          raise errors.OpPrereqError("Cannot remove the last disk of"
6261
                                     " an instance")
6262
        ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
6263
        ins_l = ins_l[pnode]
6264
        if ins_l.failed or not isinstance(ins_l.data, list):
6265
          raise errors.OpPrereqError("Can't contact node '%s'" % pnode)
6266
        if instance.name in ins_l.data:
6267
          raise errors.OpPrereqError("Instance is running, can't remove"
6268
                                     " disks.")
6269

    
6270
      if (disk_op == constants.DDM_ADD and
6271
          len(instance.nics) >= constants.MAX_DISKS):
6272
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
6273
                                   " add more" % constants.MAX_DISKS)
6274
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
6275
        # an existing disk
6276
        if disk_op < 0 or disk_op >= len(instance.disks):
6277
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
6278
                                     " are 0 to %d" %
6279
                                     (disk_op, len(instance.disks)))
6280

    
6281
    return
6282

    
6283
  def Exec(self, feedback_fn):
6284
    """Modifies an instance.
6285

6286
    All parameters take effect only at the next restart of the instance.
6287

6288
    """
6289
    # Process here the warnings from CheckPrereq, as we don't have a
6290
    # feedback_fn there.
6291
    for warn in self.warn:
6292
      feedback_fn("WARNING: %s" % warn)
6293

    
6294
    result = []
6295
    instance = self.instance
6296
    # disk changes
6297
    for disk_op, disk_dict in self.op.disks:
6298
      if disk_op == constants.DDM_REMOVE:
6299
        # remove the last disk
6300
        device = instance.disks.pop()
6301
        device_idx = len(instance.disks)
6302
        for node, disk in device.ComputeNodeTree(instance.primary_node):
6303
          self.cfg.SetDiskID(disk, node)
6304
          msg = self.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
6305
          if msg:
6306
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
6307
                            " continuing anyway", device_idx, node, msg)
6308
        result.append(("disk/%d" % device_idx, "remove"))
6309
      elif disk_op == constants.DDM_ADD:
6310
        # add a new disk
6311
        if instance.disk_template == constants.DT_FILE:
6312
          file_driver, file_path = instance.disks[0].logical_id
6313
          file_path = os.path.dirname(file_path)
6314
        else:
6315
          file_driver = file_path = None
6316
        disk_idx_base = len(instance.disks)
6317
        new_disk = _GenerateDiskTemplate(self,
6318
                                         instance.disk_template,
6319
                                         instance.name, instance.primary_node,
6320
                                         instance.secondary_nodes,
6321
                                         [disk_dict],
6322
                                         file_path,
6323
                                         file_driver,
6324
                                         disk_idx_base)[0]
6325
        instance.disks.append(new_disk)
6326
        info = _GetInstanceInfoText(instance)
6327

    
6328
        logging.info("Creating volume %s for instance %s",
6329
                     new_disk.iv_name, instance.name)
6330
        # Note: this needs to be kept in sync with _CreateDisks
6331
        #HARDCODE
6332
        for node in instance.all_nodes:
6333
          f_create = node == instance.primary_node
6334
          try:
6335
            _CreateBlockDev(self, node, instance, new_disk,
6336
                            f_create, info, f_create)
6337
          except errors.OpExecError, err:
6338
            self.LogWarning("Failed to create volume %s (%s) on"
6339
                            " node %s: %s",
6340
                            new_disk.iv_name, new_disk, node, err)
6341
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
6342
                       (new_disk.size, new_disk.mode)))
6343
      else:
6344
        # change a given disk
6345
        instance.disks[disk_op].mode = disk_dict['mode']
6346
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
6347
    # NIC changes
6348
    for nic_op, nic_dict in self.op.nics:
6349
      if nic_op == constants.DDM_REMOVE:
6350
        # remove the last nic
6351
        del instance.nics[-1]
6352
        result.append(("nic.%d" % len(instance.nics), "remove"))
6353
      elif nic_op == constants.DDM_ADD:
6354
        # mac and bridge should be set, by now
6355
        mac = nic_dict['mac']
6356
        bridge = nic_dict['bridge']
6357
        new_nic = objects.NIC(mac=mac, ip=nic_dict.get('ip', None),
6358
                              bridge=bridge)
6359
        instance.nics.append(new_nic)
6360
        result.append(("nic.%d" % (len(instance.nics) - 1),
6361
                       "add:mac=%s,ip=%s,bridge=%s" %
6362
                       (new_nic.mac, new_nic.ip, new_nic.bridge)))
6363
      else:
6364
        # change a given nic
6365
        for key in 'mac', 'ip', 'bridge':
6366
          if key in nic_dict:
6367
            setattr(instance.nics[nic_op], key, nic_dict[key])
6368
            result.append(("nic.%s/%d" % (key, nic_op), nic_dict[key]))
6369

    
6370
    # hvparams changes
6371
    if self.op.hvparams:
6372
      instance.hvparams = self.hv_inst
6373
      for key, val in self.op.hvparams.iteritems():
6374
        result.append(("hv/%s" % key, val))
6375

    
6376
    # beparams changes
6377
    if self.op.beparams:
6378
      instance.beparams = self.be_inst
6379
      for key, val in self.op.beparams.iteritems():
6380
        result.append(("be/%s" % key, val))
6381

    
6382
    self.cfg.Update(instance)
6383

    
6384
    return result
6385

    
6386

    
6387
class LUQueryExports(NoHooksLU):
6388
  """Query the exports list
6389

6390
  """
6391
  _OP_REQP = ['nodes']
6392
  REQ_BGL = False
6393

    
6394
  def ExpandNames(self):
6395
    self.needed_locks = {}
6396
    self.share_locks[locking.LEVEL_NODE] = 1
6397
    if not self.op.nodes:
6398
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6399
    else:
6400
      self.needed_locks[locking.LEVEL_NODE] = \
6401
        _GetWantedNodes(self, self.op.nodes)
6402

    
6403
  def CheckPrereq(self):
6404
    """Check prerequisites.
6405

6406
    """
6407
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
6408

    
6409
  def Exec(self, feedback_fn):
6410
    """Compute the list of all the exported system images.
6411

6412
    @rtype: dict
6413
    @return: a dictionary with the structure node->(export-list)
6414
        where export-list is a list of the instances exported on
6415
        that node.
6416

6417
    """
6418
    rpcresult = self.rpc.call_export_list(self.nodes)
6419
    result = {}
6420
    for node in rpcresult:
6421
      if rpcresult[node].failed:
6422
        result[node] = False
6423
      else:
6424
        result[node] = rpcresult[node].data
6425

    
6426
    return result
6427

    
6428

    
6429
class LUExportInstance(LogicalUnit):
6430
  """Export an instance to an image in the cluster.
6431

6432
  """
6433
  HPATH = "instance-export"
6434
  HTYPE = constants.HTYPE_INSTANCE
6435
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
6436
  REQ_BGL = False
6437

    
6438
  def ExpandNames(self):
6439
    self._ExpandAndLockInstance()
6440
    # FIXME: lock only instance primary and destination node
6441
    #
6442
    # Sad but true, for now we have do lock all nodes, as we don't know where
6443
    # the previous export might be, and and in this LU we search for it and
6444
    # remove it from its current node. In the future we could fix this by:
6445
    #  - making a tasklet to search (share-lock all), then create the new one,
6446
    #    then one to remove, after
6447
    #  - removing the removal operation altogether
6448
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6449

    
6450
  def DeclareLocks(self, level):
6451
    """Last minute lock declaration."""
6452
    # All nodes are locked anyway, so nothing to do here.
6453

    
6454
  def BuildHooksEnv(self):
6455
    """Build hooks env.
6456

6457
    This will run on the master, primary node and target node.
6458

6459
    """
6460
    env = {
6461
      "EXPORT_NODE": self.op.target_node,
6462
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
6463
      }
6464
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6465
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
6466
          self.op.target_node]
6467
    return env, nl, nl
6468

    
6469
  def CheckPrereq(self):
6470
    """Check prerequisites.
6471

6472
    This checks that the instance and node names are valid.
6473

6474
    """
6475
    instance_name = self.op.instance_name
6476
    self.instance = self.cfg.GetInstanceInfo(instance_name)
6477
    assert self.instance is not None, \
6478
          "Cannot retrieve locked instance %s" % self.op.instance_name
6479
    _CheckNodeOnline(self, self.instance.primary_node)
6480

    
6481
    self.dst_node = self.cfg.GetNodeInfo(
6482
      self.cfg.ExpandNodeName(self.op.target_node))
6483

    
6484
    if self.dst_node is None:
6485
      # This is wrong node name, not a non-locked node
6486
      raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
6487
    _CheckNodeOnline(self, self.dst_node.name)
6488
    _CheckNodeNotDrained(self, self.dst_node.name)
6489

    
6490
    # instance disk type verification
6491
    for disk in self.instance.disks:
6492
      if disk.dev_type == constants.LD_FILE:
6493
        raise errors.OpPrereqError("Export not supported for instances with"
6494
                                   " file-based disks")
6495

    
6496
  def Exec(self, feedback_fn):
6497
    """Export an instance to an image in the cluster.
6498

6499
    """
6500
    instance = self.instance
6501
    dst_node = self.dst_node
6502
    src_node = instance.primary_node
6503
    if self.op.shutdown:
6504
      # shutdown the instance, but not the disks
6505
      result = self.rpc.call_instance_shutdown(src_node, instance)
6506
      msg = result.RemoteFailMsg()
6507
      if msg:
6508
        raise errors.OpExecError("Could not shutdown instance %s on"
6509
                                 " node %s: %s" %
6510
                                 (instance.name, src_node, msg))
6511

    
6512
    vgname = self.cfg.GetVGName()
6513

    
6514
    snap_disks = []
6515

    
6516
    # set the disks ID correctly since call_instance_start needs the
6517
    # correct drbd minor to create the symlinks
6518
    for disk in instance.disks:
6519
      self.cfg.SetDiskID(disk, src_node)
6520

    
6521
    # per-disk results
6522
    dresults = []
6523
    try:
6524
      for idx, disk in enumerate(instance.disks):
6525
        # new_dev_name will be a snapshot of an lvm leaf of the one we passed
6526
        new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
6527
        if new_dev_name.failed or not new_dev_name.data:
6528
          self.LogWarning("Could not snapshot disk/%d on node %s",
6529
                          idx, src_node)
6530
          snap_disks.append(False)
6531
        else:
6532
          new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
6533
                                 logical_id=(vgname, new_dev_name.data),
6534
                                 physical_id=(vgname, new_dev_name.data),
6535
                                 iv_name=disk.iv_name)
6536
          snap_disks.append(new_dev)
6537

    
6538
    finally:
6539
      if self.op.shutdown and instance.admin_up:
6540
        result = self.rpc.call_instance_start(src_node, instance, None, None)
6541
        msg = result.RemoteFailMsg()
6542
        if msg:
6543
          _ShutdownInstanceDisks(self, instance)
6544
          raise errors.OpExecError("Could not start instance: %s" % msg)
6545

    
6546
    # TODO: check for size
6547

    
6548
    cluster_name = self.cfg.GetClusterName()
6549
    for idx, dev in enumerate(snap_disks):
6550
      if dev:
6551
        result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
6552
                                               instance, cluster_name, idx)
6553
        if result.failed or not result.data:
6554
          self.LogWarning("Could not export disk/%d from node %s to"
6555
                          " node %s", idx, src_node, dst_node.name)
6556
          dresults.append(False)
6557
        else:
6558
          dresults.append(True)
6559
        msg = self.rpc.call_blockdev_remove(src_node, dev).RemoteFailMsg()
6560
        if msg:
6561
          self.LogWarning("Could not remove snapshot for disk/%d from node"
6562
                          " %s: %s", idx, src_node, msg)
6563
      else:
6564
        dresults.append(False)
6565

    
6566
    result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
6567
    fin_resu = True
6568
    if result.failed or not result.data:
6569
      self.LogWarning("Could not finalize export for instance %s on node %s",
6570
                      instance.name, dst_node.name)
6571
      fin_resu = False
6572

    
6573
    nodelist = self.cfg.GetNodeList()
6574
    nodelist.remove(dst_node.name)
6575

    
6576
    # on one-node clusters nodelist will be empty after the removal
6577
    # if we proceed the backup would be removed because OpQueryExports
6578
    # substitutes an empty list with the full cluster node list.
6579
    if nodelist:
6580
      exportlist = self.rpc.call_export_list(nodelist)
6581
      for node in exportlist:
6582
        if exportlist[node].failed:
6583
          continue
6584
        if instance.name in exportlist[node].data:
6585
          if not self.rpc.call_export_remove(node, instance.name):
6586
            self.LogWarning("Could not remove older export for instance %s"
6587
                            " on node %s", instance.name, node)
6588
    return fin_resu, dresults
6589

    
6590

    
6591
class LURemoveExport(NoHooksLU):
6592
  """Remove exports related to the named instance.
6593

6594
  """
6595
  _OP_REQP = ["instance_name"]
6596
  REQ_BGL = False
6597

    
6598
  def ExpandNames(self):
6599
    self.needed_locks = {}
6600
    # We need all nodes to be locked in order for RemoveExport to work, but we
6601
    # don't need to lock the instance itself, as nothing will happen to it (and
6602
    # we can remove exports also for a removed instance)
6603
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6604

    
6605
  def CheckPrereq(self):
6606
    """Check prerequisites.
6607
    """
6608
    pass
6609

    
6610
  def Exec(self, feedback_fn):
6611
    """Remove any export.
6612

6613
    """
6614
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
6615
    # If the instance was not found we'll try with the name that was passed in.
6616
    # This will only work if it was an FQDN, though.
6617
    fqdn_warn = False
6618
    if not instance_name:
6619
      fqdn_warn = True
6620
      instance_name = self.op.instance_name
6621

    
6622
    exportlist = self.rpc.call_export_list(self.acquired_locks[
6623
      locking.LEVEL_NODE])
6624
    found = False
6625
    for node in exportlist:
6626
      if exportlist[node].failed:
6627
        self.LogWarning("Failed to query node %s, continuing" % node)
6628
        continue
6629
      if instance_name in exportlist[node].data:
6630
        found = True
6631
        result = self.rpc.call_export_remove(node, instance_name)
6632
        if result.failed or not result.data:
6633
          logging.error("Could not remove export for instance %s"
6634
                        " on node %s", instance_name, node)
6635

    
6636
    if fqdn_warn and not found:
6637
      feedback_fn("Export not found. If trying to remove an export belonging"
6638
                  " to a deleted instance please use its Fully Qualified"
6639
                  " Domain Name.")
6640

    
6641

    
6642
class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
6643
  """Generic tags LU.
6644

6645
  This is an abstract class which is the parent of all the other tags LUs.
6646

6647
  """
6648

    
6649
  def ExpandNames(self):
6650
    self.needed_locks = {}
6651
    if self.op.kind == constants.TAG_NODE:
6652
      name = self.cfg.ExpandNodeName(self.op.name)
6653
      if name is None:
6654
        raise errors.OpPrereqError("Invalid node name (%s)" %
6655
                                   (self.op.name,))
6656
      self.op.name = name
6657
      self.needed_locks[locking.LEVEL_NODE] = name
6658
    elif self.op.kind == constants.TAG_INSTANCE:
6659
      name = self.cfg.ExpandInstanceName(self.op.name)
6660
      if name is None:
6661
        raise errors.OpPrereqError("Invalid instance name (%s)" %
6662
                                   (self.op.name,))
6663
      self.op.name = name
6664
      self.needed_locks[locking.LEVEL_INSTANCE] = name
6665

    
6666
  def CheckPrereq(self):
6667
    """Check prerequisites.
6668

6669
    """
6670
    if self.op.kind == constants.TAG_CLUSTER:
6671
      self.target = self.cfg.GetClusterInfo()
6672
    elif self.op.kind == constants.TAG_NODE:
6673
      self.target = self.cfg.GetNodeInfo(self.op.name)
6674
    elif self.op.kind == constants.TAG_INSTANCE:
6675
      self.target = self.cfg.GetInstanceInfo(self.op.name)
6676
    else:
6677
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
6678
                                 str(self.op.kind))
6679

    
6680

    
6681
class LUGetTags(TagsLU):
6682
  """Returns the tags of a given object.
6683

6684
  """
6685
  _OP_REQP = ["kind", "name"]
6686
  REQ_BGL = False
6687

    
6688
  def Exec(self, feedback_fn):
6689
    """Returns the tag list.
6690

6691
    """
6692
    return list(self.target.GetTags())
6693

    
6694

    
6695
class LUSearchTags(NoHooksLU):
6696
  """Searches the tags for a given pattern.
6697

6698
  """
6699
  _OP_REQP = ["pattern"]
6700
  REQ_BGL = False
6701

    
6702
  def ExpandNames(self):
6703
    self.needed_locks = {}
6704

    
6705
  def CheckPrereq(self):
6706
    """Check prerequisites.
6707

6708
    This checks the pattern passed for validity by compiling it.
6709

6710
    """
6711
    try:
6712
      self.re = re.compile(self.op.pattern)
6713
    except re.error, err:
6714
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
6715
                                 (self.op.pattern, err))
6716

    
6717
  def Exec(self, feedback_fn):
6718
    """Returns the tag list.
6719

6720
    """
6721
    cfg = self.cfg
6722
    tgts = [("/cluster", cfg.GetClusterInfo())]
6723
    ilist = cfg.GetAllInstancesInfo().values()
6724
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
6725
    nlist = cfg.GetAllNodesInfo().values()
6726
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
6727
    results = []
6728
    for path, target in tgts:
6729
      for tag in target.GetTags():
6730
        if self.re.search(tag):
6731
          results.append((path, tag))
6732
    return results
6733

    
6734

    
6735
class LUAddTags(TagsLU):
6736
  """Sets a tag on a given object.
6737

6738
  """
6739
  _OP_REQP = ["kind", "name", "tags"]
6740
  REQ_BGL = False
6741

    
6742
  def CheckPrereq(self):
6743
    """Check prerequisites.
6744

6745
    This checks the type and length of the tag name and value.
6746

6747
    """
6748
    TagsLU.CheckPrereq(self)
6749
    for tag in self.op.tags:
6750
      objects.TaggableObject.ValidateTag(tag)
6751

    
6752
  def Exec(self, feedback_fn):
6753
    """Sets the tag.
6754

6755
    """
6756
    try:
6757
      for tag in self.op.tags:
6758
        self.target.AddTag(tag)
6759
    except errors.TagError, err:
6760
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
6761
    try:
6762
      self.cfg.Update(self.target)
6763
    except errors.ConfigurationError:
6764
      raise errors.OpRetryError("There has been a modification to the"
6765
                                " config file and the operation has been"
6766
                                " aborted. Please retry.")
6767

    
6768

    
6769
class LUDelTags(TagsLU):
6770
  """Delete a list of tags from a given object.
6771

6772
  """
6773
  _OP_REQP = ["kind", "name", "tags"]
6774
  REQ_BGL = False
6775

    
6776
  def CheckPrereq(self):
6777
    """Check prerequisites.
6778

6779
    This checks that we have the given tag.
6780

6781
    """
6782
    TagsLU.CheckPrereq(self)
6783
    for tag in self.op.tags:
6784
      objects.TaggableObject.ValidateTag(tag)
6785
    del_tags = frozenset(self.op.tags)
6786
    cur_tags = self.target.GetTags()
6787
    if not del_tags <= cur_tags:
6788
      diff_tags = del_tags - cur_tags
6789
      diff_names = ["'%s'" % tag for tag in diff_tags]
6790
      diff_names.sort()
6791
      raise errors.OpPrereqError("Tag(s) %s not found" %
6792
                                 (",".join(diff_names)))
6793

    
6794
  def Exec(self, feedback_fn):
6795
    """Remove the tag from the object.
6796

6797
    """
6798
    for tag in self.op.tags:
6799
      self.target.RemoveTag(tag)
6800
    try:
6801
      self.cfg.Update(self.target)
6802
    except errors.ConfigurationError:
6803
      raise errors.OpRetryError("There has been a modification to the"
6804
                                " config file and the operation has been"
6805
                                " aborted. Please retry.")
6806

    
6807

    
6808
class LUTestDelay(NoHooksLU):
6809
  """Sleep for a specified amount of time.
6810

6811
  This LU sleeps on the master and/or nodes for a specified amount of
6812
  time.
6813

6814
  """
6815
  _OP_REQP = ["duration", "on_master", "on_nodes"]
6816
  REQ_BGL = False
6817

    
6818
  def ExpandNames(self):
6819
    """Expand names and set required locks.
6820

6821
    This expands the node list, if any.
6822

6823
    """
6824
    self.needed_locks = {}
6825
    if self.op.on_nodes:
6826
      # _GetWantedNodes can be used here, but is not always appropriate to use
6827
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
6828
      # more information.
6829
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
6830
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
6831

    
6832
  def CheckPrereq(self):
6833
    """Check prerequisites.
6834

6835
    """
6836

    
6837
  def Exec(self, feedback_fn):
6838
    """Do the actual sleep.
6839

6840
    """
6841
    if self.op.on_master:
6842
      if not utils.TestDelay(self.op.duration):
6843
        raise errors.OpExecError("Error during master delay test")
6844
    if self.op.on_nodes:
6845
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
6846
      if not result:
6847
        raise errors.OpExecError("Complete failure from rpc call")
6848
      for node, node_result in result.items():
6849
        node_result.Raise()
6850
        if not node_result.data:
6851
          raise errors.OpExecError("Failure during rpc call to node %s,"
6852
                                   " result: %s" % (node, node_result.data))
6853

    
6854

    
6855
class IAllocator(object):
6856
  """IAllocator framework.
6857

6858
  An IAllocator instance has three sets of attributes:
6859
    - cfg that is needed to query the cluster
6860
    - input data (all members of the _KEYS class attribute are required)
6861
    - four buffer attributes (in|out_data|text), that represent the
6862
      input (to the external script) in text and data structure format,
6863
      and the output from it, again in two formats
6864
    - the result variables from the script (success, info, nodes) for
6865
      easy usage
6866

6867
  """
6868
  _ALLO_KEYS = [
6869
    "mem_size", "disks", "disk_template",
6870
    "os", "tags", "nics", "vcpus", "hypervisor",
6871
    ]
6872
  _RELO_KEYS = [
6873
    "relocate_from",
6874
    ]
6875

    
6876
  def __init__(self, lu, mode, name, **kwargs):
6877
    self.lu = lu
6878
    # init buffer variables
6879
    self.in_text = self.out_text = self.in_data = self.out_data = None
6880
    # init all input fields so that pylint is happy
6881
    self.mode = mode
6882
    self.name = name
6883
    self.mem_size = self.disks = self.disk_template = None
6884
    self.os = self.tags = self.nics = self.vcpus = None
6885
    self.hypervisor = None
6886
    self.relocate_from = None
6887
    # computed fields
6888
    self.required_nodes = None
6889
    # init result fields
6890
    self.success = self.info = self.nodes = None
6891
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6892
      keyset = self._ALLO_KEYS
6893
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6894
      keyset = self._RELO_KEYS
6895
    else:
6896
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
6897
                                   " IAllocator" % self.mode)
6898
    for key in kwargs:
6899
      if key not in keyset:
6900
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
6901
                                     " IAllocator" % key)
6902
      setattr(self, key, kwargs[key])
6903
    for key in keyset:
6904
      if key not in kwargs:
6905
        raise errors.ProgrammerError("Missing input parameter '%s' to"
6906
                                     " IAllocator" % key)
6907
    self._BuildInputData()
6908

    
6909
  def _ComputeClusterData(self):
6910
    """Compute the generic allocator input data.
6911

6912
    This is the data that is independent of the actual operation.
6913

6914
    """
6915
    cfg = self.lu.cfg
6916
    cluster_info = cfg.GetClusterInfo()
6917
    # cluster data
6918
    data = {
6919
      "version": constants.IALLOCATOR_VERSION,
6920
      "cluster_name": cfg.GetClusterName(),
6921
      "cluster_tags": list(cluster_info.GetTags()),
6922
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
6923
      # we don't have job IDs
6924
      }
6925
    iinfo = cfg.GetAllInstancesInfo().values()
6926
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
6927

    
6928
    # node data
6929
    node_results = {}
6930
    node_list = cfg.GetNodeList()
6931

    
6932
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6933
      hypervisor_name = self.hypervisor
6934
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6935
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
6936

    
6937
    node_data = self.lu.rpc.call_node_info(node_list, cfg.GetVGName(),
6938
                                           hypervisor_name)
6939
    node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
6940
                       cluster_info.enabled_hypervisors)
6941
    for nname, nresult in node_data.items():
6942
      # first fill in static (config-based) values
6943
      ninfo = cfg.GetNodeInfo(nname)
6944
      pnr = {
6945
        "tags": list(ninfo.GetTags()),
6946
        "primary_ip": ninfo.primary_ip,
6947
        "secondary_ip": ninfo.secondary_ip,
6948
        "offline": ninfo.offline,
6949
        "drained": ninfo.drained,
6950
        "master_candidate": ninfo.master_candidate,
6951
        }
6952

    
6953
      if not (ninfo.offline or ninfo.drained):
6954
        nresult.Raise()
6955
        if not isinstance(nresult.data, dict):
6956
          raise errors.OpExecError("Can't get data for node %s" % nname)
6957
        remote_info = nresult.data
6958
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
6959
                     'vg_size', 'vg_free', 'cpu_total']:
6960
          if attr not in remote_info:
6961
            raise errors.OpExecError("Node '%s' didn't return attribute"
6962
                                     " '%s'" % (nname, attr))
6963
          try:
6964
            remote_info[attr] = int(remote_info[attr])
6965
          except ValueError, err:
6966
            raise errors.OpExecError("Node '%s' returned invalid value"
6967
                                     " for '%s': %s" % (nname, attr, err))
6968
        # compute memory used by primary instances
6969
        i_p_mem = i_p_up_mem = 0
6970
        for iinfo, beinfo in i_list:
6971
          if iinfo.primary_node == nname:
6972
            i_p_mem += beinfo[constants.BE_MEMORY]
6973
            if iinfo.name not in node_iinfo[nname].data:
6974
              i_used_mem = 0
6975
            else:
6976
              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
6977
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
6978
            remote_info['memory_free'] -= max(0, i_mem_diff)
6979

    
6980
            if iinfo.admin_up:
6981
              i_p_up_mem += beinfo[constants.BE_MEMORY]
6982

    
6983
        # compute memory used by instances
6984
        pnr_dyn = {
6985
          "total_memory": remote_info['memory_total'],
6986
          "reserved_memory": remote_info['memory_dom0'],
6987
          "free_memory": remote_info['memory_free'],
6988
          "total_disk": remote_info['vg_size'],
6989
          "free_disk": remote_info['vg_free'],
6990
          "total_cpus": remote_info['cpu_total'],
6991
          "i_pri_memory": i_p_mem,
6992
          "i_pri_up_memory": i_p_up_mem,
6993
          }
6994
        pnr.update(pnr_dyn)
6995

    
6996
      node_results[nname] = pnr
6997
    data["nodes"] = node_results
6998

    
6999
    # instance data
7000
    instance_data = {}
7001
    for iinfo, beinfo in i_list:
7002
      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
7003
                  for n in iinfo.nics]
7004
      pir = {
7005
        "tags": list(iinfo.GetTags()),
7006
        "admin_up": iinfo.admin_up,
7007
        "vcpus": beinfo[constants.BE_VCPUS],
7008
        "memory": beinfo[constants.BE_MEMORY],
7009
        "os": iinfo.os,
7010
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
7011
        "nics": nic_data,
7012
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
7013
        "disk_template": iinfo.disk_template,
7014
        "hypervisor": iinfo.hypervisor,
7015
        }
7016
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
7017
                                                 pir["disks"])
7018
      instance_data[iinfo.name] = pir
7019

    
7020
    data["instances"] = instance_data
7021

    
7022
    self.in_data = data
7023

    
7024
  def _AddNewInstance(self):
7025
    """Add new instance data to allocator structure.
7026

7027
    This in combination with _AllocatorGetClusterData will create the
7028
    correct structure needed as input for the allocator.
7029

7030
    The checks for the completeness of the opcode must have already been
7031
    done.
7032

7033
    """
7034
    data = self.in_data
7035

    
7036
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
7037

    
7038
    if self.disk_template in constants.DTS_NET_MIRROR:
7039
      self.required_nodes = 2
7040
    else:
7041
      self.required_nodes = 1
7042
    request = {
7043
      "type": "allocate",
7044
      "name": self.name,
7045
      "disk_template": self.disk_template,
7046
      "tags": self.tags,
7047
      "os": self.os,
7048
      "vcpus": self.vcpus,
7049
      "memory": self.mem_size,
7050
      "disks": self.disks,
7051
      "disk_space_total": disk_space,
7052
      "nics": self.nics,
7053
      "required_nodes": self.required_nodes,
7054
      }
7055
    data["request"] = request
7056

    
7057
  def _AddRelocateInstance(self):
7058
    """Add relocate instance data to allocator structure.
7059

7060
    This in combination with _IAllocatorGetClusterData will create the
7061
    correct structure needed as input for the allocator.
7062

7063
    The checks for the completeness of the opcode must have already been
7064
    done.
7065

7066
    """
7067
    instance = self.lu.cfg.GetInstanceInfo(self.name)
7068
    if instance is None:
7069
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
7070
                                   " IAllocator" % self.name)
7071

    
7072
    if instance.disk_template not in constants.DTS_NET_MIRROR:
7073
      raise errors.OpPrereqError("Can't relocate non-mirrored instances")
7074

    
7075
    if len(instance.secondary_nodes) != 1:
7076
      raise errors.OpPrereqError("Instance has not exactly one secondary node")
7077

    
7078
    self.required_nodes = 1
7079
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
7080
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
7081

    
7082
    request = {
7083
      "type": "relocate",
7084
      "name": self.name,
7085
      "disk_space_total": disk_space,
7086
      "required_nodes": self.required_nodes,
7087
      "relocate_from": self.relocate_from,
7088
      }
7089
    self.in_data["request"] = request
7090

    
7091
  def _BuildInputData(self):
7092
    """Build input data structures.
7093

7094
    """
7095
    self._ComputeClusterData()
7096

    
7097
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
7098
      self._AddNewInstance()
7099
    else:
7100
      self._AddRelocateInstance()
7101

    
7102
    self.in_text = serializer.Dump(self.in_data)
7103

    
7104
  def Run(self, name, validate=True, call_fn=None):
7105
    """Run an instance allocator and return the results.
7106

7107
    """
7108
    if call_fn is None:
7109
      call_fn = self.lu.rpc.call_iallocator_runner
7110

    
7111
    result = call_fn(self.lu.cfg.GetMasterNode(), name, self.in_text)
7112
    result.Raise()
7113

    
7114
    if not isinstance(result.data, (list, tuple)) or len(result.data) != 4:
7115
      raise errors.OpExecError("Invalid result from master iallocator runner")
7116

    
7117
    rcode, stdout, stderr, fail = result.data
7118

    
7119
    if rcode == constants.IARUN_NOTFOUND:
7120
      raise errors.OpExecError("Can't find allocator '%s'" % name)
7121
    elif rcode == constants.IARUN_FAILURE:
7122
      raise errors.OpExecError("Instance allocator call failed: %s,"
7123
                               " output: %s" % (fail, stdout+stderr))
7124
    self.out_text = stdout
7125
    if validate:
7126
      self._ValidateResult()
7127

    
7128
  def _ValidateResult(self):
7129
    """Process the allocator results.
7130

7131
    This will process and if successful save the result in
7132
    self.out_data and the other parameters.
7133

7134
    """
7135
    try:
7136
      rdict = serializer.Load(self.out_text)
7137
    except Exception, err:
7138
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
7139

    
7140
    if not isinstance(rdict, dict):
7141
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
7142

    
7143
    for key in "success", "info", "nodes":
7144
      if key not in rdict:
7145
        raise errors.OpExecError("Can't parse iallocator results:"
7146
                                 " missing key '%s'" % key)
7147
      setattr(self, key, rdict[key])
7148

    
7149
    if not isinstance(rdict["nodes"], list):
7150
      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
7151
                               " is not a list")
7152
    self.out_data = rdict
7153

    
7154

    
7155
class LUTestAllocator(NoHooksLU):
7156
  """Run allocator tests.
7157

7158
  This LU runs the allocator tests
7159

7160
  """
7161
  _OP_REQP = ["direction", "mode", "name"]
7162

    
7163
  def CheckPrereq(self):
7164
    """Check prerequisites.
7165

7166
    This checks the opcode parameters depending on the director and mode test.
7167

7168
    """
7169
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
7170
      for attr in ["name", "mem_size", "disks", "disk_template",
7171
                   "os", "tags", "nics", "vcpus"]:
7172
        if not hasattr(self.op, attr):
7173
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
7174
                                     attr)
7175
      iname = self.cfg.ExpandInstanceName(self.op.name)
7176
      if iname is not None:
7177
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
7178
                                   iname)
7179
      if not isinstance(self.op.nics, list):
7180
        raise errors.OpPrereqError("Invalid parameter 'nics'")
7181
      for row in self.op.nics:
7182
        if (not isinstance(row, dict) or
7183
            "mac" not in row or
7184
            "ip" not in row or
7185
            "bridge" not in row):
7186
          raise errors.OpPrereqError("Invalid contents of the"
7187
                                     " 'nics' parameter")
7188
      if not isinstance(self.op.disks, list):
7189
        raise errors.OpPrereqError("Invalid parameter 'disks'")
7190
      for row in self.op.disks:
7191
        if (not isinstance(row, dict) or
7192
            "size" not in row or
7193
            not isinstance(row["size"], int) or
7194
            "mode" not in row or
7195
            row["mode"] not in ['r', 'w']):
7196
          raise errors.OpPrereqError("Invalid contents of the"
7197
                                     " 'disks' parameter")
7198
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
7199
        self.op.hypervisor = self.cfg.GetHypervisorType()
7200
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
7201
      if not hasattr(self.op, "name"):
7202
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
7203
      fname = self.cfg.ExpandInstanceName(self.op.name)
7204
      if fname is None:
7205
        raise errors.OpPrereqError("Instance '%s' not found for relocation" %
7206
                                   self.op.name)
7207
      self.op.name = fname
7208
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
7209
    else:
7210
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
7211
                                 self.op.mode)
7212

    
7213
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
7214
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
7215
        raise errors.OpPrereqError("Missing allocator name")
7216
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
7217
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
7218
                                 self.op.direction)
7219

    
7220
  def Exec(self, feedback_fn):
7221
    """Run the allocator test.
7222

7223
    """
7224
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
7225
      ial = IAllocator(self,
7226
                       mode=self.op.mode,
7227
                       name=self.op.name,
7228
                       mem_size=self.op.mem_size,
7229
                       disks=self.op.disks,
7230
                       disk_template=self.op.disk_template,
7231
                       os=self.op.os,
7232
                       tags=self.op.tags,
7233
                       nics=self.op.nics,
7234
                       vcpus=self.op.vcpus,
7235
                       hypervisor=self.op.hypervisor,
7236
                       )
7237
    else:
7238
      ial = IAllocator(self,
7239
                       mode=self.op.mode,
7240
                       name=self.op.name,
7241
                       relocate_from=list(self.relocate_from),
7242
                       )
7243

    
7244
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
7245
      result = ial.in_text
7246
    else:
7247
      ial.Run(self.op.allocator, validate=False)
7248
      result = ial.out_text
7249
    return result