Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 1b54fc6c

History | View | Annotate | Download (245.3 kB)

1
#
2
#
3

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

    
21

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

    
24
# pylint: disable-msg=W0613,W0201
25

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

    
36
from ganeti import ssh
37
from ganeti import utils
38
from ganeti import errors
39
from ganeti import hypervisor
40
from ganeti import locking
41
from ganeti import constants
42
from ganeti import objects
43
from ganeti import opcodes
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 overriden 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
91
    self.LogInfo = processor.LogInfo
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 aquired 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):
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
  """Builds instance related env variables for hooks
458

459
  This builds the hook environment from individual variables.
460

461
  @type name: string
462
  @param name: the name of the instance
463
  @type primary_node: string
464
  @param primary_node: the name of the instance's primary node
465
  @type secondary_nodes: list
466
  @param secondary_nodes: list of secondary nodes as strings
467
  @type os_type: string
468
  @param os_type: the name of the instance's OS
469
  @type status: boolean
470
  @param status: the should_run status of the instance
471
  @type memory: string
472
  @param memory: the memory size of the instance
473
  @type vcpus: string
474
  @param vcpus: the count of VCPUs the instance has
475
  @type nics: list
476
  @param nics: list of tuples (ip, bridge, mac) representing
477
      the NICs the instance  has
478
  @type disk_template: string
479
  @param disk_template: the distk template of the instance
480
  @type disks: list
481
  @param disks: the list of (size, mode) pairs
482
  @rtype: dict
483
  @return: the hook environment for this instance
484

485
  """
486
  if status:
487
    str_status = "up"
488
  else:
489
    str_status = "down"
490
  env = {
491
    "OP_TARGET": name,
492
    "INSTANCE_NAME": name,
493
    "INSTANCE_PRIMARY": primary_node,
494
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
495
    "INSTANCE_OS_TYPE": os_type,
496
    "INSTANCE_STATUS": str_status,
497
    "INSTANCE_MEMORY": memory,
498
    "INSTANCE_VCPUS": vcpus,
499
    "INSTANCE_DISK_TEMPLATE": disk_template,
500
  }
501

    
502
  if nics:
503
    nic_count = len(nics)
504
    for idx, (ip, bridge, mac) in enumerate(nics):
505
      if ip is None:
506
        ip = ""
507
      env["INSTANCE_NIC%d_IP" % idx] = ip
508
      env["INSTANCE_NIC%d_BRIDGE" % idx] = bridge
509
      env["INSTANCE_NIC%d_MAC" % idx] = mac
510
  else:
511
    nic_count = 0
512

    
513
  env["INSTANCE_NIC_COUNT"] = nic_count
514

    
515
  if disks:
516
    disk_count = len(disks)
517
    for idx, (size, mode) in enumerate(disks):
518
      env["INSTANCE_DISK%d_SIZE" % idx] = size
519
      env["INSTANCE_DISK%d_MODE" % idx] = mode
520
  else:
521
    disk_count = 0
522

    
523
  env["INSTANCE_DISK_COUNT"] = disk_count
524

    
525
  return env
526

    
527

    
528
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
529
  """Builds instance related env variables for hooks from an object.
530

531
  @type lu: L{LogicalUnit}
532
  @param lu: the logical unit on whose behalf we execute
533
  @type instance: L{objects.Instance}
534
  @param instance: the instance for which we should build the
535
      environment
536
  @type override: dict
537
  @param override: dictionary with key/values that will override
538
      our values
539
  @rtype: dict
540
  @return: the hook environment dictionary
541

542
  """
543
  bep = lu.cfg.GetClusterInfo().FillBE(instance)
544
  args = {
545
    'name': instance.name,
546
    'primary_node': instance.primary_node,
547
    'secondary_nodes': instance.secondary_nodes,
548
    'os_type': instance.os,
549
    'status': instance.admin_up,
550
    'memory': bep[constants.BE_MEMORY],
551
    'vcpus': bep[constants.BE_VCPUS],
552
    'nics': [(nic.ip, nic.bridge, nic.mac) for nic in instance.nics],
553
    'disk_template': instance.disk_template,
554
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
555
  }
556
  if override:
557
    args.update(override)
558
  return _BuildInstanceHookEnv(**args)
559

    
560

    
561
def _AdjustCandidatePool(lu):
562
  """Adjust the candidate pool after node operations.
563

564
  """
565
  mod_list = lu.cfg.MaintainCandidatePool()
566
  if mod_list:
567
    lu.LogInfo("Promoted nodes to master candidate role: %s",
568
               ", ".join(node.name for node in mod_list))
569
    for name in mod_list:
570
      lu.context.ReaddNode(name)
571
  mc_now, mc_max = lu.cfg.GetMasterCandidateStats()
572
  if mc_now > mc_max:
573
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
574
               (mc_now, mc_max))
575

    
576

    
577
def _CheckInstanceBridgesExist(lu, instance):
578
  """Check that the brigdes needed by an instance exist.
579

580
  """
581
  # check bridges existance
582
  brlist = [nic.bridge for nic in instance.nics]
583
  result = lu.rpc.call_bridges_exist(instance.primary_node, brlist)
584
  result.Raise()
585
  if not result.data:
586
    raise errors.OpPrereqError("One or more target bridges %s does not"
587
                               " exist on destination node '%s'" %
588
                               (brlist, instance.primary_node))
589

    
590

    
591
class LUDestroyCluster(NoHooksLU):
592
  """Logical unit for destroying the cluster.
593

594
  """
595
  _OP_REQP = []
596

    
597
  def CheckPrereq(self):
598
    """Check prerequisites.
599

600
    This checks whether the cluster is empty.
601

602
    Any errors are signalled by raising errors.OpPrereqError.
603

604
    """
605
    master = self.cfg.GetMasterNode()
606

    
607
    nodelist = self.cfg.GetNodeList()
608
    if len(nodelist) != 1 or nodelist[0] != master:
609
      raise errors.OpPrereqError("There are still %d node(s) in"
610
                                 " this cluster." % (len(nodelist) - 1))
611
    instancelist = self.cfg.GetInstanceList()
612
    if instancelist:
613
      raise errors.OpPrereqError("There are still %d instance(s) in"
614
                                 " this cluster." % len(instancelist))
615

    
616
  def Exec(self, feedback_fn):
617
    """Destroys the cluster.
618

619
    """
620
    master = self.cfg.GetMasterNode()
621
    result = self.rpc.call_node_stop_master(master, False)
622
    result.Raise()
623
    if not result.data:
624
      raise errors.OpExecError("Could not disable the master role")
625
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
626
    utils.CreateBackup(priv_key)
627
    utils.CreateBackup(pub_key)
628
    return master
629

    
630

    
631
class LUVerifyCluster(LogicalUnit):
632
  """Verifies the cluster status.
633

634
  """
635
  HPATH = "cluster-verify"
636
  HTYPE = constants.HTYPE_CLUSTER
637
  _OP_REQP = ["skip_checks"]
638
  REQ_BGL = False
639

    
640
  def ExpandNames(self):
641
    self.needed_locks = {
642
      locking.LEVEL_NODE: locking.ALL_SET,
643
      locking.LEVEL_INSTANCE: locking.ALL_SET,
644
    }
645
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
646

    
647
  def _VerifyNode(self, nodeinfo, file_list, local_cksum,
648
                  node_result, feedback_fn, master_files,
649
                  drbd_map, vg_name):
650
    """Run multiple tests against a node.
651

652
    Test list:
653

654
      - compares ganeti version
655
      - checks vg existance and size > 20G
656
      - checks config file checksum
657
      - checks ssh to other nodes
658

659
    @type nodeinfo: L{objects.Node}
660
    @param nodeinfo: the node to check
661
    @param file_list: required list of files
662
    @param local_cksum: dictionary of local files and their checksums
663
    @param node_result: the results from the node
664
    @param feedback_fn: function used to accumulate results
665
    @param master_files: list of files that only masters should have
666
    @param drbd_map: the useddrbd minors for this node, in
667
        form of minor: (instance, must_exist) which correspond to instances
668
        and their running status
669
    @param vg_name: Ganeti Volume Group (result of self.cfg.GetVGName())
670

671
    """
672
    node = nodeinfo.name
673

    
674
    # main result, node_result should be a non-empty dict
675
    if not node_result or not isinstance(node_result, dict):
676
      feedback_fn("  - ERROR: unable to verify node %s." % (node,))
677
      return True
678

    
679
    # compares ganeti version
680
    local_version = constants.PROTOCOL_VERSION
681
    remote_version = node_result.get('version', None)
682
    if not (remote_version and isinstance(remote_version, (list, tuple)) and
683
            len(remote_version) == 2):
684
      feedback_fn("  - ERROR: connection to %s failed" % (node))
685
      return True
686

    
687
    if local_version != remote_version[0]:
688
      feedback_fn("  - ERROR: incompatible protocol versions: master %s,"
689
                  " node %s %s" % (local_version, node, remote_version[0]))
690
      return True
691

    
692
    # node seems compatible, we can actually try to look into its results
693

    
694
    bad = False
695

    
696
    # full package version
697
    if constants.RELEASE_VERSION != remote_version[1]:
698
      feedback_fn("  - WARNING: software version mismatch: master %s,"
699
                  " node %s %s" %
700
                  (constants.RELEASE_VERSION, node, remote_version[1]))
701

    
702
    # checks vg existence and size > 20G
703
    if vg_name is not None:
704
      vglist = node_result.get(constants.NV_VGLIST, None)
705
      if not vglist:
706
        feedback_fn("  - ERROR: unable to check volume groups on node %s." %
707
                        (node,))
708
        bad = True
709
      else:
710
        vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
711
                                              constants.MIN_VG_SIZE)
712
        if vgstatus:
713
          feedback_fn("  - ERROR: %s on node %s" % (vgstatus, node))
714
          bad = True
715

    
716
    # checks config file checksum
717

    
718
    remote_cksum = node_result.get(constants.NV_FILELIST, None)
719
    if not isinstance(remote_cksum, dict):
720
      bad = True
721
      feedback_fn("  - ERROR: node hasn't returned file checksum data")
722
    else:
723
      for file_name in file_list:
724
        node_is_mc = nodeinfo.master_candidate
725
        must_have_file = file_name not in master_files
726
        if file_name not in remote_cksum:
727
          if node_is_mc or must_have_file:
728
            bad = True
729
            feedback_fn("  - ERROR: file '%s' missing" % file_name)
730
        elif remote_cksum[file_name] != local_cksum[file_name]:
731
          if node_is_mc or must_have_file:
732
            bad = True
733
            feedback_fn("  - ERROR: file '%s' has wrong checksum" % file_name)
734
          else:
735
            # not candidate and this is not a must-have file
736
            bad = True
737
            feedback_fn("  - ERROR: non master-candidate has old/wrong file"
738
                        " '%s'" % file_name)
739
        else:
740
          # all good, except non-master/non-must have combination
741
          if not node_is_mc and not must_have_file:
742
            feedback_fn("  - ERROR: file '%s' should not exist on non master"
743
                        " candidates" % file_name)
744

    
745
    # checks ssh to any
746

    
747
    if constants.NV_NODELIST not in node_result:
748
      bad = True
749
      feedback_fn("  - ERROR: node hasn't returned node ssh connectivity data")
750
    else:
751
      if node_result[constants.NV_NODELIST]:
752
        bad = True
753
        for node in node_result[constants.NV_NODELIST]:
754
          feedback_fn("  - ERROR: ssh communication with node '%s': %s" %
755
                          (node, node_result[constants.NV_NODELIST][node]))
756

    
757
    if constants.NV_NODENETTEST not in node_result:
758
      bad = True
759
      feedback_fn("  - ERROR: node hasn't returned node tcp connectivity data")
760
    else:
761
      if node_result[constants.NV_NODENETTEST]:
762
        bad = True
763
        nlist = utils.NiceSort(node_result[constants.NV_NODENETTEST].keys())
764
        for node in nlist:
765
          feedback_fn("  - ERROR: tcp communication with node '%s': %s" %
766
                          (node, node_result[constants.NV_NODENETTEST][node]))
767

    
768
    hyp_result = node_result.get(constants.NV_HYPERVISOR, None)
769
    if isinstance(hyp_result, dict):
770
      for hv_name, hv_result in hyp_result.iteritems():
771
        if hv_result is not None:
772
          feedback_fn("  - ERROR: hypervisor %s verify failure: '%s'" %
773
                      (hv_name, hv_result))
774

    
775
    # check used drbd list
776
    if vg_name is not None:
777
      used_minors = node_result.get(constants.NV_DRBDLIST, [])
778
      if not isinstance(used_minors, (tuple, list)):
779
        feedback_fn("  - ERROR: cannot parse drbd status file: %s" %
780
                    str(used_minors))
781
      else:
782
        for minor, (iname, must_exist) in drbd_map.items():
783
          if minor not in used_minors and must_exist:
784
            feedback_fn("  - ERROR: drbd minor %d of instance %s is"
785
                        " not active" % (minor, iname))
786
            bad = True
787
        for minor in used_minors:
788
          if minor not in drbd_map:
789
            feedback_fn("  - ERROR: unallocated drbd minor %d is in use" %
790
                        minor)
791
            bad = True
792

    
793
    return bad
794

    
795
  def _VerifyInstance(self, instance, instanceconfig, node_vol_is,
796
                      node_instance, feedback_fn, n_offline):
797
    """Verify an instance.
798

799
    This function checks to see if the required block devices are
800
    available on the instance's node.
801

802
    """
803
    bad = False
804

    
805
    node_current = instanceconfig.primary_node
806

    
807
    node_vol_should = {}
808
    instanceconfig.MapLVsByNode(node_vol_should)
809

    
810
    for node in node_vol_should:
811
      if node in n_offline:
812
        # ignore missing volumes on offline nodes
813
        continue
814
      for volume in node_vol_should[node]:
815
        if node not in node_vol_is or volume not in node_vol_is[node]:
816
          feedback_fn("  - ERROR: volume %s missing on node %s" %
817
                          (volume, node))
818
          bad = True
819

    
820
    if instanceconfig.admin_up:
821
      if ((node_current not in node_instance or
822
          not instance in node_instance[node_current]) and
823
          node_current not in n_offline):
824
        feedback_fn("  - ERROR: instance %s not running on node %s" %
825
                        (instance, node_current))
826
        bad = True
827

    
828
    for node in node_instance:
829
      if (not node == node_current):
830
        if instance in node_instance[node]:
831
          feedback_fn("  - ERROR: instance %s should not run on node %s" %
832
                          (instance, node))
833
          bad = True
834

    
835
    return bad
836

    
837
  def _VerifyOrphanVolumes(self, node_vol_should, node_vol_is, feedback_fn):
838
    """Verify if there are any unknown volumes in the cluster.
839

840
    The .os, .swap and backup volumes are ignored. All other volumes are
841
    reported as unknown.
842

843
    """
844
    bad = False
845

    
846
    for node in node_vol_is:
847
      for volume in node_vol_is[node]:
848
        if node not in node_vol_should or volume not in node_vol_should[node]:
849
          feedback_fn("  - ERROR: volume %s on node %s should not exist" %
850
                      (volume, node))
851
          bad = True
852
    return bad
853

    
854
  def _VerifyOrphanInstances(self, instancelist, node_instance, feedback_fn):
855
    """Verify the list of running instances.
856

857
    This checks what instances are running but unknown to the cluster.
858

859
    """
860
    bad = False
861
    for node in node_instance:
862
      for runninginstance in node_instance[node]:
863
        if runninginstance not in instancelist:
864
          feedback_fn("  - ERROR: instance %s on node %s should not exist" %
865
                          (runninginstance, node))
866
          bad = True
867
    return bad
868

    
869
  def _VerifyNPlusOneMemory(self, node_info, instance_cfg, feedback_fn):
870
    """Verify N+1 Memory Resilience.
871

872
    Check that if one single node dies we can still start all the instances it
873
    was primary for.
874

875
    """
876
    bad = False
877

    
878
    for node, nodeinfo in node_info.iteritems():
879
      # This code checks that every node which is now listed as secondary has
880
      # enough memory to host all instances it is supposed to should a single
881
      # other node in the cluster fail.
882
      # FIXME: not ready for failover to an arbitrary node
883
      # FIXME: does not support file-backed instances
884
      # WARNING: we currently take into account down instances as well as up
885
      # ones, considering that even if they're down someone might want to start
886
      # them even in the event of a node failure.
887
      for prinode, instances in nodeinfo['sinst-by-pnode'].iteritems():
888
        needed_mem = 0
889
        for instance in instances:
890
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
891
          if bep[constants.BE_AUTO_BALANCE]:
892
            needed_mem += bep[constants.BE_MEMORY]
893
        if nodeinfo['mfree'] < needed_mem:
894
          feedback_fn("  - ERROR: not enough memory on node %s to accomodate"
895
                      " failovers should node %s fail" % (node, prinode))
896
          bad = True
897
    return bad
898

    
899
  def CheckPrereq(self):
900
    """Check prerequisites.
901

902
    Transform the list of checks we're going to skip into a set and check that
903
    all its members are valid.
904

905
    """
906
    self.skip_set = frozenset(self.op.skip_checks)
907
    if not constants.VERIFY_OPTIONAL_CHECKS.issuperset(self.skip_set):
908
      raise errors.OpPrereqError("Invalid checks to be skipped specified")
909

    
910
  def BuildHooksEnv(self):
911
    """Build hooks env.
912

913
    Cluster-Verify hooks just rone in the post phase and their failure makes
914
    the output be logged in the verify output and the verification to fail.
915

916
    """
917
    all_nodes = self.cfg.GetNodeList()
918
    env = {
919
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
920
      }
921
    for node in self.cfg.GetAllNodesInfo().values():
922
      env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())
923

    
924
    return env, [], all_nodes
925

    
926
  def Exec(self, feedback_fn):
927
    """Verify integrity of cluster, performing various test on nodes.
928

929
    """
930
    bad = False
931
    feedback_fn("* Verifying global settings")
932
    for msg in self.cfg.VerifyConfig():
933
      feedback_fn("  - ERROR: %s" % msg)
934

    
935
    vg_name = self.cfg.GetVGName()
936
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
937
    nodelist = utils.NiceSort(self.cfg.GetNodeList())
938
    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
939
    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
940
    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
941
                        for iname in instancelist)
942
    i_non_redundant = [] # Non redundant instances
943
    i_non_a_balanced = [] # Non auto-balanced instances
944
    n_offline = [] # List of offline nodes
945
    n_drained = [] # List of nodes being drained
946
    node_volume = {}
947
    node_instance = {}
948
    node_info = {}
949
    instance_cfg = {}
950

    
951
    # FIXME: verify OS list
952
    # do local checksums
953
    master_files = [constants.CLUSTER_CONF_FILE]
954

    
955
    file_names = ssconf.SimpleStore().GetFileList()
956
    file_names.append(constants.SSL_CERT_FILE)
957
    file_names.append(constants.RAPI_CERT_FILE)
958
    file_names.extend(master_files)
959

    
960
    local_checksums = utils.FingerprintFiles(file_names)
961

    
962
    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
963
    node_verify_param = {
964
      constants.NV_FILELIST: file_names,
965
      constants.NV_NODELIST: [node.name for node in nodeinfo
966
                              if not node.offline],
967
      constants.NV_HYPERVISOR: hypervisors,
968
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
969
                                  node.secondary_ip) for node in nodeinfo
970
                                 if not node.offline],
971
      constants.NV_INSTANCELIST: hypervisors,
972
      constants.NV_VERSION: None,
973
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
974
      }
975
    if vg_name is not None:
976
      node_verify_param[constants.NV_VGLIST] = None
977
      node_verify_param[constants.NV_LVLIST] = vg_name
978
      node_verify_param[constants.NV_DRBDLIST] = None
979
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
980
                                           self.cfg.GetClusterName())
981

    
982
    cluster = self.cfg.GetClusterInfo()
983
    master_node = self.cfg.GetMasterNode()
984
    all_drbd_map = self.cfg.ComputeDRBDMap()
985

    
986
    for node_i in nodeinfo:
987
      node = node_i.name
988
      nresult = all_nvinfo[node].data
989

    
990
      if node_i.offline:
991
        feedback_fn("* Skipping offline node %s" % (node,))
992
        n_offline.append(node)
993
        continue
994

    
995
      if node == master_node:
996
        ntype = "master"
997
      elif node_i.master_candidate:
998
        ntype = "master candidate"
999
      elif node_i.drained:
1000
        ntype = "drained"
1001
        n_drained.append(node)
1002
      else:
1003
        ntype = "regular"
1004
      feedback_fn("* Verifying node %s (%s)" % (node, ntype))
1005

    
1006
      if all_nvinfo[node].failed or not isinstance(nresult, dict):
1007
        feedback_fn("  - ERROR: connection to %s failed" % (node,))
1008
        bad = True
1009
        continue
1010

    
1011
      node_drbd = {}
1012
      for minor, instance in all_drbd_map[node].items():
1013
        if instance not in instanceinfo:
1014
          feedback_fn("  - ERROR: ghost instance '%s' in temporary DRBD map" %
1015
                      instance)
1016
          # ghost instance should not be running, but otherwise we
1017
          # don't give double warnings (both ghost instance and
1018
          # unallocated minor in use)
1019
          node_drbd[minor] = (instance, False)
1020
        else:
1021
          instance = instanceinfo[instance]
1022
          node_drbd[minor] = (instance.name, instance.admin_up)
1023
      result = self._VerifyNode(node_i, file_names, local_checksums,
1024
                                nresult, feedback_fn, master_files,
1025
                                node_drbd, vg_name)
1026
      bad = bad or result
1027

    
1028
      lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
1029
      if vg_name is None:
1030
        node_volume[node] = {}
1031
      elif isinstance(lvdata, basestring):
1032
        feedback_fn("  - ERROR: LVM problem on node %s: %s" %
1033
                    (node, utils.SafeEncode(lvdata)))
1034
        bad = True
1035
        node_volume[node] = {}
1036
      elif not isinstance(lvdata, dict):
1037
        feedback_fn("  - ERROR: connection to %s failed (lvlist)" % (node,))
1038
        bad = True
1039
        continue
1040
      else:
1041
        node_volume[node] = lvdata
1042

    
1043
      # node_instance
1044
      idata = nresult.get(constants.NV_INSTANCELIST, None)
1045
      if not isinstance(idata, list):
1046
        feedback_fn("  - ERROR: connection to %s failed (instancelist)" %
1047
                    (node,))
1048
        bad = True
1049
        continue
1050

    
1051
      node_instance[node] = idata
1052

    
1053
      # node_info
1054
      nodeinfo = nresult.get(constants.NV_HVINFO, None)
1055
      if not isinstance(nodeinfo, dict):
1056
        feedback_fn("  - ERROR: connection to %s failed (hvinfo)" % (node,))
1057
        bad = True
1058
        continue
1059

    
1060
      try:
1061
        node_info[node] = {
1062
          "mfree": int(nodeinfo['memory_free']),
1063
          "pinst": [],
1064
          "sinst": [],
1065
          # dictionary holding all instances this node is secondary for,
1066
          # grouped by their primary node. Each key is a cluster node, and each
1067
          # value is a list of instances which have the key as primary and the
1068
          # current node as secondary.  this is handy to calculate N+1 memory
1069
          # availability if you can only failover from a primary to its
1070
          # secondary.
1071
          "sinst-by-pnode": {},
1072
        }
1073
        # FIXME: devise a free space model for file based instances as well
1074
        if vg_name is not None:
1075
          if (constants.NV_VGLIST not in nresult or
1076
              vg_name not in nresult[constants.NV_VGLIST]):
1077
            feedback_fn("  - ERROR: node %s didn't return data for the"
1078
                        " volume group '%s' - it is either missing or broken" %
1079
                        (node, vg_name))
1080
            bad = True
1081
            continue
1082
          node_info[node]["dfree"] = int(nresult[constants.NV_VGLIST][vg_name])
1083
      except (ValueError, KeyError):
1084
        feedback_fn("  - ERROR: invalid nodeinfo value returned"
1085
                    " from node %s" % (node,))
1086
        bad = True
1087
        continue
1088

    
1089
    node_vol_should = {}
1090

    
1091
    for instance in instancelist:
1092
      feedback_fn("* Verifying instance %s" % instance)
1093
      inst_config = instanceinfo[instance]
1094
      result =  self._VerifyInstance(instance, inst_config, node_volume,
1095
                                     node_instance, feedback_fn, n_offline)
1096
      bad = bad or result
1097
      inst_nodes_offline = []
1098

    
1099
      inst_config.MapLVsByNode(node_vol_should)
1100

    
1101
      instance_cfg[instance] = inst_config
1102

    
1103
      pnode = inst_config.primary_node
1104
      if pnode in node_info:
1105
        node_info[pnode]['pinst'].append(instance)
1106
      elif pnode not in n_offline:
1107
        feedback_fn("  - ERROR: instance %s, connection to primary node"
1108
                    " %s failed" % (instance, pnode))
1109
        bad = True
1110

    
1111
      if pnode in n_offline:
1112
        inst_nodes_offline.append(pnode)
1113

    
1114
      # If the instance is non-redundant we cannot survive losing its primary
1115
      # node, so we are not N+1 compliant. On the other hand we have no disk
1116
      # templates with more than one secondary so that situation is not well
1117
      # supported either.
1118
      # FIXME: does not support file-backed instances
1119
      if len(inst_config.secondary_nodes) == 0:
1120
        i_non_redundant.append(instance)
1121
      elif len(inst_config.secondary_nodes) > 1:
1122
        feedback_fn("  - WARNING: multiple secondaries for instance %s"
1123
                    % instance)
1124

    
1125
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
1126
        i_non_a_balanced.append(instance)
1127

    
1128
      for snode in inst_config.secondary_nodes:
1129
        if snode in node_info:
1130
          node_info[snode]['sinst'].append(instance)
1131
          if pnode not in node_info[snode]['sinst-by-pnode']:
1132
            node_info[snode]['sinst-by-pnode'][pnode] = []
1133
          node_info[snode]['sinst-by-pnode'][pnode].append(instance)
1134
        elif snode not in n_offline:
1135
          feedback_fn("  - ERROR: instance %s, connection to secondary node"
1136
                      " %s failed" % (instance, snode))
1137
          bad = True
1138
        if snode in n_offline:
1139
          inst_nodes_offline.append(snode)
1140

    
1141
      if inst_nodes_offline:
1142
        # warn that the instance lives on offline nodes, and set bad=True
1143
        feedback_fn("  - ERROR: instance lives on offline node(s) %s" %
1144
                    ", ".join(inst_nodes_offline))
1145
        bad = True
1146

    
1147
    feedback_fn("* Verifying orphan volumes")
1148
    result = self._VerifyOrphanVolumes(node_vol_should, node_volume,
1149
                                       feedback_fn)
1150
    bad = bad or result
1151

    
1152
    feedback_fn("* Verifying remaining instances")
1153
    result = self._VerifyOrphanInstances(instancelist, node_instance,
1154
                                         feedback_fn)
1155
    bad = bad or result
1156

    
1157
    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
1158
      feedback_fn("* Verifying N+1 Memory redundancy")
1159
      result = self._VerifyNPlusOneMemory(node_info, instance_cfg, feedback_fn)
1160
      bad = bad or result
1161

    
1162
    feedback_fn("* Other Notes")
1163
    if i_non_redundant:
1164
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
1165
                  % len(i_non_redundant))
1166

    
1167
    if i_non_a_balanced:
1168
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
1169
                  % len(i_non_a_balanced))
1170

    
1171
    if n_offline:
1172
      feedback_fn("  - NOTICE: %d offline node(s) found." % len(n_offline))
1173

    
1174
    if n_drained:
1175
      feedback_fn("  - NOTICE: %d drained node(s) found." % len(n_drained))
1176

    
1177
    return not bad
1178

    
1179
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
1180
    """Analize the post-hooks' result
1181

1182
    This method analyses the hook result, handles it, and sends some
1183
    nicely-formatted feedback back to the user.
1184

1185
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
1186
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
1187
    @param hooks_results: the results of the multi-node hooks rpc call
1188
    @param feedback_fn: function used send feedback back to the caller
1189
    @param lu_result: previous Exec result
1190
    @return: the new Exec result, based on the previous result
1191
        and hook results
1192

1193
    """
1194
    # We only really run POST phase hooks, and are only interested in
1195
    # their results
1196
    if phase == constants.HOOKS_PHASE_POST:
1197
      # Used to change hooks' output to proper indentation
1198
      indent_re = re.compile('^', re.M)
1199
      feedback_fn("* Hooks Results")
1200
      if not hooks_results:
1201
        feedback_fn("  - ERROR: general communication failure")
1202
        lu_result = 1
1203
      else:
1204
        for node_name in hooks_results:
1205
          show_node_header = True
1206
          res = hooks_results[node_name]
1207
          if res.failed or res.data is False or not isinstance(res.data, list):
1208
            if res.offline:
1209
              # no need to warn or set fail return value
1210
              continue
1211
            feedback_fn("    Communication failure in hooks execution")
1212
            lu_result = 1
1213
            continue
1214
          for script, hkr, output in res.data:
1215
            if hkr == constants.HKR_FAIL:
1216
              # The node header is only shown once, if there are
1217
              # failing hooks on that node
1218
              if show_node_header:
1219
                feedback_fn("  Node %s:" % node_name)
1220
                show_node_header = False
1221
              feedback_fn("    ERROR: Script %s failed, output:" % script)
1222
              output = indent_re.sub('      ', output)
1223
              feedback_fn("%s" % output)
1224
              lu_result = 1
1225

    
1226
      return lu_result
1227

    
1228

    
1229
class LUVerifyDisks(NoHooksLU):
1230
  """Verifies the cluster disks status.
1231

1232
  """
1233
  _OP_REQP = []
1234
  REQ_BGL = False
1235

    
1236
  def ExpandNames(self):
1237
    self.needed_locks = {
1238
      locking.LEVEL_NODE: locking.ALL_SET,
1239
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1240
    }
1241
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
1242

    
1243
  def CheckPrereq(self):
1244
    """Check prerequisites.
1245

1246
    This has no prerequisites.
1247

1248
    """
1249
    pass
1250

    
1251
  def Exec(self, feedback_fn):
1252
    """Verify integrity of cluster disks.
1253

1254
    """
1255
    result = res_nodes, res_nlvm, res_instances, res_missing = [], {}, [], {}
1256

    
1257
    vg_name = self.cfg.GetVGName()
1258
    nodes = utils.NiceSort(self.cfg.GetNodeList())
1259
    instances = [self.cfg.GetInstanceInfo(name)
1260
                 for name in self.cfg.GetInstanceList()]
1261

    
1262
    nv_dict = {}
1263
    for inst in instances:
1264
      inst_lvs = {}
1265
      if (not inst.admin_up or
1266
          inst.disk_template not in constants.DTS_NET_MIRROR):
1267
        continue
1268
      inst.MapLVsByNode(inst_lvs)
1269
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
1270
      for node, vol_list in inst_lvs.iteritems():
1271
        for vol in vol_list:
1272
          nv_dict[(node, vol)] = inst
1273

    
1274
    if not nv_dict:
1275
      return result
1276

    
1277
    node_lvs = self.rpc.call_volume_list(nodes, vg_name)
1278

    
1279
    to_act = set()
1280
    for node in nodes:
1281
      # node_volume
1282
      lvs = node_lvs[node]
1283
      if lvs.failed:
1284
        if not lvs.offline:
1285
          self.LogWarning("Connection to node %s failed: %s" %
1286
                          (node, lvs.data))
1287
        continue
1288
      lvs = lvs.data
1289
      if isinstance(lvs, basestring):
1290
        logging.warning("Error enumerating LVs on node %s: %s", node, lvs)
1291
        res_nlvm[node] = lvs
1292
        continue
1293
      elif not isinstance(lvs, dict):
1294
        logging.warning("Connection to node %s failed or invalid data"
1295
                        " returned", node)
1296
        res_nodes.append(node)
1297
        continue
1298

    
1299
      for lv_name, (_, lv_inactive, lv_online) in lvs.iteritems():
1300
        inst = nv_dict.pop((node, lv_name), None)
1301
        if (not lv_online and inst is not None
1302
            and inst.name not in res_instances):
1303
          res_instances.append(inst.name)
1304

    
1305
    # any leftover items in nv_dict are missing LVs, let's arrange the
1306
    # data better
1307
    for key, inst in nv_dict.iteritems():
1308
      if inst.name not in res_missing:
1309
        res_missing[inst.name] = []
1310
      res_missing[inst.name].append(key)
1311

    
1312
    return result
1313

    
1314

    
1315
class LURenameCluster(LogicalUnit):
1316
  """Rename the cluster.
1317

1318
  """
1319
  HPATH = "cluster-rename"
1320
  HTYPE = constants.HTYPE_CLUSTER
1321
  _OP_REQP = ["name"]
1322

    
1323
  def BuildHooksEnv(self):
1324
    """Build hooks env.
1325

1326
    """
1327
    env = {
1328
      "OP_TARGET": self.cfg.GetClusterName(),
1329
      "NEW_NAME": self.op.name,
1330
      }
1331
    mn = self.cfg.GetMasterNode()
1332
    return env, [mn], [mn]
1333

    
1334
  def CheckPrereq(self):
1335
    """Verify that the passed name is a valid one.
1336

1337
    """
1338
    hostname = utils.HostInfo(self.op.name)
1339

    
1340
    new_name = hostname.name
1341
    self.ip = new_ip = hostname.ip
1342
    old_name = self.cfg.GetClusterName()
1343
    old_ip = self.cfg.GetMasterIP()
1344
    if new_name == old_name and new_ip == old_ip:
1345
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
1346
                                 " cluster has changed")
1347
    if new_ip != old_ip:
1348
      if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
1349
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
1350
                                   " reachable on the network. Aborting." %
1351
                                   new_ip)
1352

    
1353
    self.op.name = new_name
1354

    
1355
  def Exec(self, feedback_fn):
1356
    """Rename the cluster.
1357

1358
    """
1359
    clustername = self.op.name
1360
    ip = self.ip
1361

    
1362
    # shutdown the master IP
1363
    master = self.cfg.GetMasterNode()
1364
    result = self.rpc.call_node_stop_master(master, False)
1365
    if result.failed or not result.data:
1366
      raise errors.OpExecError("Could not disable the master role")
1367

    
1368
    try:
1369
      cluster = self.cfg.GetClusterInfo()
1370
      cluster.cluster_name = clustername
1371
      cluster.master_ip = ip
1372
      self.cfg.Update(cluster)
1373

    
1374
      # update the known hosts file
1375
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
1376
      node_list = self.cfg.GetNodeList()
1377
      try:
1378
        node_list.remove(master)
1379
      except ValueError:
1380
        pass
1381
      result = self.rpc.call_upload_file(node_list,
1382
                                         constants.SSH_KNOWN_HOSTS_FILE)
1383
      for to_node, to_result in result.iteritems():
1384
         msg = to_result.RemoteFailMsg()
1385
         if msg:
1386
           msg = ("Copy of file %s to node %s failed: %s" %
1387
                   (constants.SSH_KNOWN_HOSTS_FILE, to_node, msg))
1388
           self.proc.LogWarning(msg)
1389

    
1390
    finally:
1391
      result = self.rpc.call_node_start_master(master, False)
1392
      if result.failed or not result.data:
1393
        self.LogWarning("Could not re-enable the master role on"
1394
                        " the master, please restart manually.")
1395

    
1396

    
1397
def _RecursiveCheckIfLVMBased(disk):
1398
  """Check if the given disk or its children are lvm-based.
1399

1400
  @type disk: L{objects.Disk}
1401
  @param disk: the disk to check
1402
  @rtype: booleean
1403
  @return: boolean indicating whether a LD_LV dev_type was found or not
1404

1405
  """
1406
  if disk.children:
1407
    for chdisk in disk.children:
1408
      if _RecursiveCheckIfLVMBased(chdisk):
1409
        return True
1410
  return disk.dev_type == constants.LD_LV
1411

    
1412

    
1413
class LUSetClusterParams(LogicalUnit):
1414
  """Change the parameters of the cluster.
1415

1416
  """
1417
  HPATH = "cluster-modify"
1418
  HTYPE = constants.HTYPE_CLUSTER
1419
  _OP_REQP = []
1420
  REQ_BGL = False
1421

    
1422
  def CheckArguments(self):
1423
    """Check parameters
1424

1425
    """
1426
    if not hasattr(self.op, "candidate_pool_size"):
1427
      self.op.candidate_pool_size = None
1428
    if self.op.candidate_pool_size is not None:
1429
      try:
1430
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1431
      except (ValueError, TypeError), err:
1432
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1433
                                   str(err))
1434
      if self.op.candidate_pool_size < 1:
1435
        raise errors.OpPrereqError("At least one master candidate needed")
1436

    
1437
  def ExpandNames(self):
1438
    # FIXME: in the future maybe other cluster params won't require checking on
1439
    # all nodes to be modified.
1440
    self.needed_locks = {
1441
      locking.LEVEL_NODE: locking.ALL_SET,
1442
    }
1443
    self.share_locks[locking.LEVEL_NODE] = 1
1444

    
1445
  def BuildHooksEnv(self):
1446
    """Build hooks env.
1447

1448
    """
1449
    env = {
1450
      "OP_TARGET": self.cfg.GetClusterName(),
1451
      "NEW_VG_NAME": self.op.vg_name,
1452
      }
1453
    mn = self.cfg.GetMasterNode()
1454
    return env, [mn], [mn]
1455

    
1456
  def CheckPrereq(self):
1457
    """Check prerequisites.
1458

1459
    This checks whether the given params don't conflict and
1460
    if the given volume group is valid.
1461

1462
    """
1463
    if self.op.vg_name is not None and not self.op.vg_name:
1464
      instances = self.cfg.GetAllInstancesInfo().values()
1465
      for inst in instances:
1466
        for disk in inst.disks:
1467
          if _RecursiveCheckIfLVMBased(disk):
1468
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1469
                                       " lvm-based instances exist")
1470

    
1471
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1472

    
1473
    # if vg_name not None, checks given volume group on all nodes
1474
    if self.op.vg_name:
1475
      vglist = self.rpc.call_vg_list(node_list)
1476
      for node in node_list:
1477
        if vglist[node].failed:
1478
          # ignoring down node
1479
          self.LogWarning("Node %s unreachable/error, ignoring" % node)
1480
          continue
1481
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].data,
1482
                                              self.op.vg_name,
1483
                                              constants.MIN_VG_SIZE)
1484
        if vgstatus:
1485
          raise errors.OpPrereqError("Error on node '%s': %s" %
1486
                                     (node, vgstatus))
1487

    
1488
    self.cluster = cluster = self.cfg.GetClusterInfo()
1489
    # validate beparams changes
1490
    if self.op.beparams:
1491
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
1492
      self.new_beparams = cluster.FillDict(
1493
        cluster.beparams[constants.BEGR_DEFAULT], self.op.beparams)
1494

    
1495
    # hypervisor list/parameters
1496
    self.new_hvparams = cluster.FillDict(cluster.hvparams, {})
1497
    if self.op.hvparams:
1498
      if not isinstance(self.op.hvparams, dict):
1499
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1500
      for hv_name, hv_dict in self.op.hvparams.items():
1501
        if hv_name not in self.new_hvparams:
1502
          self.new_hvparams[hv_name] = hv_dict
1503
        else:
1504
          self.new_hvparams[hv_name].update(hv_dict)
1505

    
1506
    if self.op.enabled_hypervisors is not None:
1507
      self.hv_list = self.op.enabled_hypervisors
1508
    else:
1509
      self.hv_list = cluster.enabled_hypervisors
1510

    
1511
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1512
      # either the enabled list has changed, or the parameters have, validate
1513
      for hv_name, hv_params in self.new_hvparams.items():
1514
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1515
            (self.op.enabled_hypervisors and
1516
             hv_name in self.op.enabled_hypervisors)):
1517
          # either this is a new hypervisor, or its parameters have changed
1518
          hv_class = hypervisor.GetHypervisor(hv_name)
1519
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1520
          hv_class.CheckParameterSyntax(hv_params)
1521
          _CheckHVParams(self, node_list, hv_name, hv_params)
1522

    
1523
  def Exec(self, feedback_fn):
1524
    """Change the parameters of the cluster.
1525

1526
    """
1527
    if self.op.vg_name is not None:
1528
      new_volume = self.op.vg_name
1529
      if not new_volume:
1530
        new_volume = None
1531
      if new_volume != self.cfg.GetVGName():
1532
        self.cfg.SetVGName(new_volume)
1533
      else:
1534
        feedback_fn("Cluster LVM configuration already in desired"
1535
                    " state, not changing")
1536
    if self.op.hvparams:
1537
      self.cluster.hvparams = self.new_hvparams
1538
    if self.op.enabled_hypervisors is not None:
1539
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1540
    if self.op.beparams:
1541
      self.cluster.beparams[constants.BEGR_DEFAULT] = self.new_beparams
1542
    if self.op.candidate_pool_size is not None:
1543
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1544

    
1545
    self.cfg.Update(self.cluster)
1546

    
1547
    # we want to update nodes after the cluster so that if any errors
1548
    # happen, we have recorded and saved the cluster info
1549
    if self.op.candidate_pool_size is not None:
1550
      _AdjustCandidatePool(self)
1551

    
1552

    
1553
def _RedistributeAncillaryFiles(lu, additional_nodes=None):
1554
  """Distribute additional files which are part of the cluster configuration.
1555

1556
  ConfigWriter takes care of distributing the config and ssconf files, but
1557
  there are more files which should be distributed to all nodes. This function
1558
  makes sure those are copied.
1559

1560
  @param lu: calling logical unit
1561
  @param additional_nodes: list of nodes not in the config to distribute to
1562

1563
  """
1564
  # 1. Gather target nodes
1565
  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
1566
  dist_nodes = lu.cfg.GetNodeList()
1567
  if additional_nodes is not None:
1568
    dist_nodes.extend(additional_nodes)
1569
  if myself.name in dist_nodes:
1570
    dist_nodes.remove(myself.name)
1571
  # 2. Gather files to distribute
1572
  dist_files = set([constants.ETC_HOSTS,
1573
                    constants.SSH_KNOWN_HOSTS_FILE,
1574
                    constants.RAPI_CERT_FILE,
1575
                    constants.RAPI_USERS_FILE,
1576
                   ])
1577

    
1578
  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
1579
  for hv_name in enabled_hypervisors:
1580
    hv_class = hypervisor.GetHypervisor(hv_name)
1581
    dist_files.update(hv_class.GetAncillaryFiles())
1582

    
1583
  # 3. Perform the files upload
1584
  for fname in dist_files:
1585
    if os.path.exists(fname):
1586
      result = lu.rpc.call_upload_file(dist_nodes, fname)
1587
      for to_node, to_result in result.items():
1588
         msg = to_result.RemoteFailMsg()
1589
         if msg:
1590
           msg = ("Copy of file %s to node %s failed: %s" %
1591
                   (fname, to_node, msg))
1592
           lu.proc.LogWarning(msg)
1593

    
1594

    
1595
class LURedistributeConfig(NoHooksLU):
1596
  """Force the redistribution of cluster configuration.
1597

1598
  This is a very simple LU.
1599

1600
  """
1601
  _OP_REQP = []
1602
  REQ_BGL = False
1603

    
1604
  def ExpandNames(self):
1605
    self.needed_locks = {
1606
      locking.LEVEL_NODE: locking.ALL_SET,
1607
    }
1608
    self.share_locks[locking.LEVEL_NODE] = 1
1609

    
1610
  def CheckPrereq(self):
1611
    """Check prerequisites.
1612

1613
    """
1614

    
1615
  def Exec(self, feedback_fn):
1616
    """Redistribute the configuration.
1617

1618
    """
1619
    self.cfg.Update(self.cfg.GetClusterInfo())
1620
    _RedistributeAncillaryFiles(self)
1621

    
1622

    
1623
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1624
  """Sleep and poll for an instance's disk to sync.
1625

1626
  """
1627
  if not instance.disks:
1628
    return True
1629

    
1630
  if not oneshot:
1631
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1632

    
1633
  node = instance.primary_node
1634

    
1635
  for dev in instance.disks:
1636
    lu.cfg.SetDiskID(dev, node)
1637

    
1638
  retries = 0
1639
  while True:
1640
    max_time = 0
1641
    done = True
1642
    cumul_degraded = False
1643
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1644
    if rstats.failed or not rstats.data:
1645
      lu.LogWarning("Can't get any data from node %s", node)
1646
      retries += 1
1647
      if retries >= 10:
1648
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1649
                                 " aborting." % node)
1650
      time.sleep(6)
1651
      continue
1652
    rstats = rstats.data
1653
    retries = 0
1654
    for i, mstat in enumerate(rstats):
1655
      if mstat is None:
1656
        lu.LogWarning("Can't compute data for node %s/%s",
1657
                           node, instance.disks[i].iv_name)
1658
        continue
1659
      # we ignore the ldisk parameter
1660
      perc_done, est_time, is_degraded, _ = mstat
1661
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1662
      if perc_done is not None:
1663
        done = False
1664
        if est_time is not None:
1665
          rem_time = "%d estimated seconds remaining" % est_time
1666
          max_time = est_time
1667
        else:
1668
          rem_time = "no time estimate"
1669
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1670
                        (instance.disks[i].iv_name, perc_done, rem_time))
1671
    if done or oneshot:
1672
      break
1673

    
1674
    time.sleep(min(60, max_time))
1675

    
1676
  if done:
1677
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1678
  return not cumul_degraded
1679

    
1680

    
1681
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1682
  """Check that mirrors are not degraded.
1683

1684
  The ldisk parameter, if True, will change the test from the
1685
  is_degraded attribute (which represents overall non-ok status for
1686
  the device(s)) to the ldisk (representing the local storage status).
1687

1688
  """
1689
  lu.cfg.SetDiskID(dev, node)
1690
  if ldisk:
1691
    idx = 6
1692
  else:
1693
    idx = 5
1694

    
1695
  result = True
1696
  if on_primary or dev.AssembleOnSecondary():
1697
    rstats = lu.rpc.call_blockdev_find(node, dev)
1698
    msg = rstats.RemoteFailMsg()
1699
    if msg:
1700
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1701
      result = False
1702
    elif not rstats.payload:
1703
      lu.LogWarning("Can't find disk on node %s", node)
1704
      result = False
1705
    else:
1706
      result = result and (not rstats.payload[idx])
1707
  if dev.children:
1708
    for child in dev.children:
1709
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1710

    
1711
  return result
1712

    
1713

    
1714
class LUDiagnoseOS(NoHooksLU):
1715
  """Logical unit for OS diagnose/query.
1716

1717
  """
1718
  _OP_REQP = ["output_fields", "names"]
1719
  REQ_BGL = False
1720
  _FIELDS_STATIC = utils.FieldSet()
1721
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1722

    
1723
  def ExpandNames(self):
1724
    if self.op.names:
1725
      raise errors.OpPrereqError("Selective OS query not supported")
1726

    
1727
    _CheckOutputFields(static=self._FIELDS_STATIC,
1728
                       dynamic=self._FIELDS_DYNAMIC,
1729
                       selected=self.op.output_fields)
1730

    
1731
    # Lock all nodes, in shared mode
1732
    # Temporary removal of locks, should be reverted later
1733
    # TODO: reintroduce locks when they are lighter-weight
1734
    self.needed_locks = {}
1735
    #self.share_locks[locking.LEVEL_NODE] = 1
1736
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1737

    
1738
  def CheckPrereq(self):
1739
    """Check prerequisites.
1740

1741
    """
1742

    
1743
  @staticmethod
1744
  def _DiagnoseByOS(node_list, rlist):
1745
    """Remaps a per-node return list into an a per-os per-node dictionary
1746

1747
    @param node_list: a list with the names of all nodes
1748
    @param rlist: a map with node names as keys and OS objects as values
1749

1750
    @rtype: dict
1751
    @return: a dictionary with osnames as keys and as value another map, with
1752
        nodes as keys and list of OS objects as values, eg::
1753

1754
          {"debian-etch": {"node1": [<object>,...],
1755
                           "node2": [<object>,]}
1756
          }
1757

1758
    """
1759
    all_os = {}
1760
    # we build here the list of nodes that didn't fail the RPC (at RPC
1761
    # level), so that nodes with a non-responding node daemon don't
1762
    # make all OSes invalid
1763
    good_nodes = [node_name for node_name in rlist
1764
                  if not rlist[node_name].failed]
1765
    for node_name, nr in rlist.iteritems():
1766
      if nr.failed or not nr.data:
1767
        continue
1768
      for os_obj in nr.data:
1769
        if os_obj.name not in all_os:
1770
          # build a list of nodes for this os containing empty lists
1771
          # for each node in node_list
1772
          all_os[os_obj.name] = {}
1773
          for nname in good_nodes:
1774
            all_os[os_obj.name][nname] = []
1775
        all_os[os_obj.name][node_name].append(os_obj)
1776
    return all_os
1777

    
1778
  def Exec(self, feedback_fn):
1779
    """Compute the list of OSes.
1780

1781
    """
1782
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
1783
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1784
    if node_data == False:
1785
      raise errors.OpExecError("Can't gather the list of OSes")
1786
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1787
    output = []
1788
    for os_name, os_data in pol.iteritems():
1789
      row = []
1790
      for field in self.op.output_fields:
1791
        if field == "name":
1792
          val = os_name
1793
        elif field == "valid":
1794
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1795
        elif field == "node_status":
1796
          val = {}
1797
          for node_name, nos_list in os_data.iteritems():
1798
            val[node_name] = [(v.status, v.path) for v in nos_list]
1799
        else:
1800
          raise errors.ParameterError(field)
1801
        row.append(val)
1802
      output.append(row)
1803

    
1804
    return output
1805

    
1806

    
1807
class LURemoveNode(LogicalUnit):
1808
  """Logical unit for removing a node.
1809

1810
  """
1811
  HPATH = "node-remove"
1812
  HTYPE = constants.HTYPE_NODE
1813
  _OP_REQP = ["node_name"]
1814

    
1815
  def BuildHooksEnv(self):
1816
    """Build hooks env.
1817

1818
    This doesn't run on the target node in the pre phase as a failed
1819
    node would then be impossible to remove.
1820

1821
    """
1822
    env = {
1823
      "OP_TARGET": self.op.node_name,
1824
      "NODE_NAME": self.op.node_name,
1825
      }
1826
    all_nodes = self.cfg.GetNodeList()
1827
    all_nodes.remove(self.op.node_name)
1828
    return env, all_nodes, all_nodes
1829

    
1830
  def CheckPrereq(self):
1831
    """Check prerequisites.
1832

1833
    This checks:
1834
     - the node exists in the configuration
1835
     - it does not have primary or secondary instances
1836
     - it's not the master
1837

1838
    Any errors are signalled by raising errors.OpPrereqError.
1839

1840
    """
1841
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1842
    if node is None:
1843
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1844

    
1845
    instance_list = self.cfg.GetInstanceList()
1846

    
1847
    masternode = self.cfg.GetMasterNode()
1848
    if node.name == masternode:
1849
      raise errors.OpPrereqError("Node is the master node,"
1850
                                 " you need to failover first.")
1851

    
1852
    for instance_name in instance_list:
1853
      instance = self.cfg.GetInstanceInfo(instance_name)
1854
      if node.name in instance.all_nodes:
1855
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1856
                                   " please remove first." % instance_name)
1857
    self.op.node_name = node.name
1858
    self.node = node
1859

    
1860
  def Exec(self, feedback_fn):
1861
    """Removes the node from the cluster.
1862

1863
    """
1864
    node = self.node
1865
    logging.info("Stopping the node daemon and removing configs from node %s",
1866
                 node.name)
1867

    
1868
    self.context.RemoveNode(node.name)
1869

    
1870
    self.rpc.call_node_leave_cluster(node.name)
1871

    
1872
    # Promote nodes to master candidate as needed
1873
    _AdjustCandidatePool(self)
1874

    
1875

    
1876
class LUQueryNodes(NoHooksLU):
1877
  """Logical unit for querying nodes.
1878

1879
  """
1880
  _OP_REQP = ["output_fields", "names", "use_locking"]
1881
  REQ_BGL = False
1882
  _FIELDS_DYNAMIC = utils.FieldSet(
1883
    "dtotal", "dfree",
1884
    "mtotal", "mnode", "mfree",
1885
    "bootid",
1886
    "ctotal", "cnodes", "csockets",
1887
    )
1888

    
1889
  _FIELDS_STATIC = utils.FieldSet(
1890
    "name", "pinst_cnt", "sinst_cnt",
1891
    "pinst_list", "sinst_list",
1892
    "pip", "sip", "tags",
1893
    "serial_no",
1894
    "master_candidate",
1895
    "master",
1896
    "offline",
1897
    "drained",
1898
    )
1899

    
1900
  def ExpandNames(self):
1901
    _CheckOutputFields(static=self._FIELDS_STATIC,
1902
                       dynamic=self._FIELDS_DYNAMIC,
1903
                       selected=self.op.output_fields)
1904

    
1905
    self.needed_locks = {}
1906
    self.share_locks[locking.LEVEL_NODE] = 1
1907

    
1908
    if self.op.names:
1909
      self.wanted = _GetWantedNodes(self, self.op.names)
1910
    else:
1911
      self.wanted = locking.ALL_SET
1912

    
1913
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
1914
    self.do_locking = self.do_node_query and self.op.use_locking
1915
    if self.do_locking:
1916
      # if we don't request only static fields, we need to lock the nodes
1917
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
1918

    
1919

    
1920
  def CheckPrereq(self):
1921
    """Check prerequisites.
1922

1923
    """
1924
    # The validation of the node list is done in the _GetWantedNodes,
1925
    # if non empty, and if empty, there's no validation to do
1926
    pass
1927

    
1928
  def Exec(self, feedback_fn):
1929
    """Computes the list of nodes and their attributes.
1930

1931
    """
1932
    all_info = self.cfg.GetAllNodesInfo()
1933
    if self.do_locking:
1934
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
1935
    elif self.wanted != locking.ALL_SET:
1936
      nodenames = self.wanted
1937
      missing = set(nodenames).difference(all_info.keys())
1938
      if missing:
1939
        raise errors.OpExecError(
1940
          "Some nodes were removed before retrieving their data: %s" % missing)
1941
    else:
1942
      nodenames = all_info.keys()
1943

    
1944
    nodenames = utils.NiceSort(nodenames)
1945
    nodelist = [all_info[name] for name in nodenames]
1946

    
1947
    # begin data gathering
1948

    
1949
    if self.do_node_query:
1950
      live_data = {}
1951
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
1952
                                          self.cfg.GetHypervisorType())
1953
      for name in nodenames:
1954
        nodeinfo = node_data[name]
1955
        if not nodeinfo.failed and nodeinfo.data:
1956
          nodeinfo = nodeinfo.data
1957
          fn = utils.TryConvert
1958
          live_data[name] = {
1959
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
1960
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
1961
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
1962
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
1963
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
1964
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
1965
            "bootid": nodeinfo.get('bootid', None),
1966
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
1967
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
1968
            }
1969
        else:
1970
          live_data[name] = {}
1971
    else:
1972
      live_data = dict.fromkeys(nodenames, {})
1973

    
1974
    node_to_primary = dict([(name, set()) for name in nodenames])
1975
    node_to_secondary = dict([(name, set()) for name in nodenames])
1976

    
1977
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
1978
                             "sinst_cnt", "sinst_list"))
1979
    if inst_fields & frozenset(self.op.output_fields):
1980
      instancelist = self.cfg.GetInstanceList()
1981

    
1982
      for instance_name in instancelist:
1983
        inst = self.cfg.GetInstanceInfo(instance_name)
1984
        if inst.primary_node in node_to_primary:
1985
          node_to_primary[inst.primary_node].add(inst.name)
1986
        for secnode in inst.secondary_nodes:
1987
          if secnode in node_to_secondary:
1988
            node_to_secondary[secnode].add(inst.name)
1989

    
1990
    master_node = self.cfg.GetMasterNode()
1991

    
1992
    # end data gathering
1993

    
1994
    output = []
1995
    for node in nodelist:
1996
      node_output = []
1997
      for field in self.op.output_fields:
1998
        if field == "name":
1999
          val = node.name
2000
        elif field == "pinst_list":
2001
          val = list(node_to_primary[node.name])
2002
        elif field == "sinst_list":
2003
          val = list(node_to_secondary[node.name])
2004
        elif field == "pinst_cnt":
2005
          val = len(node_to_primary[node.name])
2006
        elif field == "sinst_cnt":
2007
          val = len(node_to_secondary[node.name])
2008
        elif field == "pip":
2009
          val = node.primary_ip
2010
        elif field == "sip":
2011
          val = node.secondary_ip
2012
        elif field == "tags":
2013
          val = list(node.GetTags())
2014
        elif field == "serial_no":
2015
          val = node.serial_no
2016
        elif field == "master_candidate":
2017
          val = node.master_candidate
2018
        elif field == "master":
2019
          val = node.name == master_node
2020
        elif field == "offline":
2021
          val = node.offline
2022
        elif field == "drained":
2023
          val = node.drained
2024
        elif self._FIELDS_DYNAMIC.Matches(field):
2025
          val = live_data[node.name].get(field, None)
2026
        else:
2027
          raise errors.ParameterError(field)
2028
        node_output.append(val)
2029
      output.append(node_output)
2030

    
2031
    return output
2032

    
2033

    
2034
class LUQueryNodeVolumes(NoHooksLU):
2035
  """Logical unit for getting volumes on node(s).
2036

2037
  """
2038
  _OP_REQP = ["nodes", "output_fields"]
2039
  REQ_BGL = False
2040
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
2041
  _FIELDS_STATIC = utils.FieldSet("node")
2042

    
2043
  def ExpandNames(self):
2044
    _CheckOutputFields(static=self._FIELDS_STATIC,
2045
                       dynamic=self._FIELDS_DYNAMIC,
2046
                       selected=self.op.output_fields)
2047

    
2048
    self.needed_locks = {}
2049
    self.share_locks[locking.LEVEL_NODE] = 1
2050
    if not self.op.nodes:
2051
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2052
    else:
2053
      self.needed_locks[locking.LEVEL_NODE] = \
2054
        _GetWantedNodes(self, self.op.nodes)
2055

    
2056
  def CheckPrereq(self):
2057
    """Check prerequisites.
2058

2059
    This checks that the fields required are valid output fields.
2060

2061
    """
2062
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
2063

    
2064
  def Exec(self, feedback_fn):
2065
    """Computes the list of nodes and their attributes.
2066

2067
    """
2068
    nodenames = self.nodes
2069
    volumes = self.rpc.call_node_volumes(nodenames)
2070

    
2071
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
2072
             in self.cfg.GetInstanceList()]
2073

    
2074
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
2075

    
2076
    output = []
2077
    for node in nodenames:
2078
      if node not in volumes or volumes[node].failed or not volumes[node].data:
2079
        continue
2080

    
2081
      node_vols = volumes[node].data[:]
2082
      node_vols.sort(key=lambda vol: vol['dev'])
2083

    
2084
      for vol in node_vols:
2085
        node_output = []
2086
        for field in self.op.output_fields:
2087
          if field == "node":
2088
            val = node
2089
          elif field == "phys":
2090
            val = vol['dev']
2091
          elif field == "vg":
2092
            val = vol['vg']
2093
          elif field == "name":
2094
            val = vol['name']
2095
          elif field == "size":
2096
            val = int(float(vol['size']))
2097
          elif field == "instance":
2098
            for inst in ilist:
2099
              if node not in lv_by_node[inst]:
2100
                continue
2101
              if vol['name'] in lv_by_node[inst][node]:
2102
                val = inst.name
2103
                break
2104
            else:
2105
              val = '-'
2106
          else:
2107
            raise errors.ParameterError(field)
2108
          node_output.append(str(val))
2109

    
2110
        output.append(node_output)
2111

    
2112
    return output
2113

    
2114

    
2115
class LUAddNode(LogicalUnit):
2116
  """Logical unit for adding node to the cluster.
2117

2118
  """
2119
  HPATH = "node-add"
2120
  HTYPE = constants.HTYPE_NODE
2121
  _OP_REQP = ["node_name"]
2122

    
2123
  def BuildHooksEnv(self):
2124
    """Build hooks env.
2125

2126
    This will run on all nodes before, and on all nodes + the new node after.
2127

2128
    """
2129
    env = {
2130
      "OP_TARGET": self.op.node_name,
2131
      "NODE_NAME": self.op.node_name,
2132
      "NODE_PIP": self.op.primary_ip,
2133
      "NODE_SIP": self.op.secondary_ip,
2134
      }
2135
    nodes_0 = self.cfg.GetNodeList()
2136
    nodes_1 = nodes_0 + [self.op.node_name, ]
2137
    return env, nodes_0, nodes_1
2138

    
2139
  def CheckPrereq(self):
2140
    """Check prerequisites.
2141

2142
    This checks:
2143
     - the new node is not already in the config
2144
     - it is resolvable
2145
     - its parameters (single/dual homed) matches the cluster
2146

2147
    Any errors are signalled by raising errors.OpPrereqError.
2148

2149
    """
2150
    node_name = self.op.node_name
2151
    cfg = self.cfg
2152

    
2153
    dns_data = utils.HostInfo(node_name)
2154

    
2155
    node = dns_data.name
2156
    primary_ip = self.op.primary_ip = dns_data.ip
2157
    secondary_ip = getattr(self.op, "secondary_ip", None)
2158
    if secondary_ip is None:
2159
      secondary_ip = primary_ip
2160
    if not utils.IsValidIP(secondary_ip):
2161
      raise errors.OpPrereqError("Invalid secondary IP given")
2162
    self.op.secondary_ip = secondary_ip
2163

    
2164
    node_list = cfg.GetNodeList()
2165
    if not self.op.readd and node in node_list:
2166
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2167
                                 node)
2168
    elif self.op.readd and node not in node_list:
2169
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2170

    
2171
    for existing_node_name in node_list:
2172
      existing_node = cfg.GetNodeInfo(existing_node_name)
2173

    
2174
      if self.op.readd and node == existing_node_name:
2175
        if (existing_node.primary_ip != primary_ip or
2176
            existing_node.secondary_ip != secondary_ip):
2177
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2178
                                     " address configuration as before")
2179
        continue
2180

    
2181
      if (existing_node.primary_ip == primary_ip or
2182
          existing_node.secondary_ip == primary_ip or
2183
          existing_node.primary_ip == secondary_ip or
2184
          existing_node.secondary_ip == secondary_ip):
2185
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2186
                                   " existing node %s" % existing_node.name)
2187

    
2188
    # check that the type of the node (single versus dual homed) is the
2189
    # same as for the master
2190
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2191
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2192
    newbie_singlehomed = secondary_ip == primary_ip
2193
    if master_singlehomed != newbie_singlehomed:
2194
      if master_singlehomed:
2195
        raise errors.OpPrereqError("The master has no private ip but the"
2196
                                   " new node has one")
2197
      else:
2198
        raise errors.OpPrereqError("The master has a private ip but the"
2199
                                   " new node doesn't have one")
2200

    
2201
    # checks reachablity
2202
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2203
      raise errors.OpPrereqError("Node not reachable by ping")
2204

    
2205
    if not newbie_singlehomed:
2206
      # check reachability from my secondary ip to newbie's secondary ip
2207
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2208
                           source=myself.secondary_ip):
2209
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2210
                                   " based ping to noded port")
2211

    
2212
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2213
    mc_now, _ = self.cfg.GetMasterCandidateStats()
2214
    master_candidate = mc_now < cp_size
2215

    
2216
    self.new_node = objects.Node(name=node,
2217
                                 primary_ip=primary_ip,
2218
                                 secondary_ip=secondary_ip,
2219
                                 master_candidate=master_candidate,
2220
                                 offline=False, drained=False)
2221

    
2222
  def Exec(self, feedback_fn):
2223
    """Adds the new node to the cluster.
2224

2225
    """
2226
    new_node = self.new_node
2227
    node = new_node.name
2228

    
2229
    # check connectivity
2230
    result = self.rpc.call_version([node])[node]
2231
    result.Raise()
2232
    if result.data:
2233
      if constants.PROTOCOL_VERSION == result.data:
2234
        logging.info("Communication to node %s fine, sw version %s match",
2235
                     node, result.data)
2236
      else:
2237
        raise errors.OpExecError("Version mismatch master version %s,"
2238
                                 " node version %s" %
2239
                                 (constants.PROTOCOL_VERSION, result.data))
2240
    else:
2241
      raise errors.OpExecError("Cannot get version from the new node")
2242

    
2243
    # setup ssh on node
2244
    logging.info("Copy ssh key to node %s", node)
2245
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2246
    keyarray = []
2247
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2248
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2249
                priv_key, pub_key]
2250

    
2251
    for i in keyfiles:
2252
      f = open(i, 'r')
2253
      try:
2254
        keyarray.append(f.read())
2255
      finally:
2256
        f.close()
2257

    
2258
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2259
                                    keyarray[2],
2260
                                    keyarray[3], keyarray[4], keyarray[5])
2261

    
2262
    msg = result.RemoteFailMsg()
2263
    if msg:
2264
      raise errors.OpExecError("Cannot transfer ssh keys to the"
2265
                               " new node: %s" % msg)
2266

    
2267
    # Add node to our /etc/hosts, and add key to known_hosts
2268
    utils.AddHostToEtcHosts(new_node.name)
2269

    
2270
    if new_node.secondary_ip != new_node.primary_ip:
2271
      result = self.rpc.call_node_has_ip_address(new_node.name,
2272
                                                 new_node.secondary_ip)
2273
      if result.failed or not result.data:
2274
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2275
                                 " you gave (%s). Please fix and re-run this"
2276
                                 " command." % new_node.secondary_ip)
2277

    
2278
    node_verify_list = [self.cfg.GetMasterNode()]
2279
    node_verify_param = {
2280
      'nodelist': [node],
2281
      # TODO: do a node-net-test as well?
2282
    }
2283

    
2284
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2285
                                       self.cfg.GetClusterName())
2286
    for verifier in node_verify_list:
2287
      if result[verifier].failed or not result[verifier].data:
2288
        raise errors.OpExecError("Cannot communicate with %s's node daemon"
2289
                                 " for remote verification" % verifier)
2290
      if result[verifier].data['nodelist']:
2291
        for failed in result[verifier].data['nodelist']:
2292
          feedback_fn("ssh/hostname verification failed %s -> %s" %
2293
                      (verifier, result[verifier].data['nodelist'][failed]))
2294
        raise errors.OpExecError("ssh/hostname verification failed.")
2295

    
2296
    if self.op.readd:
2297
      _RedistributeAncillaryFiles(self)
2298
      self.context.ReaddNode(new_node)
2299
    else:
2300
      _RedistributeAncillaryFiles(self, additional_nodes=node)
2301
      self.context.AddNode(new_node)
2302

    
2303

    
2304
class LUSetNodeParams(LogicalUnit):
2305
  """Modifies the parameters of a node.
2306

2307
  """
2308
  HPATH = "node-modify"
2309
  HTYPE = constants.HTYPE_NODE
2310
  _OP_REQP = ["node_name"]
2311
  REQ_BGL = False
2312

    
2313
  def CheckArguments(self):
2314
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2315
    if node_name is None:
2316
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2317
    self.op.node_name = node_name
2318
    _CheckBooleanOpField(self.op, 'master_candidate')
2319
    _CheckBooleanOpField(self.op, 'offline')
2320
    _CheckBooleanOpField(self.op, 'drained')
2321
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
2322
    if all_mods.count(None) == 3:
2323
      raise errors.OpPrereqError("Please pass at least one modification")
2324
    if all_mods.count(True) > 1:
2325
      raise errors.OpPrereqError("Can't set the node into more than one"
2326
                                 " state at the same time")
2327

    
2328
  def ExpandNames(self):
2329
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2330

    
2331
  def BuildHooksEnv(self):
2332
    """Build hooks env.
2333

2334
    This runs on the master node.
2335

2336
    """
2337
    env = {
2338
      "OP_TARGET": self.op.node_name,
2339
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2340
      "OFFLINE": str(self.op.offline),
2341
      "DRAINED": str(self.op.drained),
2342
      }
2343
    nl = [self.cfg.GetMasterNode(),
2344
          self.op.node_name]
2345
    return env, nl, nl
2346

    
2347
  def CheckPrereq(self):
2348
    """Check prerequisites.
2349

2350
    This only checks the instance list against the existing names.
2351

2352
    """
2353
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2354

    
2355
    if ((self.op.master_candidate == False or self.op.offline == True or
2356
         self.op.drained == True) and node.master_candidate):
2357
      # we will demote the node from master_candidate
2358
      if self.op.node_name == self.cfg.GetMasterNode():
2359
        raise errors.OpPrereqError("The master node has to be a"
2360
                                   " master candidate, online and not drained")
2361
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2362
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2363
      if num_candidates <= cp_size:
2364
        msg = ("Not enough master candidates (desired"
2365
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2366
        if self.op.force:
2367
          self.LogWarning(msg)
2368
        else:
2369
          raise errors.OpPrereqError(msg)
2370

    
2371
    if (self.op.master_candidate == True and
2372
        ((node.offline and not self.op.offline == False) or
2373
         (node.drained and not self.op.drained == False))):
2374
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
2375
                                 " to master_candidate" % node.name)
2376

    
2377
    return
2378

    
2379
  def Exec(self, feedback_fn):
2380
    """Modifies a node.
2381

2382
    """
2383
    node = self.node
2384

    
2385
    result = []
2386
    changed_mc = False
2387

    
2388
    if self.op.offline is not None:
2389
      node.offline = self.op.offline
2390
      result.append(("offline", str(self.op.offline)))
2391
      if self.op.offline == True:
2392
        if node.master_candidate:
2393
          node.master_candidate = False
2394
          changed_mc = True
2395
          result.append(("master_candidate", "auto-demotion due to offline"))
2396
        if node.drained:
2397
          node.drained = False
2398
          result.append(("drained", "clear drained status due to offline"))
2399

    
2400
    if self.op.master_candidate is not None:
2401
      node.master_candidate = self.op.master_candidate
2402
      changed_mc = True
2403
      result.append(("master_candidate", str(self.op.master_candidate)))
2404
      if self.op.master_candidate == False:
2405
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2406
        msg = rrc.RemoteFailMsg()
2407
        if msg:
2408
          self.LogWarning("Node failed to demote itself: %s" % msg)
2409

    
2410
    if self.op.drained is not None:
2411
      node.drained = self.op.drained
2412
      result.append(("drained", str(self.op.drained)))
2413
      if self.op.drained == True:
2414
        if node.master_candidate:
2415
          node.master_candidate = False
2416
          changed_mc = True
2417
          result.append(("master_candidate", "auto-demotion due to drain"))
2418
        if node.offline:
2419
          node.offline = False
2420
          result.append(("offline", "clear offline status due to drain"))
2421

    
2422
    # this will trigger configuration file update, if needed
2423
    self.cfg.Update(node)
2424
    # this will trigger job queue propagation or cleanup
2425
    if changed_mc:
2426
      self.context.ReaddNode(node)
2427

    
2428
    return result
2429

    
2430

    
2431
class LUPowercycleNode(NoHooksLU):
2432
  """Powercycles a node.
2433

2434
  """
2435
  _OP_REQP = ["node_name", "force"]
2436
  REQ_BGL = False
2437

    
2438
  def CheckArguments(self):
2439
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2440
    if node_name is None:
2441
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2442
    self.op.node_name = node_name
2443
    if node_name == self.cfg.GetMasterNode() and not self.op.force:
2444
      raise errors.OpPrereqError("The node is the master and the force"
2445
                                 " parameter was not set")
2446

    
2447
  def ExpandNames(self):
2448
    """Locking for PowercycleNode.
2449

2450
    This is a last-resource option and shouldn't block on other
2451
    jobs. Therefore, we grab no locks.
2452

2453
    """
2454
    self.needed_locks = {}
2455

    
2456
  def CheckPrereq(self):
2457
    """Check prerequisites.
2458

2459
    This LU has no prereqs.
2460

2461
    """
2462
    pass
2463

    
2464
  def Exec(self, feedback_fn):
2465
    """Reboots a node.
2466

2467
    """
2468
    result = self.rpc.call_node_powercycle(self.op.node_name,
2469
                                           self.cfg.GetHypervisorType())
2470
    msg = result.RemoteFailMsg()
2471
    if msg:
2472
      raise errors.OpExecError("Failed to schedule the reboot: %s" % msg)
2473
    return result.payload
2474

    
2475

    
2476
class LUQueryClusterInfo(NoHooksLU):
2477
  """Query cluster configuration.
2478

2479
  """
2480
  _OP_REQP = []
2481
  REQ_BGL = False
2482

    
2483
  def ExpandNames(self):
2484
    self.needed_locks = {}
2485

    
2486
  def CheckPrereq(self):
2487
    """No prerequsites needed for this LU.
2488

2489
    """
2490
    pass
2491

    
2492
  def Exec(self, feedback_fn):
2493
    """Return cluster config.
2494

2495
    """
2496
    cluster = self.cfg.GetClusterInfo()
2497
    result = {
2498
      "software_version": constants.RELEASE_VERSION,
2499
      "protocol_version": constants.PROTOCOL_VERSION,
2500
      "config_version": constants.CONFIG_VERSION,
2501
      "os_api_version": constants.OS_API_VERSION,
2502
      "export_version": constants.EXPORT_VERSION,
2503
      "architecture": (platform.architecture()[0], platform.machine()),
2504
      "name": cluster.cluster_name,
2505
      "master": cluster.master_node,
2506
      "default_hypervisor": cluster.default_hypervisor,
2507
      "enabled_hypervisors": cluster.enabled_hypervisors,
2508
      "hvparams": dict([(hypervisor, cluster.hvparams[hypervisor])
2509
                        for hypervisor in cluster.enabled_hypervisors]),
2510
      "beparams": cluster.beparams,
2511
      "candidate_pool_size": cluster.candidate_pool_size,
2512
      "default_bridge": cluster.default_bridge,
2513
      "master_netdev": cluster.master_netdev,
2514
      "volume_group_name": cluster.volume_group_name,
2515
      "file_storage_dir": cluster.file_storage_dir,
2516
      }
2517

    
2518
    return result
2519

    
2520

    
2521
class LUQueryConfigValues(NoHooksLU):
2522
  """Return configuration values.
2523

2524
  """
2525
  _OP_REQP = []
2526
  REQ_BGL = False
2527
  _FIELDS_DYNAMIC = utils.FieldSet()
2528
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2529

    
2530
  def ExpandNames(self):
2531
    self.needed_locks = {}
2532

    
2533
    _CheckOutputFields(static=self._FIELDS_STATIC,
2534
                       dynamic=self._FIELDS_DYNAMIC,
2535
                       selected=self.op.output_fields)
2536

    
2537
  def CheckPrereq(self):
2538
    """No prerequisites.
2539

2540
    """
2541
    pass
2542

    
2543
  def Exec(self, feedback_fn):
2544
    """Dump a representation of the cluster config to the standard output.
2545

2546
    """
2547
    values = []
2548
    for field in self.op.output_fields:
2549
      if field == "cluster_name":
2550
        entry = self.cfg.GetClusterName()
2551
      elif field == "master_node":
2552
        entry = self.cfg.GetMasterNode()
2553
      elif field == "drain_flag":
2554
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2555
      else:
2556
        raise errors.ParameterError(field)
2557
      values.append(entry)
2558
    return values
2559

    
2560

    
2561
class LUActivateInstanceDisks(NoHooksLU):
2562
  """Bring up an instance's disks.
2563

2564
  """
2565
  _OP_REQP = ["instance_name"]
2566
  REQ_BGL = False
2567

    
2568
  def ExpandNames(self):
2569
    self._ExpandAndLockInstance()
2570
    self.needed_locks[locking.LEVEL_NODE] = []
2571
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2572

    
2573
  def DeclareLocks(self, level):
2574
    if level == locking.LEVEL_NODE:
2575
      self._LockInstancesNodes()
2576

    
2577
  def CheckPrereq(self):
2578
    """Check prerequisites.
2579

2580
    This checks that the instance is in the cluster.
2581

2582
    """
2583
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2584
    assert self.instance is not None, \
2585
      "Cannot retrieve locked instance %s" % self.op.instance_name
2586
    _CheckNodeOnline(self, self.instance.primary_node)
2587

    
2588
  def Exec(self, feedback_fn):
2589
    """Activate the disks.
2590

2591
    """
2592
    disks_ok, disks_info = _AssembleInstanceDisks(self, self.instance)
2593
    if not disks_ok:
2594
      raise errors.OpExecError("Cannot activate block devices")
2595

    
2596
    return disks_info
2597

    
2598

    
2599
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False):
2600
  """Prepare the block devices for an instance.
2601

2602
  This sets up the block devices on all nodes.
2603

2604
  @type lu: L{LogicalUnit}
2605
  @param lu: the logical unit on whose behalf we execute
2606
  @type instance: L{objects.Instance}
2607
  @param instance: the instance for whose disks we assemble
2608
  @type ignore_secondaries: boolean
2609
  @param ignore_secondaries: if true, errors on secondary nodes
2610
      won't result in an error return from the function
2611
  @return: False if the operation failed, otherwise a list of
2612
      (host, instance_visible_name, node_visible_name)
2613
      with the mapping from node devices to instance devices
2614

2615
  """
2616
  device_info = []
2617
  disks_ok = True
2618
  iname = instance.name
2619
  # With the two passes mechanism we try to reduce the window of
2620
  # opportunity for the race condition of switching DRBD to primary
2621
  # before handshaking occured, but we do not eliminate it
2622

    
2623
  # The proper fix would be to wait (with some limits) until the
2624
  # connection has been made and drbd transitions from WFConnection
2625
  # into any other network-connected state (Connected, SyncTarget,
2626
  # SyncSource, etc.)
2627

    
2628
  # 1st pass, assemble on all nodes in secondary mode
2629
  for inst_disk in instance.disks:
2630
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2631
      lu.cfg.SetDiskID(node_disk, node)
2632
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2633
      msg = result.RemoteFailMsg()
2634
      if msg:
2635
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2636
                           " (is_primary=False, pass=1): %s",
2637
                           inst_disk.iv_name, node, msg)
2638
        if not ignore_secondaries:
2639
          disks_ok = False
2640

    
2641
  # FIXME: race condition on drbd migration to primary
2642

    
2643
  # 2nd pass, do only the primary node
2644
  for inst_disk in instance.disks:
2645
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2646
      if node != instance.primary_node:
2647
        continue
2648
      lu.cfg.SetDiskID(node_disk, node)
2649
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2650
      msg = result.RemoteFailMsg()
2651
      if msg:
2652
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2653
                           " (is_primary=True, pass=2): %s",
2654
                           inst_disk.iv_name, node, msg)
2655
        disks_ok = False
2656
    device_info.append((instance.primary_node, inst_disk.iv_name,
2657
                        result.payload))
2658

    
2659
  # leave the disks configured for the primary node
2660
  # this is a workaround that would be fixed better by
2661
  # improving the logical/physical id handling
2662
  for disk in instance.disks:
2663
    lu.cfg.SetDiskID(disk, instance.primary_node)
2664

    
2665
  return disks_ok, device_info
2666

    
2667

    
2668
def _StartInstanceDisks(lu, instance, force):
2669
  """Start the disks of an instance.
2670

2671
  """
2672
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2673
                                           ignore_secondaries=force)
2674
  if not disks_ok:
2675
    _ShutdownInstanceDisks(lu, instance)
2676
    if force is not None and not force:
2677
      lu.proc.LogWarning("", hint="If the message above refers to a"
2678
                         " secondary node,"
2679
                         " you can retry the operation using '--force'.")
2680
    raise errors.OpExecError("Disk consistency error")
2681

    
2682

    
2683
class LUDeactivateInstanceDisks(NoHooksLU):
2684
  """Shutdown an instance's disks.
2685

2686
  """
2687
  _OP_REQP = ["instance_name"]
2688
  REQ_BGL = False
2689

    
2690
  def ExpandNames(self):
2691
    self._ExpandAndLockInstance()
2692
    self.needed_locks[locking.LEVEL_NODE] = []
2693
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2694

    
2695
  def DeclareLocks(self, level):
2696
    if level == locking.LEVEL_NODE:
2697
      self._LockInstancesNodes()
2698

    
2699
  def CheckPrereq(self):
2700
    """Check prerequisites.
2701

2702
    This checks that the instance is in the cluster.
2703

2704
    """
2705
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2706
    assert self.instance is not None, \
2707
      "Cannot retrieve locked instance %s" % self.op.instance_name
2708

    
2709
  def Exec(self, feedback_fn):
2710
    """Deactivate the disks
2711

2712
    """
2713
    instance = self.instance
2714
    _SafeShutdownInstanceDisks(self, instance)
2715

    
2716

    
2717
def _SafeShutdownInstanceDisks(lu, instance):
2718
  """Shutdown block devices of an instance.
2719

2720
  This function checks if an instance is running, before calling
2721
  _ShutdownInstanceDisks.
2722

2723
  """
2724
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2725
                                      [instance.hypervisor])
2726
  ins_l = ins_l[instance.primary_node]
2727
  if ins_l.failed or not isinstance(ins_l.data, list):
2728
    raise errors.OpExecError("Can't contact node '%s'" %
2729
                             instance.primary_node)
2730

    
2731
  if instance.name in ins_l.data:
2732
    raise errors.OpExecError("Instance is running, can't shutdown"
2733
                             " block devices.")
2734

    
2735
  _ShutdownInstanceDisks(lu, instance)
2736

    
2737

    
2738
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2739
  """Shutdown block devices of an instance.
2740

2741
  This does the shutdown on all nodes of the instance.
2742

2743
  If the ignore_primary is false, errors on the primary node are
2744
  ignored.
2745

2746
  """
2747
  all_result = True
2748
  for disk in instance.disks:
2749
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2750
      lu.cfg.SetDiskID(top_disk, node)
2751
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2752
      msg = result.RemoteFailMsg()
2753
      if msg:
2754
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2755
                      disk.iv_name, node, msg)
2756
        if not ignore_primary or node != instance.primary_node:
2757
          all_result = False
2758
  return all_result
2759

    
2760

    
2761
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2762
  """Checks if a node has enough free memory.
2763

2764
  This function check if a given node has the needed amount of free
2765
  memory. In case the node has less memory or we cannot get the
2766
  information from the node, this function raise an OpPrereqError
2767
  exception.
2768

2769
  @type lu: C{LogicalUnit}
2770
  @param lu: a logical unit from which we get configuration data
2771
  @type node: C{str}
2772
  @param node: the node to check
2773
  @type reason: C{str}
2774
  @param reason: string to use in the error message
2775
  @type requested: C{int}
2776
  @param requested: the amount of memory in MiB to check for
2777
  @type hypervisor_name: C{str}
2778
  @param hypervisor_name: the hypervisor to ask for memory stats
2779
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2780
      we cannot check the node
2781

2782
  """
2783
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2784
  nodeinfo[node].Raise()
2785
  free_mem = nodeinfo[node].data.get('memory_free')
2786
  if not isinstance(free_mem, int):
2787
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2788
                             " was '%s'" % (node, free_mem))
2789
  if requested > free_mem:
2790
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2791
                             " needed %s MiB, available %s MiB" %
2792
                             (node, reason, requested, free_mem))
2793

    
2794

    
2795
class LUStartupInstance(LogicalUnit):
2796
  """Starts an instance.
2797

2798
  """
2799
  HPATH = "instance-start"
2800
  HTYPE = constants.HTYPE_INSTANCE
2801
  _OP_REQP = ["instance_name", "force"]
2802
  REQ_BGL = False
2803

    
2804
  def ExpandNames(self):
2805
    self._ExpandAndLockInstance()
2806

    
2807
  def BuildHooksEnv(self):
2808
    """Build hooks env.
2809

2810
    This runs on master, primary and secondary nodes of the instance.
2811

2812
    """
2813
    env = {
2814
      "FORCE": self.op.force,
2815
      }
2816
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2817
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2818
    return env, nl, nl
2819

    
2820
  def CheckPrereq(self):
2821
    """Check prerequisites.
2822

2823
    This checks that the instance is in the cluster.
2824

2825
    """
2826
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2827
    assert self.instance is not None, \
2828
      "Cannot retrieve locked instance %s" % self.op.instance_name
2829

    
2830
    # extra beparams
2831
    self.beparams = getattr(self.op, "beparams", {})
2832
    if self.beparams:
2833
      if not isinstance(self.beparams, dict):
2834
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
2835
                                   " dict" % (type(self.beparams), ))
2836
      # fill the beparams dict
2837
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
2838
      self.op.beparams = self.beparams
2839

    
2840
    # extra hvparams
2841
    self.hvparams = getattr(self.op, "hvparams", {})
2842
    if self.hvparams:
2843
      if not isinstance(self.hvparams, dict):
2844
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
2845
                                   " dict" % (type(self.hvparams), ))
2846

    
2847
      # check hypervisor parameter syntax (locally)
2848
      cluster = self.cfg.GetClusterInfo()
2849
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
2850
      filled_hvp = cluster.FillDict(cluster.hvparams[instance.hypervisor],
2851
                                    instance.hvparams)
2852
      filled_hvp.update(self.hvparams)
2853
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
2854
      hv_type.CheckParameterSyntax(filled_hvp)
2855
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
2856
      self.op.hvparams = self.hvparams
2857

    
2858
    _CheckNodeOnline(self, instance.primary_node)
2859

    
2860
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2861
    # check bridges existance
2862
    _CheckInstanceBridgesExist(self, instance)
2863

    
2864
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2865
                                              instance.name,
2866
                                              instance.hypervisor)
2867
    remote_info.Raise()
2868
    if not remote_info.data:
2869
      _CheckNodeFreeMemory(self, instance.primary_node,
2870
                           "starting instance %s" % instance.name,
2871
                           bep[constants.BE_MEMORY], instance.hypervisor)
2872

    
2873
  def Exec(self, feedback_fn):
2874
    """Start the instance.
2875

2876
    """
2877
    instance = self.instance
2878
    force = self.op.force
2879

    
2880
    self.cfg.MarkInstanceUp(instance.name)
2881

    
2882
    node_current = instance.primary_node
2883

    
2884
    _StartInstanceDisks(self, instance, force)
2885

    
2886
    result = self.rpc.call_instance_start(node_current, instance,
2887
                                          self.hvparams, self.beparams)
2888
    msg = result.RemoteFailMsg()
2889
    if msg:
2890
      _ShutdownInstanceDisks(self, instance)
2891
      raise errors.OpExecError("Could not start instance: %s" % msg)
2892

    
2893

    
2894
class LURebootInstance(LogicalUnit):
2895
  """Reboot an instance.
2896

2897
  """
2898
  HPATH = "instance-reboot"
2899
  HTYPE = constants.HTYPE_INSTANCE
2900
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2901
  REQ_BGL = False
2902

    
2903
  def ExpandNames(self):
2904
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2905
                                   constants.INSTANCE_REBOOT_HARD,
2906
                                   constants.INSTANCE_REBOOT_FULL]:
2907
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2908
                                  (constants.INSTANCE_REBOOT_SOFT,
2909
                                   constants.INSTANCE_REBOOT_HARD,
2910
                                   constants.INSTANCE_REBOOT_FULL))
2911
    self._ExpandAndLockInstance()
2912

    
2913
  def BuildHooksEnv(self):
2914
    """Build hooks env.
2915

2916
    This runs on master, primary and secondary nodes of the instance.
2917

2918
    """
2919
    env = {
2920
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2921
      "REBOOT_TYPE": self.op.reboot_type,
2922
      }
2923
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2924
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2925
    return env, nl, nl
2926

    
2927
  def CheckPrereq(self):
2928
    """Check prerequisites.
2929

2930
    This checks that the instance is in the cluster.
2931

2932
    """
2933
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2934
    assert self.instance is not None, \
2935
      "Cannot retrieve locked instance %s" % self.op.instance_name
2936

    
2937
    _CheckNodeOnline(self, instance.primary_node)
2938

    
2939
    # check bridges existance
2940
    _CheckInstanceBridgesExist(self, instance)
2941

    
2942
  def Exec(self, feedback_fn):
2943
    """Reboot the instance.
2944

2945
    """
2946
    instance = self.instance
2947
    ignore_secondaries = self.op.ignore_secondaries
2948
    reboot_type = self.op.reboot_type
2949

    
2950
    node_current = instance.primary_node
2951

    
2952
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
2953
                       constants.INSTANCE_REBOOT_HARD]:
2954
      for disk in instance.disks:
2955
        self.cfg.SetDiskID(disk, node_current)
2956
      result = self.rpc.call_instance_reboot(node_current, instance,
2957
                                             reboot_type)
2958
      msg = result.RemoteFailMsg()
2959
      if msg:
2960
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
2961
    else:
2962
      result = self.rpc.call_instance_shutdown(node_current, instance)
2963
      msg = result.RemoteFailMsg()
2964
      if msg:
2965
        raise errors.OpExecError("Could not shutdown instance for"
2966
                                 " full reboot: %s" % msg)
2967
      _ShutdownInstanceDisks(self, instance)
2968
      _StartInstanceDisks(self, instance, ignore_secondaries)
2969
      result = self.rpc.call_instance_start(node_current, instance, None, None)
2970
      msg = result.RemoteFailMsg()
2971
      if msg:
2972
        _ShutdownInstanceDisks(self, instance)
2973
        raise errors.OpExecError("Could not start instance for"
2974
                                 " full reboot: %s" % msg)
2975

    
2976
    self.cfg.MarkInstanceUp(instance.name)
2977

    
2978

    
2979
class LUShutdownInstance(LogicalUnit):
2980
  """Shutdown an instance.
2981

2982
  """
2983
  HPATH = "instance-stop"
2984
  HTYPE = constants.HTYPE_INSTANCE
2985
  _OP_REQP = ["instance_name"]
2986
  REQ_BGL = False
2987

    
2988
  def ExpandNames(self):
2989
    self._ExpandAndLockInstance()
2990

    
2991
  def BuildHooksEnv(self):
2992
    """Build hooks env.
2993

2994
    This runs on master, primary and secondary nodes of the instance.
2995

2996
    """
2997
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2998
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2999
    return env, nl, nl
3000

    
3001
  def CheckPrereq(self):
3002
    """Check prerequisites.
3003

3004
    This checks that the instance is in the cluster.
3005

3006
    """
3007
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3008
    assert self.instance is not None, \
3009
      "Cannot retrieve locked instance %s" % self.op.instance_name
3010
    _CheckNodeOnline(self, self.instance.primary_node)
3011

    
3012
  def Exec(self, feedback_fn):
3013
    """Shutdown the instance.
3014

3015
    """
3016
    instance = self.instance
3017
    node_current = instance.primary_node
3018
    self.cfg.MarkInstanceDown(instance.name)
3019
    result = self.rpc.call_instance_shutdown(node_current, instance)
3020
    msg = result.RemoteFailMsg()
3021
    if msg:
3022
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
3023

    
3024
    _ShutdownInstanceDisks(self, instance)
3025

    
3026

    
3027
class LUReinstallInstance(LogicalUnit):
3028
  """Reinstall an instance.
3029

3030
  """
3031
  HPATH = "instance-reinstall"
3032
  HTYPE = constants.HTYPE_INSTANCE
3033
  _OP_REQP = ["instance_name"]
3034
  REQ_BGL = False
3035

    
3036
  def ExpandNames(self):
3037
    self._ExpandAndLockInstance()
3038

    
3039
  def BuildHooksEnv(self):
3040
    """Build hooks env.
3041

3042
    This runs on master, primary and secondary nodes of the instance.
3043

3044
    """
3045
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3046
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3047
    return env, nl, nl
3048

    
3049
  def CheckPrereq(self):
3050
    """Check prerequisites.
3051

3052
    This checks that the instance is in the cluster and is not running.
3053

3054
    """
3055
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3056
    assert instance is not None, \
3057
      "Cannot retrieve locked instance %s" % self.op.instance_name
3058
    _CheckNodeOnline(self, instance.primary_node)
3059

    
3060
    if instance.disk_template == constants.DT_DISKLESS:
3061
      raise errors.OpPrereqError("Instance '%s' has no disks" %
3062
                                 self.op.instance_name)
3063
    if instance.admin_up:
3064
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3065
                                 self.op.instance_name)
3066
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3067
                                              instance.name,
3068
                                              instance.hypervisor)
3069
    remote_info.Raise()
3070
    if remote_info.data:
3071
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3072
                                 (self.op.instance_name,
3073
                                  instance.primary_node))
3074

    
3075
    self.op.os_type = getattr(self.op, "os_type", None)
3076
    if self.op.os_type is not None:
3077
      # OS verification
3078
      pnode = self.cfg.GetNodeInfo(
3079
        self.cfg.ExpandNodeName(instance.primary_node))
3080
      if pnode is None:
3081
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
3082
                                   self.op.pnode)
3083
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
3084
      result.Raise()
3085
      if not isinstance(result.data, objects.OS):
3086
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
3087
                                   " primary node"  % self.op.os_type)
3088

    
3089
    self.instance = instance
3090

    
3091
  def Exec(self, feedback_fn):
3092
    """Reinstall the instance.
3093

3094
    """
3095
    inst = self.instance
3096

    
3097
    if self.op.os_type is not None:
3098
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3099
      inst.os = self.op.os_type
3100
      self.cfg.Update(inst)
3101

    
3102
    _StartInstanceDisks(self, inst, None)
3103
    try:
3104
      feedback_fn("Running the instance OS create scripts...")
3105
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True)
3106
      msg = result.RemoteFailMsg()
3107
      if msg:
3108
        raise errors.OpExecError("Could not install OS for instance %s"
3109
                                 " on node %s: %s" %
3110
                                 (inst.name, inst.primary_node, msg))
3111
    finally:
3112
      _ShutdownInstanceDisks(self, inst)
3113

    
3114

    
3115
class LURenameInstance(LogicalUnit):
3116
  """Rename an instance.
3117

3118
  """
3119
  HPATH = "instance-rename"
3120
  HTYPE = constants.HTYPE_INSTANCE
3121
  _OP_REQP = ["instance_name", "new_name"]
3122

    
3123
  def BuildHooksEnv(self):
3124
    """Build hooks env.
3125

3126
    This runs on master, primary and secondary nodes of the instance.
3127

3128
    """
3129
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3130
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3131
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3132
    return env, nl, nl
3133

    
3134
  def CheckPrereq(self):
3135
    """Check prerequisites.
3136

3137
    This checks that the instance is in the cluster and is not running.
3138

3139
    """
3140
    instance = self.cfg.GetInstanceInfo(
3141
      self.cfg.ExpandInstanceName(self.op.instance_name))
3142
    if instance is None:
3143
      raise errors.OpPrereqError("Instance '%s' not known" %
3144
                                 self.op.instance_name)
3145
    _CheckNodeOnline(self, instance.primary_node)
3146

    
3147
    if instance.admin_up:
3148
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3149
                                 self.op.instance_name)
3150
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3151
                                              instance.name,
3152
                                              instance.hypervisor)
3153
    remote_info.Raise()
3154
    if remote_info.data:
3155
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3156
                                 (self.op.instance_name,
3157
                                  instance.primary_node))
3158
    self.instance = instance
3159

    
3160
    # new name verification
3161
    name_info = utils.HostInfo(self.op.new_name)
3162

    
3163
    self.op.new_name = new_name = name_info.name
3164
    instance_list = self.cfg.GetInstanceList()
3165
    if new_name in instance_list:
3166
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3167
                                 new_name)
3168

    
3169
    if not getattr(self.op, "ignore_ip", False):
3170
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3171
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3172
                                   (name_info.ip, new_name))
3173

    
3174

    
3175
  def Exec(self, feedback_fn):
3176
    """Reinstall the instance.
3177

3178
    """
3179
    inst = self.instance
3180
    old_name = inst.name
3181

    
3182
    if inst.disk_template == constants.DT_FILE:
3183
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3184

    
3185
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3186
    # Change the instance lock. This is definitely safe while we hold the BGL
3187
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3188
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3189

    
3190
    # re-read the instance from the configuration after rename
3191
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3192

    
3193
    if inst.disk_template == constants.DT_FILE:
3194
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3195
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3196
                                                     old_file_storage_dir,
3197
                                                     new_file_storage_dir)
3198
      result.Raise()
3199
      if not result.data:
3200
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3201
                                 " directory '%s' to '%s' (but the instance"
3202
                                 " has been renamed in Ganeti)" % (
3203
                                 inst.primary_node, old_file_storage_dir,
3204
                                 new_file_storage_dir))
3205

    
3206
      if not result.data[0]:
3207
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3208
                                 " (but the instance has been renamed in"
3209
                                 " Ganeti)" % (old_file_storage_dir,
3210
                                               new_file_storage_dir))
3211

    
3212
    _StartInstanceDisks(self, inst, None)
3213
    try:
3214
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3215
                                                 old_name)
3216
      msg = result.RemoteFailMsg()
3217
      if msg:
3218
        msg = ("Could not run OS rename script for instance %s on node %s"
3219
               " (but the instance has been renamed in Ganeti): %s" %
3220
               (inst.name, inst.primary_node, msg))
3221
        self.proc.LogWarning(msg)
3222
    finally:
3223
      _ShutdownInstanceDisks(self, inst)
3224

    
3225

    
3226
class LURemoveInstance(LogicalUnit):
3227
  """Remove an instance.
3228

3229
  """
3230
  HPATH = "instance-remove"
3231
  HTYPE = constants.HTYPE_INSTANCE
3232
  _OP_REQP = ["instance_name", "ignore_failures"]
3233
  REQ_BGL = False
3234

    
3235
  def ExpandNames(self):
3236
    self._ExpandAndLockInstance()
3237
    self.needed_locks[locking.LEVEL_NODE] = []
3238
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3239

    
3240
  def DeclareLocks(self, level):
3241
    if level == locking.LEVEL_NODE:
3242
      self._LockInstancesNodes()
3243

    
3244
  def BuildHooksEnv(self):
3245
    """Build hooks env.
3246

3247
    This runs on master, primary and secondary nodes of the instance.
3248

3249
    """
3250
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3251
    nl = [self.cfg.GetMasterNode()]
3252
    return env, nl, nl
3253

    
3254
  def CheckPrereq(self):
3255
    """Check prerequisites.
3256

3257
    This checks that the instance is in the cluster.
3258

3259
    """
3260
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3261
    assert self.instance is not None, \
3262
      "Cannot retrieve locked instance %s" % self.op.instance_name
3263

    
3264
  def Exec(self, feedback_fn):
3265
    """Remove the instance.
3266

3267
    """
3268
    instance = self.instance
3269
    logging.info("Shutting down instance %s on node %s",
3270
                 instance.name, instance.primary_node)
3271

    
3272
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3273
    msg = result.RemoteFailMsg()
3274
    if msg:
3275
      if self.op.ignore_failures:
3276
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3277
      else:
3278
        raise errors.OpExecError("Could not shutdown instance %s on"
3279
                                 " node %s: %s" %
3280
                                 (instance.name, instance.primary_node, msg))
3281

    
3282
    logging.info("Removing block devices for instance %s", instance.name)
3283

    
3284
    if not _RemoveDisks(self, instance):
3285
      if self.op.ignore_failures:
3286
        feedback_fn("Warning: can't remove instance's disks")
3287
      else:
3288
        raise errors.OpExecError("Can't remove instance's disks")
3289

    
3290
    logging.info("Removing instance %s out of cluster config", instance.name)
3291

    
3292
    self.cfg.RemoveInstance(instance.name)
3293
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3294

    
3295

    
3296
class LUQueryInstances(NoHooksLU):
3297
  """Logical unit for querying instances.
3298

3299
  """
3300
  _OP_REQP = ["output_fields", "names", "use_locking"]
3301
  REQ_BGL = False
3302
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3303
                                    "admin_state",
3304
                                    "disk_template", "ip", "mac", "bridge",
3305
                                    "sda_size", "sdb_size", "vcpus", "tags",
3306
                                    "network_port", "beparams",
3307
                                    r"(disk)\.(size)/([0-9]+)",
3308
                                    r"(disk)\.(sizes)", "disk_usage",
3309
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3310
                                    r"(nic)\.(macs|ips|bridges)",
3311
                                    r"(disk|nic)\.(count)",
3312
                                    "serial_no", "hypervisor", "hvparams",] +
3313
                                  ["hv/%s" % name
3314
                                   for name in constants.HVS_PARAMETERS] +
3315
                                  ["be/%s" % name
3316
                                   for name in constants.BES_PARAMETERS])
3317
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3318

    
3319

    
3320
  def ExpandNames(self):
3321
    _CheckOutputFields(static=self._FIELDS_STATIC,
3322
                       dynamic=self._FIELDS_DYNAMIC,
3323
                       selected=self.op.output_fields)
3324

    
3325
    self.needed_locks = {}
3326
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3327
    self.share_locks[locking.LEVEL_NODE] = 1
3328

    
3329
    if self.op.names:
3330
      self.wanted = _GetWantedInstances(self, self.op.names)
3331
    else:
3332
      self.wanted = locking.ALL_SET
3333

    
3334
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3335
    self.do_locking = self.do_node_query and self.op.use_locking
3336
    if self.do_locking:
3337
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3338
      self.needed_locks[locking.LEVEL_NODE] = []
3339
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3340

    
3341
  def DeclareLocks(self, level):
3342
    if level == locking.LEVEL_NODE and self.do_locking:
3343
      self._LockInstancesNodes()
3344

    
3345
  def CheckPrereq(self):
3346
    """Check prerequisites.
3347

3348
    """
3349
    pass
3350

    
3351
  def Exec(self, feedback_fn):
3352
    """Computes the list of nodes and their attributes.
3353

3354
    """
3355
    all_info = self.cfg.GetAllInstancesInfo()
3356
    if self.wanted == locking.ALL_SET:
3357
      # caller didn't specify instance names, so ordering is not important
3358
      if self.do_locking:
3359
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3360
      else:
3361
        instance_names = all_info.keys()
3362
      instance_names = utils.NiceSort(instance_names)
3363
    else:
3364
      # caller did specify names, so we must keep the ordering
3365
      if self.do_locking:
3366
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3367
      else:
3368
        tgt_set = all_info.keys()
3369
      missing = set(self.wanted).difference(tgt_set)
3370
      if missing:
3371
        raise errors.OpExecError("Some instances were removed before"
3372
                                 " retrieving their data: %s" % missing)
3373
      instance_names = self.wanted
3374

    
3375
    instance_list = [all_info[iname] for iname in instance_names]
3376

    
3377
    # begin data gathering
3378

    
3379
    nodes = frozenset([inst.primary_node for inst in instance_list])
3380
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3381

    
3382
    bad_nodes = []
3383
    off_nodes = []
3384
    if self.do_node_query:
3385
      live_data = {}
3386
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3387
      for name in nodes:
3388
        result = node_data[name]
3389
        if result.offline:
3390
          # offline nodes will be in both lists
3391
          off_nodes.append(name)
3392
        if result.failed:
3393
          bad_nodes.append(name)
3394
        else:
3395
          if result.data:
3396
            live_data.update(result.data)
3397
            # else no instance is alive
3398
    else:
3399
      live_data = dict([(name, {}) for name in instance_names])
3400

    
3401
    # end data gathering
3402

    
3403
    HVPREFIX = "hv/"
3404
    BEPREFIX = "be/"
3405
    output = []
3406
    for instance in instance_list:
3407
      iout = []
3408
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3409
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3410
      for field in self.op.output_fields:
3411
        st_match = self._FIELDS_STATIC.Matches(field)
3412
        if field == "name":
3413
          val = instance.name
3414
        elif field == "os":
3415
          val = instance.os
3416
        elif field == "pnode":
3417
          val = instance.primary_node
3418
        elif field == "snodes":
3419
          val = list(instance.secondary_nodes)
3420
        elif field == "admin_state":
3421
          val = instance.admin_up
3422
        elif field == "oper_state":
3423
          if instance.primary_node in bad_nodes:
3424
            val = None
3425
          else:
3426
            val = bool(live_data.get(instance.name))
3427
        elif field == "status":
3428
          if instance.primary_node in off_nodes:
3429
            val = "ERROR_nodeoffline"
3430
          elif instance.primary_node in bad_nodes:
3431
            val = "ERROR_nodedown"
3432
          else:
3433
            running = bool(live_data.get(instance.name))
3434
            if running:
3435
              if instance.admin_up:
3436
                val = "running"
3437
              else:
3438
                val = "ERROR_up"
3439
            else:
3440
              if instance.admin_up:
3441
                val = "ERROR_down"
3442
              else:
3443
                val = "ADMIN_down"
3444
        elif field == "oper_ram":
3445
          if instance.primary_node in bad_nodes:
3446
            val = None
3447
          elif instance.name in live_data:
3448
            val = live_data[instance.name].get("memory", "?")
3449
          else:
3450
            val = "-"
3451
        elif field == "disk_template":
3452
          val = instance.disk_template
3453
        elif field == "ip":
3454
          val = instance.nics[0].ip
3455
        elif field == "bridge":
3456
          val = instance.nics[0].bridge
3457
        elif field == "mac":
3458
          val = instance.nics[0].mac
3459
        elif field == "sda_size" or field == "sdb_size":
3460
          idx = ord(field[2]) - ord('a')
3461
          try:
3462
            val = instance.FindDisk(idx).size
3463
          except errors.OpPrereqError:
3464
            val = None
3465
        elif field == "disk_usage": # total disk usage per node
3466
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3467
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3468
        elif field == "tags":
3469
          val = list(instance.GetTags())
3470
        elif field == "serial_no":
3471
          val = instance.serial_no
3472
        elif field == "network_port":
3473
          val = instance.network_port
3474
        elif field == "hypervisor":
3475
          val = instance.hypervisor
3476
        elif field == "hvparams":
3477
          val = i_hv
3478
        elif (field.startswith(HVPREFIX) and
3479
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3480
          val = i_hv.get(field[len(HVPREFIX):], None)
3481
        elif field == "beparams":
3482
          val = i_be
3483
        elif (field.startswith(BEPREFIX) and
3484
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3485
          val = i_be.get(field[len(BEPREFIX):], None)
3486
        elif st_match and st_match.groups():
3487
          # matches a variable list
3488
          st_groups = st_match.groups()
3489
          if st_groups and st_groups[0] == "disk":
3490
            if st_groups[1] == "count":
3491
              val = len(instance.disks)
3492
            elif st_groups[1] == "sizes":
3493
              val = [disk.size for disk in instance.disks]
3494
            elif st_groups[1] == "size":
3495
              try:
3496
                val = instance.FindDisk(st_groups[2]).size
3497
              except errors.OpPrereqError:
3498
                val = None
3499
            else:
3500
              assert False, "Unhandled disk parameter"
3501
          elif st_groups[0] == "nic":
3502
            if st_groups[1] == "count":
3503
              val = len(instance.nics)
3504
            elif st_groups[1] == "macs":
3505
              val = [nic.mac for nic in instance.nics]
3506
            elif st_groups[1] == "ips":
3507
              val = [nic.ip for nic in instance.nics]
3508
            elif st_groups[1] == "bridges":
3509
              val = [nic.bridge for nic in instance.nics]
3510
            else:
3511
              # index-based item
3512
              nic_idx = int(st_groups[2])
3513
              if nic_idx >= len(instance.nics):
3514
                val = None
3515
              else:
3516
                if st_groups[1] == "mac":
3517
                  val = instance.nics[nic_idx].mac
3518
                elif st_groups[1] == "ip":
3519
                  val = instance.nics[nic_idx].ip
3520
                elif st_groups[1] == "bridge":
3521
                  val = instance.nics[nic_idx].bridge
3522
                else:
3523
                  assert False, "Unhandled NIC parameter"
3524
          else:
3525
            assert False, "Unhandled variable parameter"
3526
        else:
3527
          raise errors.ParameterError(field)
3528
        iout.append(val)
3529
      output.append(iout)
3530

    
3531
    return output
3532

    
3533

    
3534
class LUFailoverInstance(LogicalUnit):
3535
  """Failover an instance.
3536

3537
  """
3538
  HPATH = "instance-failover"
3539
  HTYPE = constants.HTYPE_INSTANCE
3540
  _OP_REQP = ["instance_name", "ignore_consistency"]
3541
  REQ_BGL = False
3542

    
3543
  def ExpandNames(self):
3544
    self._ExpandAndLockInstance()
3545
    self.needed_locks[locking.LEVEL_NODE] = []
3546
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3547

    
3548
  def DeclareLocks(self, level):
3549
    if level == locking.LEVEL_NODE:
3550
      self._LockInstancesNodes()
3551

    
3552
  def BuildHooksEnv(self):
3553
    """Build hooks env.
3554

3555
    This runs on master, primary and secondary nodes of the instance.
3556

3557
    """
3558
    env = {
3559
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3560
      }
3561
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3562
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3563
    return env, nl, nl
3564

    
3565
  def CheckPrereq(self):
3566
    """Check prerequisites.
3567

3568
    This checks that the instance is in the cluster.
3569

3570
    """
3571
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3572
    assert self.instance is not None, \
3573
      "Cannot retrieve locked instance %s" % self.op.instance_name
3574

    
3575
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3576
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3577
      raise errors.OpPrereqError("Instance's disk layout is not"
3578
                                 " network mirrored, cannot failover.")
3579

    
3580
    secondary_nodes = instance.secondary_nodes
3581
    if not secondary_nodes:
3582
      raise errors.ProgrammerError("no secondary node but using "
3583
                                   "a mirrored disk template")
3584

    
3585
    target_node = secondary_nodes[0]
3586
    _CheckNodeOnline(self, target_node)
3587
    _CheckNodeNotDrained(self, target_node)
3588
    # check memory requirements on the secondary node
3589
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3590
                         instance.name, bep[constants.BE_MEMORY],
3591
                         instance.hypervisor)
3592

    
3593
    # check bridge existance
3594
    brlist = [nic.bridge for nic in instance.nics]
3595
    result = self.rpc.call_bridges_exist(target_node, brlist)
3596
    result.Raise()
3597
    if not result.data:
3598
      raise errors.OpPrereqError("One or more target bridges %s does not"
3599
                                 " exist on destination node '%s'" %
3600
                                 (brlist, target_node))
3601

    
3602
  def Exec(self, feedback_fn):
3603
    """Failover an instance.
3604

3605
    The failover is done by shutting it down on its present node and
3606
    starting it on the secondary.
3607

3608
    """
3609
    instance = self.instance
3610

    
3611
    source_node = instance.primary_node
3612
    target_node = instance.secondary_nodes[0]
3613

    
3614
    feedback_fn("* checking disk consistency between source and target")
3615
    for dev in instance.disks:
3616
      # for drbd, these are drbd over lvm
3617
      if not _CheckDiskConsistency(self, dev, target_node, False):
3618
        if instance.admin_up and not self.op.ignore_consistency:
3619
          raise errors.OpExecError("Disk %s is degraded on target node,"
3620
                                   " aborting failover." % dev.iv_name)
3621

    
3622
    feedback_fn("* shutting down instance on source node")
3623
    logging.info("Shutting down instance %s on node %s",
3624
                 instance.name, source_node)
3625

    
3626
    result = self.rpc.call_instance_shutdown(source_node, instance)
3627
    msg = result.RemoteFailMsg()
3628
    if msg:
3629
      if self.op.ignore_consistency:
3630
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3631
                             " Proceeding anyway. Please make sure node"
3632
                             " %s is down. Error details: %s",
3633
                             instance.name, source_node, source_node, msg)
3634
      else:
3635
        raise errors.OpExecError("Could not shutdown instance %s on"
3636
                                 " node %s: %s" %
3637
                                 (instance.name, source_node, msg))
3638

    
3639
    feedback_fn("* deactivating the instance's disks on source node")
3640
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3641
      raise errors.OpExecError("Can't shut down the instance's disks.")
3642

    
3643
    instance.primary_node = target_node
3644
    # distribute new instance config to the other nodes
3645
    self.cfg.Update(instance)
3646

    
3647
    # Only start the instance if it's marked as up
3648
    if instance.admin_up:
3649
      feedback_fn("* activating the instance's disks on target node")
3650
      logging.info("Starting instance %s on node %s",
3651
                   instance.name, target_node)
3652

    
3653
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3654
                                               ignore_secondaries=True)
3655
      if not disks_ok:
3656
        _ShutdownInstanceDisks(self, instance)
3657
        raise errors.OpExecError("Can't activate the instance's disks")
3658

    
3659
      feedback_fn("* starting the instance on the target node")
3660
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3661
      msg = result.RemoteFailMsg()
3662
      if msg:
3663
        _ShutdownInstanceDisks(self, instance)
3664
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3665
                                 (instance.name, target_node, msg))
3666

    
3667

    
3668
class LUMigrateInstance(LogicalUnit):
3669
  """Migrate an instance.
3670

3671
  This is migration without shutting down, compared to the failover,
3672
  which is done with shutdown.
3673

3674
  """
3675
  HPATH = "instance-migrate"
3676
  HTYPE = constants.HTYPE_INSTANCE
3677
  _OP_REQP = ["instance_name", "live", "cleanup"]
3678

    
3679
  REQ_BGL = False
3680

    
3681
  def ExpandNames(self):
3682
    self._ExpandAndLockInstance()
3683
    self.needed_locks[locking.LEVEL_NODE] = []
3684
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3685

    
3686
  def DeclareLocks(self, level):
3687
    if level == locking.LEVEL_NODE:
3688
      self._LockInstancesNodes()
3689

    
3690
  def BuildHooksEnv(self):
3691
    """Build hooks env.
3692

3693
    This runs on master, primary and secondary nodes of the instance.
3694

3695
    """
3696
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3697
    env["MIGRATE_LIVE"] = self.op.live
3698
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3699
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3700
    return env, nl, nl
3701

    
3702
  def CheckPrereq(self):
3703
    """Check prerequisites.
3704

3705
    This checks that the instance is in the cluster.
3706

3707
    """
3708
    instance = self.cfg.GetInstanceInfo(
3709
      self.cfg.ExpandInstanceName(self.op.instance_name))
3710
    if instance is None:
3711
      raise errors.OpPrereqError("Instance '%s' not known" %
3712
                                 self.op.instance_name)
3713

    
3714
    if instance.disk_template != constants.DT_DRBD8:
3715
      raise errors.OpPrereqError("Instance's disk layout is not"
3716
                                 " drbd8, cannot migrate.")
3717

    
3718
    secondary_nodes = instance.secondary_nodes
3719
    if not secondary_nodes:
3720
      raise errors.ConfigurationError("No secondary node but using"
3721
                                      " drbd8 disk template")
3722

    
3723
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3724

    
3725
    target_node = secondary_nodes[0]
3726
    # check memory requirements on the secondary node
3727
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3728
                         instance.name, i_be[constants.BE_MEMORY],
3729
                         instance.hypervisor)
3730

    
3731
    # check bridge existance
3732
    brlist = [nic.bridge for nic in instance.nics]
3733
    result = self.rpc.call_bridges_exist(target_node, brlist)
3734
    if result.failed or not result.data:
3735
      raise errors.OpPrereqError("One or more target bridges %s does not"
3736
                                 " exist on destination node '%s'" %
3737
                                 (brlist, target_node))
3738

    
3739
    if not self.op.cleanup:
3740
      _CheckNodeNotDrained(self, target_node)
3741
      result = self.rpc.call_instance_migratable(instance.primary_node,
3742
                                                 instance)
3743
      msg = result.RemoteFailMsg()
3744
      if msg:
3745
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3746
                                   msg)
3747

    
3748
    self.instance = instance
3749

    
3750
  def _WaitUntilSync(self):
3751
    """Poll with custom rpc for disk sync.
3752

3753
    This uses our own step-based rpc call.
3754

3755
    """
3756
    self.feedback_fn("* wait until resync is done")
3757
    all_done = False
3758
    while not all_done:
3759
      all_done = True
3760
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3761
                                            self.nodes_ip,
3762
                                            self.instance.disks)
3763
      min_percent = 100
3764
      for node, nres in result.items():
3765
        msg = nres.RemoteFailMsg()
3766
        if msg:
3767
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3768
                                   (node, msg))
3769
        node_done, node_percent = nres.payload
3770
        all_done = all_done and node_done
3771
        if node_percent is not None:
3772
          min_percent = min(min_percent, node_percent)
3773
      if not all_done:
3774
        if min_percent < 100:
3775
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3776
        time.sleep(2)
3777

    
3778
  def _EnsureSecondary(self, node):
3779
    """Demote a node to secondary.
3780

3781
    """
3782
    self.feedback_fn("* switching node %s to secondary mode" % node)
3783

    
3784
    for dev in self.instance.disks:
3785
      self.cfg.SetDiskID(dev, node)
3786

    
3787
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3788
                                          self.instance.disks)
3789
    msg = result.RemoteFailMsg()
3790
    if msg:
3791
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3792
                               " error %s" % (node, msg))
3793

    
3794
  def _GoStandalone(self):
3795
    """Disconnect from the network.
3796

3797
    """
3798
    self.feedback_fn("* changing into standalone mode")
3799
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3800
                                               self.instance.disks)
3801
    for node, nres in result.items():
3802
      msg = nres.RemoteFailMsg()
3803
      if msg:
3804
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3805
                                 " error %s" % (node, msg))
3806

    
3807
  def _GoReconnect(self, multimaster):
3808
    """Reconnect to the network.
3809

3810
    """
3811
    if multimaster:
3812
      msg = "dual-master"
3813
    else:
3814
      msg = "single-master"
3815
    self.feedback_fn("* changing disks into %s mode" % msg)
3816
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3817
                                           self.instance.disks,
3818
                                           self.instance.name, multimaster)
3819
    for node, nres in result.items():
3820
      msg = nres.RemoteFailMsg()
3821
      if msg:
3822
        raise errors.OpExecError("Cannot change disks config on node %s,"
3823
                                 " error: %s" % (node, msg))
3824

    
3825
  def _ExecCleanup(self):
3826
    """Try to cleanup after a failed migration.
3827

3828
    The cleanup is done by:
3829
      - check that the instance is running only on one node
3830
        (and update the config if needed)
3831
      - change disks on its secondary node to secondary
3832
      - wait until disks are fully synchronized
3833
      - disconnect from the network
3834
      - change disks into single-master mode
3835
      - wait again until disks are fully synchronized
3836

3837
    """
3838
    instance = self.instance
3839
    target_node = self.target_node
3840
    source_node = self.source_node
3841

    
3842
    # check running on only one node
3843
    self.feedback_fn("* checking where the instance actually runs"
3844
                     " (if this hangs, the hypervisor might be in"
3845
                     " a bad state)")
3846
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3847
    for node, result in ins_l.items():
3848
      result.Raise()
3849
      if not isinstance(result.data, list):
3850
        raise errors.OpExecError("Can't contact node '%s'" % node)
3851

    
3852
    runningon_source = instance.name in ins_l[source_node].data
3853
    runningon_target = instance.name in ins_l[target_node].data
3854

    
3855
    if runningon_source and runningon_target:
3856
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3857
                               " or the hypervisor is confused. You will have"
3858
                               " to ensure manually that it runs only on one"
3859
                               " and restart this operation.")
3860

    
3861
    if not (runningon_source or runningon_target):
3862
      raise errors.OpExecError("Instance does not seem to be running at all."
3863
                               " In this case, it's safer to repair by"
3864
                               " running 'gnt-instance stop' to ensure disk"
3865
                               " shutdown, and then restarting it.")
3866

    
3867
    if runningon_target:
3868
      # the migration has actually succeeded, we need to update the config
3869
      self.feedback_fn("* instance running on secondary node (%s),"
3870
                       " updating config" % target_node)
3871
      instance.primary_node = target_node
3872
      self.cfg.Update(instance)
3873
      demoted_node = source_node
3874
    else:
3875
      self.feedback_fn("* instance confirmed to be running on its"
3876
                       " primary node (%s)" % source_node)
3877
      demoted_node = target_node
3878

    
3879
    self._EnsureSecondary(demoted_node)
3880
    try:
3881
      self._WaitUntilSync()
3882
    except errors.OpExecError:
3883
      # we ignore here errors, since if the device is standalone, it
3884
      # won't be able to sync
3885
      pass
3886
    self._GoStandalone()
3887
    self._GoReconnect(False)
3888
    self._WaitUntilSync()
3889

    
3890
    self.feedback_fn("* done")
3891

    
3892
  def _RevertDiskStatus(self):
3893
    """Try to revert the disk status after a failed migration.
3894

3895
    """
3896
    target_node = self.target_node
3897
    try:
3898
      self._EnsureSecondary(target_node)
3899
      self._GoStandalone()
3900
      self._GoReconnect(False)
3901
      self._WaitUntilSync()
3902
    except errors.OpExecError, err:
3903
      self.LogWarning("Migration failed and I can't reconnect the"
3904
                      " drives: error '%s'\n"
3905
                      "Please look and recover the instance status" %
3906
                      str(err))
3907

    
3908
  def _AbortMigration(self):
3909
    """Call the hypervisor code to abort a started migration.
3910

3911
    """
3912
    instance = self.instance
3913
    target_node = self.target_node
3914
    migration_info = self.migration_info
3915

    
3916
    abort_result = self.rpc.call_finalize_migration(target_node,
3917
                                                    instance,
3918
                                                    migration_info,
3919
                                                    False)
3920
    abort_msg = abort_result.RemoteFailMsg()
3921
    if abort_msg:
3922
      logging.error("Aborting migration failed on target node %s: %s" %
3923
                    (target_node, abort_msg))
3924
      # Don't raise an exception here, as we stil have to try to revert the
3925
      # disk status, even if this step failed.
3926

    
3927
  def _ExecMigration(self):
3928
    """Migrate an instance.
3929

3930
    The migrate is done by:
3931
      - change the disks into dual-master mode
3932
      - wait until disks are fully synchronized again
3933
      - migrate the instance
3934
      - change disks on the new secondary node (the old primary) to secondary
3935
      - wait until disks are fully synchronized
3936
      - change disks into single-master mode
3937

3938
    """
3939
    instance = self.instance
3940
    target_node = self.target_node
3941
    source_node = self.source_node
3942

    
3943
    self.feedback_fn("* checking disk consistency between source and target")
3944
    for dev in instance.disks:
3945
      if not _CheckDiskConsistency(self, dev, target_node, False):
3946
        raise errors.OpExecError("Disk %s is degraded or not fully"
3947
                                 " synchronized on target node,"
3948
                                 " aborting migrate." % dev.iv_name)
3949

    
3950
    # First get the migration information from the remote node
3951
    result = self.rpc.call_migration_info(source_node, instance)
3952
    msg = result.RemoteFailMsg()
3953
    if msg:
3954
      log_err = ("Failed fetching source migration information from %s: %s" %
3955
                 (source_node, msg))
3956
      logging.error(log_err)
3957
      raise errors.OpExecError(log_err)
3958

    
3959
    self.migration_info = migration_info = result.payload
3960

    
3961
    # Then switch the disks to master/master mode
3962
    self._EnsureSecondary(target_node)
3963
    self._GoStandalone()
3964
    self._GoReconnect(True)
3965
    self._WaitUntilSync()
3966

    
3967
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
3968
    result = self.rpc.call_accept_instance(target_node,
3969
                                           instance,
3970
                                           migration_info,
3971
                                           self.nodes_ip[target_node])
3972

    
3973
    msg = result.RemoteFailMsg()
3974
    if msg:
3975
      logging.error("Instance pre-migration failed, trying to revert"
3976
                    " disk status: %s", msg)
3977
      self._AbortMigration()
3978
      self._RevertDiskStatus()
3979
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
3980
                               (instance.name, msg))
3981

    
3982
    self.feedback_fn("* migrating instance to %s" % target_node)
3983
    time.sleep(10)
3984
    result = self.rpc.call_instance_migrate(source_node, instance,
3985